Toronto Machine Learning Summit

Meet Past Speakers

Advanced Technical / Research

Click Speaker to Read Abstract

Business Strategy

Click Speaker to Read Abstract

Case Studies

Click Speaker to Read Abstract

Hands-on Workshops

Click Speaker to Read Abstract

Virtual Workshops

Click Speaker to Read Abstract

Lightning Talks

Click Speaker to Read Abstract

Tickets

This event has ended

This event is no longer available.

Agenda

This agenda is still subject to changes

This event has ended

This event is no longer available.

Ihab Ilyas

Professor, Cheriton School of Computer Science, University of Waterloo
Director, Head of Apple Knowledge Platform, Apple

Ihab Ilyas is a professor in the Cheriton School of Computer Science and the NSERC-Thomson Reuters Research Chair on data quality at the University of Waterloo. He is currently on leave at Apple to lead the Apple Knowledge Platform team. His main research focuses on the areas of Data Science and data management, with special interest in data quality and integration, managing uncertain data, machine learning for data curation, and information extraction. Ihab is a co-founder of Tamr, a startup focusing on large-scale data integration, and the co-founder of inductiv (acquired by Apple), a Waterloo-based startup on using AI for structured data cleaning. He is an ACM Fellow and IEEE Fellow, a recipient of the Ontario Early Researcher Award, a Cheriton Faculty Fellowship, an NSERC Discovery Accelerator Award, and a Google Faculty Award. Ihab was an elected member of the VLDB Endowment board of trustees (2016-2021), elected SIGMOD vice chair (2016-2021), an associate editor of the ACM Transactions of Database Systems (2014-2020), and an associate editor of Foundations of Database Systems. He holds a PhD in computer science from Purdue University, West Lafayette.

Talk: Saga: Continuous Construction and Serving of Large Scale Knowledge Graphs

Abstract: In this talk I present Saga, an end-to-end platform for incremental and continuous construction of large scale knowledge graphs we built at Apple. Saga demonstrates the complexity of building such platform in industrial settings with strong consistency, latency, and coverage requirements. In the talk, I will discuss challenges around the following: building source adapters for ingesting heterogenous data sources; building entity linking and fusion pipelines for constructing coherent knowledge graphs that adhere to a common controlled vocabulary; updating the knowledge graphs with real-time streams; and finally, exposing the constructed knowledge via a variety of services. Graph services include: low-latency query answering; graph analytics; ML-biased entity disambiguation and semantic annotation; and other graph-embedding services to power multiple downstream applications. Saga is used in production at large scale to power a variety of user-facing knowledge features.

What You’ll Learn: Complexity of building large scale knowledge graphs

Track: Technical

Technical Level: 5

Location: Seattle

Talk: Saga: Continuous Construction and Serving of Large Scale Knowledge Graphs

Presenter:
Ihab Ilyas, Professor in the Cheriton School of Computer Science and the NSERC-Thomson Reuters Research Chair on Data quality at the University of Waterloo

About the Speaker:
Ihab Ilyas is a professor in the Cheriton School of Computer Science and the NSERC-Thomson Reuters Research Chair on data quality at the University of Waterloo. He is currently on leave at Apple to lead the Apple Knowledge Platform team. His main research focuses on the areas of Data Science and data management, with special interest in data quality and integration, managing uncertain data, machine learning for data curation, and information extraction. Ihab is a co-founder of Tamr, a startup focusing on large-scale data integration, and the co-founder of inductiv (acquired by Apple), a Waterloo-based startup on using AI for structured data cleaning. He is an ACM Fellow and IEEE Fellow, a recipient of the Ontario Early Researcher Award, a Cheriton Faculty Fellowship, an NSERC Discovery Accelerator Award, and a Google Faculty Award. Ihab was an elected member of the VLDB Endowment board of trustees (2016-2021), elected SIGMOD vice chair (2016-2021), an associate editor of the ACM Transactions of Database Systems (2014-2020), and an associate editor of Foundations of Database Systems. He holds a PhD in computer science from Purdue University, West Lafayette.

Which talk track does this best fit into?
Technical / Research

Technical level of your talk?
(Technical level: 5/7)

Are there any industries (in particular) that are relevant for this talk?
Information Technology & Service

What are the main core message (learning) you want attendees to take away from this talk?
Complexity of building large scale knowledge graphs.

Abstract of Talk:
In this talk I present Saga, an end-to-end platform for incremental and continuous construction of large scale knowledge graphs we built at Apple. Saga demonstrates the complexity of building such platform in industrial settings with strong consistency, latency, and coverage requirements. In the talk, I will discuss challenges around the following: building source adapters for ingesting heterogenous data sources; building entity linking and fusion pipelines for constructing coherent knowledge graphs that adhere to a common controlled vocabulary; updating the knowledge graphs with real-time streams; and finally, exposing the constructed knowledge via a variety of services. Graph services include: low-latency query answering; graph analytics; ML-biased entity disambiguation and semantic annotation; and other graph-embedding services to power multiple downstream applications. Saga is used in production at large scale to power a variety of user-facing knowledge features.

Anne Martel

Professor, University of Toronto

Anne Martel is a Professor in Medical Biophysics at the University of Toronto, the Tory Family Chair in Oncology at Sunnybrook Research Institute, and a Faculty Affiliate at the Vector Institute, Toronto. Her research program is focused on medical image and digital pathology analysis, particularly on the development of self-supervised and weakly supervised methods for segmentation, diagnosis, and prediction/prognosis. In 2006 she co-founded Pathcore, a software company developing complete workflow solutions for digital pathology.

Dr Martel is an active member of the medical image analysis community and is a fellow of the MICCAI Society which represents engineers and computer scientists working in this field. She has served as board member of MICCAI and is currently on the editorial board of Medical Image Analysis, on of the leading journals in the field.

Talk: Artificial Intelligence And Digital Pathology: Making The Most of Limited Annotated Data

Abstract: Obtaining large datasets with detailed annotations for medical imaging AI projects is a time consuming and expensive process as it usually requires the input of expert radiologists and pathologists. Collecting data to train outcome prediction models is even more challenging as the number of patients with both imaging and follow up data may be small, and only weak labels are available.

This talk will describe several semi-supervised and self-supervised approaches which can make more efficient use of small and/or weakly labelled datasets. The focus will be on digital pathology but the methods described are applicable any medical imaging modality.

What You’ll Learn: Self-supervision and smart sampling strategies are essential in digital pathology

Track: Advanced Technical/Research

Technical Level: 6

Location: Toronto

Talk: Artificial Intelligence And Digital Pathology: Making The Most of Limited Annotated Data

Presenter:
Anne Martel, Professor, University of Toronto

About the Speaker:
Anne Martel is a Professor in Medical Biophysics at the University of Toronto, the Tory Family Chair in Oncology at Sunnybrook Research Institute, and a Faculty Affiliate at the Vector Institute, Toronto. Her research program is focused on medical image and digital pathology analysis, particularly on the development of self-supervised and weakly supervised methods for segmentation, diagnosis, and prediction/prognosis. In 2006 she co-founded Pathcore, a software company developing complete workflow solutions for digital pathology.

Which talk track does this best fit into?
Advanced Technical / Research

Technical level of your talk?
(Technical Level: 6/7)

What you’ll learn:
Self-supervision and smart sampling strategies are essential in digital pathology

Abstract of Talk:
Obtaining large datasets with detailed annotations for medical imaging AI projects is a time consuming and expensive process as it usually requires the input of expert radiologists and pathologists. Collecting data to train outcome prediction models is even more challenging as the number of patients with both imaging and follow up data may be small, and only weak labels are available.

Varun Raj Kompella

Senior Research Scientist, Sony AI

Varun Kompella is currently a senior research scientist at Sony AI. He earned his master’s of science degree in informatics with a specialization in graphics, vision and robotics from Institut Nationale Polytechnique de Grenoble (INRIA Grenoble), and a Ph.D degree from Università della Svizzera Italiana (IDSIA Lugano), Switzerland, working with Prof. Juergen Schmidhuber. In his thesis work he developed algorithms that use the slowness principle for driving exploration in reinforcement learning agents. After completing his Ph.D., he worked as a postdoctoral researcher at the Institute for Neural Computation (INI), Germany. His research contributions led to several patents, publications in peer-reviewed journals and conference proceedings.

Talk: Outracing Champion Gran Turismo Drivers With Deep Reinforcement Learning

Abstract: Many potential applications of artificial intelligence involve making real-time decisions in physical systems while interacting with humans. Automobile racing represents an extreme example of these conditions; drivers must execute complex tactical manoeuvres to pass or block opponents while operating their vehicles at their traction limits. Racing simulations, such as the PlayStation game Gran Turismo, faithfully reproduce the non-linear control challenges of real race cars while also encapsulating the complex multi-agent interactions. Here we describe how we trained agents for Gran Turismo that can compete with the world’s best e-sports drivers. We combine state-of-the-art, model-free, deep reinforcement learning algorithms with mixed-scenario training to learn an integrated control policy that combines exceptional speed with impressive tactics.

In addition, we construct a reward function that enables the agent to be competitive while adhering to racing’s important, but under-specified, sportsmanship rules. We demonstrate the capabilities of our agent, Gran Turismo Sophy, by winning a head-to-head competition against four of the world’s best Gran Turismo drivers. By describing how we trained championship-level racers, we demonstrate the possibilities and challenges of using these techniques to control complex dynamical systems in domains where agents must respect imprecisely defined human norms.

What You’ll Learn: We demonstrate the possibilities and challenges of using deep RL techniques to control complex dynamical systems in domains such as Gran Turismo where agents must respect imprecisely defined human norms.

Track: Technical / Research

Technical Level: 7

Location: Ottawa

Talk: Outracing Champion Gran Turismo Drivers With Deep Reinforcement Learning

Presenter:
Varun Raj Kompella, Senior Research Scientist, Sony AI

About the Speaker:
Varun Kompella is currently a senior research scientist at Sony AI. He earned his master’s of science degree in informatics with a specialization in graphics, vision and robotics from Institut Nationale Polytechnique de Grenoble (INRIA Grenoble), and a Ph.D degree from Università della Svizzera Italiana (IDSIA Lugano), Switzerland, working with Prof. Juergen Schmidhuber. In his thesis work he developed algorithms that use the slowness principle for driving exploration in reinforcement learning agents. After completing his Ph.D., he worked as a postdoctoral researcher at the Institute for Neural Computation (INI), Germany. His research contributions led to several patents, publications in peer-reviewed journals and conference proceedings.

Which talk track does this best fit into?
Technical / Research

Technical level of your talk?
(Technical Level: 7/7)

What you’ll learn:
We demonstrate the possibilities and challenges of using deep RL techniques to control complex dynamical systems in domains such as Gran Turismo where agents must respect imprecisely defined human norms.

Abstract of Talk:
Many potential applications of artificial intelligence involve making real-time decisions in physical systems while interacting with humans. Automobile racing represents an extreme example of these conditions; drivers must execute complex tactical manoeuvres to pass or block opponents while operating their vehicles at their traction limits. Racing simulations, such as the PlayStation game Gran Turismo, faithfully reproduce the non-linear control challenges of real race cars while also encapsulating the complex multi-agent interactions. Here we describe how we trained agents for Gran Turismo that can compete with the world’s best e-sports drivers. We combine state-of-the-art, model-free, deep reinforcement learning algorithms with mixed-scenario training to learn an integrated control policy that combines exceptional speed with impressive tactics.

Jinen Setpal

Machine Learning Engineer, DagsHub

I’m a second-year undergraduate studying Data Science at Purdue University. I also work part-time as a Machine Learning Engineer @ DagsHub.

I love research – especially within academia! My interests lie firmly within Machine Vision, NLP & Cybersecurity; so far, I’ve published some peer-reviewed papers and have pending patents within these domains.

Talk: Interpretability Tools are Feedback Loops

Abstract: Fundamentally – Machine Learning as a field is designed to emulate the way humans think; hence, *neural* networks. When we train our models, we use optimizers and loss functions to measure their success. While these functions make sense mathematically, they are far from intuitive or explaining what happened behind the scenes. It’s hard to pick the correct functions, and performing huge grid searches to hyperparameter tune at scale is as logical as bruteforcing an SHA-256 hash.

On the other hand – Interpretability techniques can’t really be used in a training context but are intuitive in helping us understand how a given model interprets a set of data.

This talk aims to bridge the gap between the two, connecting them within a single training loop to maximize training effectiveness without disproportionately increasing compute or training time. Making training intuitive to how humans learn should help develop models that actually work, without resorting to “useless” training.

I aim to showcase – with a practical demonstration – learning techniques to build feedback loops wherein interpretability is used to better optimize a training sequence. I also aim to discuss how this carries forward to complex architectures, and a potential approach for their relevant implementation.

Structurally: the talk would provide an overview on machine interpretability, provide a brief overview on optimizers and loss functions before jumping into the implementation walkthrough of a case study. The case study uses TensorFlow, but can generally be applied to any desired framework.

What You’ll Learn: If, by the end my presentation, attendees are able to identify techniques for applying the proposed approach within their internal systems, or find themselves motivated to further research the ideas presented, I’d consider the talk a success.

Track: Technical

Technical Level: 6

Location: West Lafayette, Indiana, United States

Talk: Interpretability Tools are Feedback Loops

Presenter:
Jinen Setpal, Machine Learning Engineer, DagsHub

About the Speaker:
I’m a second-year undergraduate studying Data Science at Purdue University. I also work part-time as a Machine Learning Engineer @ DagsHub.

Which talk track does this best fit into?
Technical / Research

Technical level of your talk?
(Technical level: 6/7)

Are there any industries (in particular) that are relevant for this talk?
Computer Software, Researchers within Academia

Who is this presentation for?
Data Scientists/ ML Engineers, ML Engineers, Researchers

What you’ll learn:
Most guides on Interpretable Machine Learning, while focusing on the seemingly black-box nature of models, fail to address the challenge of it only working post-training. It’s addressed as a final check tool to ensure it isn’t learning tragically differently from what the researcher expects, but not directly as part of the training feedback loop. I aim to address this.

What are the main core message (learning) you want attendees to take away from this talk?
If, by the end my presentation, attendees are able to identify techniques for applying the proposed approach within their internal systems, or find themselves motivated to further research the ideas presented, I’d consider the talk a success.

Pre-requisite Knowledge:
It’s a technical presentation, and requires participants to be familiar on a high level with model optimization techniques and machine interpretability. They should also be very familiar with the standard classification pipeline of a feedforward neural network.

What is unique about this speech, from other speeches given on the topic?
I work extensively on ML Reproducibility. In fact, it is the fundamental work done within my research grant funded by Google. Time and again, I find that papers tend to document the end-result and the problem statement extensively, but not everything in between. That’s where the real learning takes place, understanding what DIDN’T work and WHY.

This is rarely ever documented. Papers list working parameters, both functions and hyperparameters, but fail to explain why. I’ve struggled a lot with this and hope to relay potential solutions through the extent of my presentation.

Abstract of Talk:
Fundamentally – Machine Learning as a field is designed to emulate the way humans think; hence, *neural* networks. When we train our models, we use optimizers and loss functions to measure their success. While these functions make sense mathematically, they are far from intuitive or explaining what happened behind the scenes. It’s hard to pick the correct functions, and performing huge grid searches to hyperparameter tune at scale is as logical as bruteforcing an SHA-256 hash.

On the other hand – Interpretability techniques can’t really be used in a training context but are intuitive in helping us understand how a given model interprets a set of data.

Can you suggest 2-3 topics for post-discussion?
Information Retreival, Explainable AI, relevance of the above in product development

Jesse Cresswell

Senior Machine Learning Scientist, Layer 6 AI

Jesse is a Senior Machine Learning Scientist at Layer 6 AI within TD, and is the Team Lead for Credit Risk. His applied work centers on building machine learning models in high risk and highly regulated domains. Jesse leads research on privacy enhancing technologies for machine learning including topics of Federated Learning and Differential Privacy.

Talk: Navigating the Tradeoff Between Privacy and Fairness in ML

Abstract: As machine learning becomes more widespread throughout society, aspects including data privacy and fairness must be carefully considered, and are crucial for deployment in highly regulated industries. Unfortunately, the application of privacy enhancing technologies often worsens unfair tendencies in models. In this talk we address the intersection of privacy and fairness in machine learning, and offer research-based solutions for navigating the tradeoffs.

What You’ll Learn: Applying privacy enhancing technologies can increase bias and unfairness in ML models. Practitioners need to consider the intersection of these important ethical ideas.

Track: Technical

Technical Level: 5

Location: East York

Talk: Navigating the Tradeoff Between Privacy and Fairness in ML

Presenter:
Jesse Cresswell, Senior Machine Learning Scientist, Layer 6 AI

About the Speaker:
Jesse is a Senior Machine Learning Scientist at Layer 6 AI within TD, and is the Team Lead for Credit Risk. His applied work centers on building machine learning models in high risk and highly regulated domains. Jesse leads research on privacy enhancing technologies for machine learning including topics of Federated Learning and Differential Privacy.

Which talk track does this best fit into?
Technical

Technical level of your talk?
(Technical Level: 5/7)

What you’ll learn:
Applying privacy enhancing technologies can increase bias and unfairness in ML models. Practitioners need to consider the intersection of these important ethical ideas.

Abstract of Talk:
As machine learning becomes more widespread throughout society, aspects including data privacy and fairness must be carefully considered, and are crucial for deployment in highly regulated industries. Unfortunately, the application of privacy enhancing technologies often worsens unfair tendencies in models. In this talk we address the intersection of privacy and fairness in machine learning, and offer research-based solutions for navigating the tradeoffs.

Sophia Yang

Senior Data Scientist, Anaconda

Sophia Yang is a Senior Data Scientist at Anaconda, Inc., where she uses data science to facilitate decision-making for various departments across the company. She volunteers as a Project Incubator at NumFOCUS to help Open Source Scientific projects grow. She is also the author of multiple Python open-source libraries such as condastats, cranlogs, PyPowerUp, intake-stripe, and intake-salesforce. She holds an M.S. in Statistics and Ph.D. in Educational Psychology from The University of Texas at Austin.

Talk: PyScript for Data Science

Abstract: Are you a data scientist or a developer who mostly uses Python? Are you jealous of developers who write Javascript code and build fancy websites in a browser? How nice would it be if we can write websites in Python? PyScript makes it possible! PyScript allows users to write Python in the browser. In this talk, I will introduce PyScript and discuss what does PyScript mean for data scientists, how PyScript might change the way data scientists work, and how PyScript can be incorporated into the data science workflow.

What You’ll Learn: Use PyScript to run Python in Your HTML

Track: Technical

Technical Level: 4

Location: Austin

Talk: PyScript for Data Science

Presenter:
Sophia Yang, Senior Data Scientist, Anaconda

About the Speaker:
Sophia Yang is a Senior Data Scientist at Anaconda, Inc., where she uses data science to facilitate decision-making for various departments across the company. She volunteers as a Project Incubator at NumFOCUS to help Open Source Scientific projects grow. She is also the author of multiple Python open-source libraries such as condastats, cranlogs, PyPowerUp, intake-stripe, and intake-salesforce. She holds an M.S. in Statistics and Ph.D. in Educational Psychology from The University of Texas at Austin.

Which talk track does this best fit into?
Technical / Research

Technical level of your talk?
(Technical level: 3/7)

Are there any industries (in particular) that are relevant for this talk?
Computer Software, Information Technology & Service

Who is this presentation for?
Senior Business Executives, Product Managers, Data Scientists/ ML Engineers and High-level Researchers, Product Managers, Data Scientists/ ML Engineers

What you’ll learn:
PyScript was announced early this year. There is not much tutorial and content online yet.

Pre-requisite Knowledge:
Knowledge of Python is recommended

What is unique about this speech, from other speeches given on the topic?
I will introduce PyScript from a data science perspective.

Abstract of Talk:
Are you a data scientist or a developer who mostly uses Python? Are you jealous of developers who write Javascript code and build fancy websites in a browser? How nice would it be if we can write websites in Python? PyScript makes it possible! PyScript allows users to write Python in the browser. In this talk, I will introduce PyScript and discuss what does PyScript mean for data scientists, how PyScript might change the way data scientists work, and how PyScript can be incorporated into the data science workflow.

Can you suggest 2-3 topics for post-discussion?
Visualization, Model Deployment

Danielle Goldfarb

Vice President and General Manager of Global Affairs, Economics and Public Policy, RIWI (Real-Time Interactive Worldwide Intelligence)

Opening speaker, Tedx Toronto: The Smartest Way to Predict the Future; Developing a U of Toronto course on new data tools for global affairs / public policy; see also linkedin

Talk: The Risks of Excluding the Disengaged From your Dataset

Abstract: Analysts use machine learning tools to analyze the vast amounts of data now available. Big data is appealing, but even the best tools and techniques will give you noise if the underlying data aren’t robust and inclusive. To reliably predict elections, correctly anticipate consumer demand or to accurately predict the trajectory of a pandemic, leaders need to ask themselves who is (and isn’t) included in their dataset.

What You’ll Learn: There is a huge emphasis on tools to analyze big data but we need to spend at least as much time on the quality of the underlying dataset. If, for example, we exclude disengaged populations, as some common methods do, we risk getting things wrong, whether understanding a market’s potential, predicting an election result, or anticipating the trajectory of the pandemic, the war, or the economy.

Track: Business

Technical Level: 1

Location: Toronto

Talk: The Risks of Excluding the Disengaged From your Dataset

Presenter:
Danielle Goldfarb, Vice President and General Manager of Global Affairs, Economics and Public Policy, RIWI (Real-Time Interactive Worldwide Intelligence)

About the Speaker:
Opening speaker, Tedx Toronto: The Smartest Way to Predict the Future; Developing a U of Toronto course on new data tools for global affairs / public policy; see also linkedin

Which talk track does this best fit into?
Business Strategy

Technical level of your talk?
(Technical level: 1/7)

What you’ll learn:
There is a huge emphasis on tools to analyze big data but we need to spend at least as much time on the quality of the underlying dataset. If, for example, we exclude disengaged populations, as some common methods do, we risk getting things wrong, whether understanding a market’s potential, predicting an election result, or anticipating the trajectory of the pandemic, the war, or the economy.

Abstract of Talk:
Analysts use machine learning tools to analyze the vast amounts of data now available. Big data is appealing, but even the best tools and techniques will give you noise if the underlying data aren’t robust and inclusive. To reliably predict elections, correctly anticipate consumer demand or to accurately predict the trajectory of a pandemic, leaders need to ask themselves who is (and isn’t) included in their dataset.

Danielle Goldfarb

Vice President and General Manager of Global Affairs, Economics and Public Policy, RIWI (Real-Time Interactive Worldwide Intelligence)

Opening speaker, Tedx Toronto: The Smartest Way to Predict the Future; Developing a U of Toronto course on new data tools for global affairs / public policy; see also linkedin

Talk: The Risks of Excluding the Disengaged From your Dataset

Track: Business

Technical Level: 1

Location: Toronto

Talk: Managing AI and the Associated Risks in a Regulatory Environment

Presenters:
David Van Bruwaene, Founder and CEO, Fairly AI & Fion Lee-Madan, Co-Founder and COO, Fairly AI & Susie Lindsay, Counsel, Law Commission of Ontario

About the Speakers:
Fion has over 20+ years of experience in enterprise software as a Solutions Architect (ex-Sapient, ex-Intuit, ex-ATG – acquired by Oracle.) She double majored in Computer Science and Human Biology at the University of Toronto and has an MBA from Boston University. She is a technical committee member of the CIO Strategy Council of Canada. She is a champion of DE&I, and a major supporter for women in tech as both a mentor and coach. She guest lectures for AI Ethics at Lighthouse Labs, an education company with a goal to bring more diversity into the Data Science field.

David has developed a deep understanding of ethics and formal logic, model theory, and Natural Language Processing (NLP) throughout his career in business and academia. He taught Business Ethics at the University of Waterloo and graduate level Ethics at Cornell and is a sought-after speaker for AI Ethics, Compliance and Risk Management at conferences around the world. On an exchange scholarship from Cornell to Berkeley, he became fascinated with powerful Natural Language Processing. David applied this background to cyberbullying and related NLP in his first AI startup ViSR (acquired by SafetoNet) where he was the Head of Data Science, then became the CEO and Board Member. His unique perspective as a practicing Data Scientist to being the CEO and Board Member at the top made David conscious of tensions between technical vs business decisions and the impact on people’s lives in the resulting automation of decision-making.

Susie is Counsel at the Law Commission of Ontario where she leads numerous LCO projects including: AI and the Civil Justice System, Protection Orders, and the LCO’s joint initiative on AI and Human Rights with the Ontario Human Rights Commission and Canadian Human Rights Commission. Before joining the Commission Susie practised regulatory law at a large communications company, and civil defence litigation at a boutique litigation firm. Susie is a graduate of Queen’s Law School, a Fulbright Scholar, has a Master of Laws from Harvard Law School and was a fellow at the Berkman Klein Centre for Internet & Society at Harvard University.

Which talk track does this best fit into?
Business Strategy

Technical level of your talk?
(Technical level: 2/7)

What you’ll learn:
Latest expert views on managing AI risks in the fast changing regulatory environment

Abstract of Talk:
Artificial Intelligence (AI) is a tool that can be used both to manage risks in high-stake industries such as financial services as well as poses some risks by itself. This expert panel will discuss AI yesterday versus AI today, focusing on how AI has evolved and developed from simple automation to complex decisioning. Our experts will cover AI regulatory trends globally and in the US, best practices in risk and compliance management, and complementing technologies that can be utilized to counter new and emerging AI risks.

Panel: Bias and Relevance in Application of ML in Healthcare

Presenters:
Shazia Akbar, Lead Machine Learning Engineer, Altis Labs & Ali Madani, Director of Machine Learning, Cyclica & Santosh Hariharan, Principal Scientist, Pfizer & Shiva Amiri, VP, Head of AI and Data Intelligence, Pivotal Life Sciences & Javier Diaz-Mejia, Head of Data Science, Phenomic AI

About the Speakers:
Dr. Shazia Akbar is the lead machine learning engineer at Altis Labs, a Toronto-based startup which leverages deep learning technologies to gain prognostic insight from medical imaging data. Since joining Altis Labs, Shazia has designed and developed artificial intelligent systems which ingest millions of imaging data to predict patient outcomes. Some of the applications she has developed to date include a fully automated model to quantify mortality risk in early stage lung cancer patients, and an x-ray model which determines in-patient admission risk of hospital patients diagnosed with community acquired pneumonia.

Shazia gained her PhD from the University of Dundee, UK, after which she joined the department of Radiology at New York University, US. In 2018, Shazia completed her postdoctoral fellowship at Sunnybrook Research Institute and the Vector Institute, designing novel deep learning algorithms for digital pathology. Her research interests include explainable AI, weakly supervised learning and applications of AI in healthcare.

Ali Madani leads ML technology development at Cyclica, a leading Canadian biotechnology company focused on AI based drug discovery. He is also editor of special Topic Artificial Intelligence in Cancer Diagnosis and Therapy at MDPI and work as an AI educator with companies like WeCloudData. Ali is a PhD graduate of University of Toronto; an alumni of University of Waterloo School of Engineering; and, attained a master of mathematics from the University of Waterloo. He is an active member of the machine learning community in Toronto and speaks in world and Canada wide conferences, webinars and workshops about technology development, machine learning, drug discovery and cancer therapeutics. He has also published more than 20 scientific articles in high impact factor journals on these subjects.

Santosh Hariharan is a Principal Scientist at Pfizer, committed to curing disease and improving patient lives. He enjoys solving complex biological problems using simple blocks with a motto of “”Seeing is Believing””. He develops and analyzes complex biology by looking at individual cells, evaluating their response to drugs/genetic perturbation and developing predictive models using AI and machine learning (Phenotyping).

Shiva Amiri is the VP, Head of AI and Data Intelligence at Pivotal Life Sciences, working at the intersection of computing and biology with experience in large-scale, multi-stakeholder technology development in data science and biology with a focus on computational biology, bioinformatics, machine learning, and big data systems. Shiva is a team builder and entrepreneurial in cutting edge computational methods in biology, digital health and medical research with a track record in big data/data science and program execution and strategy.

Javier is a data scientist with 15+ years of experience in projects aiming to solve problems of relevance for human health. He has experience in the academic, non-profit and industry sectors in Mexico, USA and Canada. Javier’s role involves identifying organization data science needs, building teams to implement solutions addressing those needs, and serving as a bridge between technical and executive stakeholders. Javier made his PhD studies in Mexico, postdoctoral training in Toronto and currently, works as Head of Data Science at Phenomic AI, a biotech startup developing machine learning tools to fight cancer.

Which talk track does this best fit into?
Business

Technical level of your talk?
(Technical level: 4/7)

What you’ll learn:
Examples of application of ML in healthcare; Impact of ML in healthcare technologies on patients; Potential biases in ML in Healthcare technologies

Abstract of Talk:
Machine Learning (ML) technologies learn how to accomplish tasks and identify patterns from available data. This data-based learning has been critical in helping the healthcare institute, biotechnology and pharmaceutical companies in developing new technologies to improve processes like disease diagnosis from radiological and pathological images and drug design. In this panel, we will discuss how companies design ML technologies that are not only in-line with their business model but eventually will impact patients and healthcare systems. We will also discuss the technological and sociological biases that need to be taken into account in the design of such technologies.

Muhammad Mamdan

Unity Health Toronto – VP: Data Science and Advanced Analytics; Director: Temerty Centre for Artificial Intelligence Research and Education in Medicine of the University of Toronto; Professor – University of Toronto

Dr. Mamdani is Vice President of Data Science and Advanced Analytics at Unity Health Toronto and Director of the University of Toronto Temerty Faculty of Medicine Centre for Artificial Intelligence Research and Education in Medicine (T-CAIREM). Dr. Mamdani’s team bridges advanced analytics including machine learning with clinical and management decision making to improve patient outcomes and hospital efficiency. Dr. Mamdani is also Professor in the Department of Medicine of the Temerty Faculty of Medicine, the Leslie Dan Faculty of Pharmacy, and the Institute of Health Policy, Management and Evaluation of the Dalla Lana Faculty of Public Health. He is also adjunct Senior Scientist at the Institute for Clinical Evaluative Sciences (ICES) and a Faculty Affiliate of the Vector Institute. In 2010, Dr. Mamdani was named among Canada’s Top 40 under 40. He has published over 500 studies in peer-reviewed medical journals. Dr. Mamdani obtained a Doctor of Pharmacy degree (PharmD) from the University of Michigan (Ann Arbor) and completed a fellowship in pharmacoeconomics and outcomes research at the Detroit Medical Center. During his fellowship, Dr. Mamdani obtained a Master of Arts degree in Economics from Wayne State University in Detroit, Michigan with a concentration in econometric theory. He then completed a Master of Public Health degree from Harvard University in 1998 with a concentration in quantitative methods.

Talk: Saving Lives with ML: Applications and Learnings

Abstract: Machine learning (ML) has transformed numerous industries but its application in healthcare has been limited. ML applications are expected to permeate healthcare in the near future with a recent explosion in academic and commercial activity. The application of ML in healthcare, however, is complicated by a variety of factors including the significant variability in needs, healthcare settings and patients served in these settings, workflows, and available resources. This talk will present a case study of Unity Health Toronto and its journey in developing and deploying numerous ML solutions into clinical practice, including bridging public and private sector partnerships to spread innovations internationally. The talk will also present a novel Canadian academic centre dedicated to artificial intelligence (AI) in medicine – the Temerty Centre for Artificial Intelligence Research and Education in Medicine (T-CAIREM) at the University of Toronto.

What You’ll Learn: The successful application of ML in healthcare is multifaceted and highly dependent on end-user engagement.
Innovative public-private partnerships are needed to spread ML applications globally.

Multidisciplinary, collaborative efforts will fuel innovations in the development and application of ML in healthcare.

Track: Case Study

Technical Level: 3

Location: Toronto

Talk: Saving Lives with ML: Applications and Learnings

Presenter:
Muhammad Mamdani, Unity Health Toronto – VP: Data Science and Advanced Analytics; Director: Temerty Centre for Artificial Intelligence Research and Education in Medicine of the University of Toronto; Professor – University of Toronto

About the Speaker:
Dr. Mamdani is Vice President of Data Science and Advanced Analytics at Unity Health Toronto and Director of the University of Toronto Temerty Faculty of Medicine Centre for Artificial Intelligence Research and Education in Medicine (T-CAIREM). Dr. Mamdani’s team bridges advanced analytics including machine learning with clinical and management decision making to improve patient outcomes and hospital efficiency.
Dr. Mamdani is also Professor in the Department of Medicine of the Temerty Faculty of Medicine, the Leslie Dan Faculty of Pharmacy, and the Institute of Health Policy, Management and Evaluation of the Dalla Lana Faculty of Public Health. He is also adjunct Senior Scientist at the Institute for Clinical Evaluative Sciences (ICES) and a Faculty Affiliate of the Vector Institute. In 2010, Dr. Mamdani was named among Canada’s Top 40 under 40. He has published over 500 studies in peer-reviewed medical journals.
Dr. Mamdani obtained a Doctor of Pharmacy degree (PharmD) from the University of Michigan (Ann Arbor) and completed a fellowship in pharmacoeconomics and outcomes research at the Detroit Medical Center. During his fellowship, Dr. Mamdani obtained a Master of Arts degree in Economics from Wayne State University in Detroit, Michigan with a concentration in econometric theory. He then completed a Master of Public Health degree from Harvard University in 1998 with a concentration in quantitative methods.

Which talk track does this best fit into?
Case Study

Technical level of your talk?
(Technical level: 3/7)

Are there any industries (in particular) that are relevant for this talk?
Hospital & Health Care

Who is this presentation for?
The successful application of ML in healthcare is multifaceted and highly dependent on end-user engagement.
Innovative public-private partnerships are needed to spread ML applications globally.
Multidisciplinary, collaborative efforts will fuel innovations in the development and application of ML in healthcare.

Abstract of Talk:
Machine learning (ML) has transformed numerous industries but its application in healthcare has been limited. ML applications are expected to permeate healthcare in the near future with a recent explosion in academic and commercial activity. The application of ML in healthcare, however, is complicated by a variety of factors including the significant variability in needs, healthcare settings and patients served in these settings, workflows, and available resources. This talk will present a case study of Unity Health Toronto and its journey in developing and deploying numerous ML solutions into clinical practice, including bridging public and private sector partnerships to spread innovations internationally. The talk will also present a novel Canadian academic centre dedicated to artificial intelligence (AI) in medicine – the Temerty Centre for Artificial Intelligence Research and Education in Medicine (T-CAIREM) at the University of Toronto.

Nikita Medvedev

Director of Advanced Analytics, Coca Cola

Nikita has over 10 years of experience in the Retail and Consumer Packaged Goods industries, working for companies like Loblaw and Sears. He is also an alumnus of the Master of Management Analytics program from Queen’s University, and holds a Bachelor of Finance & Economics degree from University of Toronto

Co-Presenter: Winston Li

Talk: The Application of Mobile Location Data for Vending Machine Site Selection and Revenue Optimization.

Abstract: In this presentation, we present an innovative approach to utilizing mobility data to optimize the placement of vending machines in Canada. Coca-Cola has more than 10k vending machines in various locations and their ROI heavily depends on the amount of foot traffic next to them as well as who those people are. For this use case, we’ll be concentrating on using the super detailed mobility data to understand the difference between our best machines and worst at scale, and optimizing their location based on the mobility data to increase the ROI. In addition to the practical and business application, we’ll also be able to share the algorithms used and the tech stack with the audience.

What You’ll Learn: Mobility data as an alternative data source for consumer related analytics and its recency and granularity and really drive measurable business outcomes.

Track: Case Study

Technical Level: 4

Location: Toronto

Talk: The Application of Mobile Location Data for Vending Machine Site Selection and Revenue Optimization

Presenters:
Nikita Medvedev, Director of Advanced Analytics & Winston Li, Founder, Coca Cola & Arima

About the Speaker:
Winston is the founder of Arima, a Canadian based startup that provides consumer data to its users. Our flagship product, the Synthetic Society, is a privacy-by-design, individual level database that mirrors the real society. Built using trusted sources like census, market research, mobility and purchase patterns, it contains 10k+ attributes across North America and enables advanced modelling at the most granular level.

Prior to founding Arima, Winston was the Director of Data Science at PwC and Omnicom. Winston is also a part-time faculty member at Northeastern University Toronto and sits on the advisory board of the Master of Analytics program.

Nikita is the Director of Advanced Analytics at Coca-Cola Canada Bottling Limited. Together with his team he is transforming terabytes of business operations data into actionable insights to drive growth and innovate in the Consumer Packaged Goods industry. He loves finding novel solutions to old problems and is obsessed with driving real lasting change through better use of data.

Which talk track does this best fit into?
Case Study

Technical level of your talk?
(Technical level: 4/7)

Are there any industries (in particular) that are relevant for this talk?
Food & Beverages, Information Technology & Service, Marketing & Advertising

What are the main core message (learning) you want attendees to take away from this talk?
Mobility data as an alternative data source for consumer related analytics and its recency and granularity and really drive measurable business outcomes.

Abstract of Talk:
In this presentation, we present an innovative approach to utilizing mobility data to optimize the placement of vending machines in Canada. Coca-Cola has more than 10k vending machines in various locations and their ROI heavily depends on the amount of foot traffic next to them as well as who those people are. For this use case, we’ll be concentrating on using the super detailed mobility data to understand the difference between our best machines and worst at scale, and optimizing their location based on the mobility data to increase the ROI. In addition to the practical and business application, we’ll also be able to share the algorithms used and the tech stack with the audience.

Shiming Ren

Senior Engineering Manager – Safety, MLOps and Infrastructure, Amazon/Twitch

I worked as a Software Engineer Manager at Twitch about MLOps and Tooling in Safety team. Here is my linkedin. I spoke at Meta’s At Scale about Scaling ML Workflows for Real-Time Moderation Challenges at Twitch, I also spoke at TwitchCon about Integrating Data into Twitch at Scale. I worked in engineering leadership role for 5 years and our team made several company wide MLOps tooling such as orchstration and feature store.

Co-Presenter: Chen Liu

Talk: From Silo to Collaboration – Building Tooling to Support Distributed ML Teams at Twitch

Abstract: In this talk, we will cover Twitch’s current ML team structure and challenges of it. Then we dive deep into some solutions we have built to support ML development at Twitch, including what they are and how they will benefit the situation. We close with a discussion of Twitch’s distributed ML team style and how we collaborate using Conductor as an example.

ML has been playing a more and more important role in Twitch’s products (e.g. Recommendation, Safety). In order to allow products to iterate fast, we keep ML practitioners in the product teams and empower the teams to work independently. Undoubtedly, there are common challenges in ML development regardless of product areas. So we are striving to develop tooling and infrastructures for general ML development in order to reduce duplicate work across ML teams. We will dive into those efforts we made in this presentation. For example, Twitch machine learning feature store is developed to have a single control plane serving as feature registry but facilitates distributed feature ownership (e.g. storage, pipelines). Conductor, a in-house ML orchestration system, promotes best practices in pipeline management with templated process control flow and distributed infrastructure management. Meanwhile, we are promoting collaborative ML culture among Twitch engineering teams. It is similar to community owned open source projects where teams share the same interests and encourage cross team contribution and development.

What You’ll Learn: Twitch’s strategy of scaling our ML infra and MLOps tooling has never been discussed online. And we aim to help audience figure out the best strategy to utilize ML tooling for enhancing collaborations between ML teams and boost scientists self-service / efficiency. This is a good lesson if companies are seeking to start MLOps from stratch.

Track: Case Study

Technical Level: 4

Talk: From Silo to Collaboration - Building Tooling to Support Distributed ML Teams at Twitch

Presenters:
Shiming Ren, Sr. Engineering Manager – Safety, MLOps and Infrastructure & Chen Liu, Twitch Sr. Engineering Manager on Personalization and ML Infra, Amazon/Twitch

About the Speaker:
I Shiming worked as a Software Engineer Manager at Twitch about MLOps and Tooling in Safety team. Here is my linkedin. I spoke at Meta’s At Scale about Scaling ML Workflows for Real-Time Moderation Challenges at Twitch, I also spoke at TwitchCon about Integrating Data into Twitch at Scale. I worked in engineering leadership role for 5 years and our team made several company wide MLOps tooling such as orchstration and feature store.

Chen is currently supporting teams working on personalization and ML infrastructures at Twitch. He is passionate about building scalable ML products and democratizing ML in the organization.

Which talk track does this best fit into?
Technical / Research

Technical level of your talk?
(Technical level: 4/7)

Are there any industries (in particular) that are relevant for this talk?
Computer Software, Information Technology & Service

Who is this presentation for?
Senior Business Executives, Product Managers, Data Scientists/ ML Engineers and High-level Researchers, Data Scientists/ ML Engineers, ML Engineers

What you’ll learn:
Twitch’s strategy of scaling our ML infra and MLOps tooling has never been discussed online. And we aim to help audience figure out the best strategy to utilize ML tooling for enhancing collaborations between ML teams and boost scientists self-service / efficiency. This is a good lesson if companies are seeking to start MLOps from stratch.

On a scale of 1-10 how mature is this applied AI application you plan to discuss?
7/10

Pre-requisite Knowledge:
Feature store, Orchstration, Large Scale Data Handling

What kind of DevOps tools you plan to discuss? Open source?
N/A Our tools are all in house

What are some of the languages you plan to discuss?
Python, Golang

What are some of the infrastructures you plan to discuss?
Feature Store, ML Orchstration, Realtime Inference, Distributed ML team collaborations

What is unique about this speech, from other speeches given on the topic?
We aim to use examples how Twitch build in house feature store, realtime inference and orchstration system to demonstrate from technology perspective about MLOps collaborations in a company. This is more like a hybrid tech and management talk which will benefit both engineer and leadership groups.

Abstract of Talk:
[High level intro]
In this talk, we will cover Twitch’s current ML team structure and its challenges of it. Then we dive deep into some solutions we have built to support ML development at Twitch, including what they are and how they will benefit the situation. We close with a discussion of Twitch’s distributed ML team style and how we collaborate using Conductor as an example.

[Actual abstract]
ML has been playing a more and more important role in Twitch’s products (e.g. Recommendation, Safety). In order to allow products to iterate fast, we keep ML practitioners in the product teams and empower the teams to work independently. Undoubtedly, there are common challenges in ML development regardless of product areas. So we are striving to develop tooling and infrastructures for general ML development in order to reduce duplicate work across ML teams. We will dive into those efforts we made in this presentation. For example, Twitch machine learning feature store is developed to have a single control plane serving as feature registry but facilitates distributed feature ownership (e.g. storage, pipelines). Conductor, a in-house ML orchestration system, promotes best practices in pipeline management with templated process control flow and distributed infrastructure management. Meanwhile, we are promoting collaborative ML culture among Twitch engineering teams. It is similar to community-owned open source projects where teams share the same interests and encourage cross team contribution and development.

Can you suggest 2-3 topics for post-discussion?
Manage ML teams collaboration in a distributed manner; ML tooling development from 0 to 10; Implementation details for feature store and ML orchestration system.

Valerii Podymov

Lead Data Scientist, FreshBooks

Valerii joined FreshBooks a year ago to lead and grow a team of Data Scientists and Machine Learning Engineers. He has an experience in multiple industries ranging from Electronics to Clean Tech and has contributed to the development of innovative solutions for a variety of brands such as LG Electronics, Panasonic, Samsung, Toyota, Scotiabank, Cineplex. He has a University Degree in Telecom Engineering and PhD in Automated Control Systems. Author of 20 patented inventions in Signal Processing, Electronics and Computing.

Talk: Builidng a Fully Automated ML Platform Using Kubeflow and Declarative Approach to Development of End-to-End ML Pipelines

Abstract: Recent innovations in the ML ecosystem have seen the emergence of operationally-focused technology like declarative systems and data-centric AI. These techniques appear to be a radical change for AI practitioners, who can now more simply frame use cases and manage workflows. In this talk, we’ll take a look at the history of AI to see the progress that has been made and how we’ve arrived at where we are now. How are high-tech companies handling AI initiatives internally, and why aren’t we all copying them? Has MLOps been the promised solution to simplifying deployment and monitoring of production AI? How do we create a simpler paradigm for operationalizing AI? All these questions and more will be addressed.

What You’ll Learn: A journey to higher levels of MLOps maturity is unique for any company and has no recipes due to experimental nature of MLOps. Many insights and ideas in this area are the results of investments by big names (Google, Microsoft, Amazon) and knowledge sharing between smaller companies like us working on similar problems. We are grateful for this opportunity to contribute to the ecosystem so that others can learn from us.

Track: Case Study

Technical Level: 6

Location: Toronto

Talk: Building a Fully Automated ML Platform Using Kubeflow and Declarative Approach to Development of End-to-End ML Pipelines

Presenters:
Valerii Podymov, Lead Data Scientist, FreshBooks & Roshan Isaac, Machine Learning Engineer, FreshBooks & Vlad Ryzhkov, Senior Data Engineer, FreshBooks & Joey Zhou, Senior Data Engineer, FreshBooks

About the Speaker:
Valerii joined FreshBooks a year ago to lead and grow a team of Data Scientists and Machine Learning Engineers. He has an experience in multiple industries ranging from Electronics to Clean Tech and has contributed to the development of innovative solutions for a variety of brands such as LG Electronics, Panasonic, Samsung, Toyota, Scotiabank, Cineplex. He has a University Degree in Telecom Engineering and PhD in Automated Control Systems. Author of 20 patented inventions in Signal Processing, Electronics and Computing.

Roshan works as a Machine Learning Engineer at FreshBooks where he is building ML Platform on Vertex AI and bringing MLOps best practices to the organization. He was previously at the same role with Cineplex. He has a Bachelor Degree in Computer Science and Engineering and hold graduate certificates in AI & Project Management. Overall he has 8+ years of experience in Machine Learning, Data Analytics and CRM software working in different startups and companies in Canada and India. He published papers in IEEE conferences and was a speaker at Libre Software Meeting (LSM), France.

Vlad joined FreshBooks a year ago with extensive Data Engineering background and he works on building ML Platform bringing best practices in large-scale data processing to the company. He has a PhD in System Analysis, Management and Information Processing. Overall, his 15+ years of software development experience comprises such areas as financial systems, e-commerce, e-sport and airlines in Canada and overseas.

Joey joined FreshBooks three months ago and works on the continuous monitoring framework for the ML team. Before, he had an experience in the tech industry, ranging from social-dating to e-commerce, in multiple roles such as Data Scientist and Machine Learning Engineer. He built a recommender systems for one of the largest e-commerce platforms in China. With hands-on experience in building and productionizing ML models, he is ready to pursue his passion for MLOps at FreshBooks.

Which talk track does this best fit into?
Technical / Research

Technical level of your talk?
(Technical level: 6/7)

Are there any industries (in particular) that are relevant for this talk?
Banking & Financial Services, Computer Software

Who is this presentation for?
Data Scientists/ ML Engineers, ML Engineers

What you’ll learn:
How we tackled existing challenges with Kubeflow pipelines changing the imperative approach to the declarative.

What are the main core message (learning) you want attendees to take away from this talk?
A journey to higher levels of MLOps maturity is unique for any company and has no recipes due to experimental nature of MLOps. Many insights and ideas in this area are the results of investments by big names (Google, Microsoft, Amazon) and knowledge sharing between smaller companies like us working on similar problems. We are grateful for this opportunity to contribute to the ecosystem so that others can learn from us.

On a scale of 1-10 how mature is this applied AI application you plan to discuss?
9/10

Pre-requisite Knowledge:
Machine Learning Lifecycle

What kind of DevOps tools you plan to discuss? Open source?
GitHub Actions, Kubeflow

What are some of the languages you plan to discuss?
Python, SQL

What are some of the infrastructures you plan to discuss?
BigQuery, Airflow, Vertex AI, containers

What is unique about this speech, from other speeches given on the topic?
Managing MLOps is highly immature topic with lack or absence of commonly accepted best practice, so the experience of any company in growing over MLOps maturity levels is always unique.

Abstract of Talk:
This talk is about our journey at FreshBooks from mostly manual processes in productionizing of our ML models to the highest levels of maturity in MLOps. First, we briefly go over a list of challenges we faced when working on the ML platform as a hybrid team of Data Scientists, ML Engineers and Data Ops Engineers. And then we provide more detailed overview of our end-to-end Kubeflow pipelines and a declarative MLOps framework that has been designed to speed up, simplify and improve the reliability of ML pipelines at each stage from development to production. Lessons learned and what’s next will be provided at the end of the talk.

Can you suggest 2-3 topics for post-discussion?
ML Ops, ML Model Governance

Eric Hart

Staff Data Scientist, Anheuser-Busch

Eric is a Staff Data Scientist with more than 7 years of experience working at Altair Engineering and Anheuser-Busch. He has a PhD in probability from the University of Toronto, and a masters degree in Applied Math and an undergraduate degree in Engineering from Queen’s university. He’s also a world champion Blokus player.

Talk: Optimal Beer Pricing: An Optimization Layer for Price Elasticities

Abstract: At Anheuser-Busch, we’re obsessed with price elasticities. When the price of beer changes, how will that affect the volume of beer that we sell? These questions (yes, this is more than one question) have implications all over the business, from price setting to procurement to financial planning. We’ve worked hard to make sure our answers to these questions are as data driven as possible. But once we have a model to produce (and predict) these elasticities, how do we make business decisions based on that? And how do we make sure those business decisions are also as data driven as possible?

In this talk we’ll discuss an optimal pricing layer for beer elasticities. We’ll cover how to use mathematical optimization to make specific price change suggestions at a variety of granularities to help achieve specific business objectives. We’ll consider what objective we actually want to optimize (Profit? Revenue? Market Share?) and see how to use constraints to help smooth the trade-off between these objectives. Finally, we’ll investigate how to ensure our price suggestions stay within the regions where the underlying elasticities models make sense.

Ever wanted to see a real-world example of levelling up your analytics from predictive- to prescriptive-, and do so in the context of price setting (or beer drinking)? Now’s your chance!

What You’ll Learn: How to add an optimization layer to ml models.

Track: Case Study

Technical Level: 2

Location: Toronto

Talk: Optimal Beer Pricing: An Optimization Layer for Price Elasticities

Presenter:
Eric Hart, Staff Data Scientist at Anheuser-Busch

About the Speaker:
Eric is a Staff Data Scientist with more than 7 years of experience working at Altair Engineering and Anheuser-Busch. He has a PhD in probability from the University of Toronto, and a masters degree in Applied Math and an undergraduate degree in Engineering from Queen’s university. He’s also a world champion Blokus player.

Which talk track does this best fit into?
Case Study

Technical level of your talk?
(Technical level: 2 /7)

Are there any industries (in particular) that are relevant for this talk?
Banking & Financial Services, Food & Beverages, Marketing & Advertising

Who is this presentation for?
Senior Business Executives, Product Managers, Data Scientists/ ML Engineers and High-level Researchers

What you’ll learn:
Putting a mathematical optimization layer on top of predictive models is still a mostly unused tool in the ML space. It’s very difficult to learn about that from existing resources.

What are the main core message (learning) you want attendees to take away from this talk?
How to add an optimization layer to ml models.

Pre-requisite Knowledge:
Not a lot. We’ll briefly discuss what price-elasticities and mathematical optimization are, but having heard those terms before (with a basic understanding) would help.

What is unique about this speech, from other speeches given on the topic?
I would argue the whole topic is fairly unique (optimization layers for predictive models are not widely used or discussed). In addition, the specifics of trying to work around the realities of the beer industry (especially varying laws about beer pricing across different geographies) add an extra layer of complexity to this already deep problem.

Abstract of Talk:
At Anheuser-Busch, we’re obsessed with price elasticities. When the price of beer changes, how will that affect the volume of beer that we sell? These questions (yes, this is more than one question) have implications all over the business, from price setting to procurement to financial planning. We’ve worked hard to make sure our answers to these questions are as data driven as possible. But once we have a model to produce (and predict) these elasticities, how do we make business decisions based on that? And how do we make sure those business decisions are also as data driven as possible?

Ever wanted to see a real-world example of levelling up your analytics from predictive- to prescriptive-, and do so in the context of price setting (or beer drinking)? Now’s your chance!

Can you suggest 2-3 topics for post-discussion?
Optimization Layers. Price Elasticities.

Jawad Ahmed

Staff Applied Scientist, Loblaw Digital

Jawad currently works as a Staff Applied Scientist at Loblaw Digital, supporting ML teams building personalization and recommender systems for different lines of business of Loblaw Companies.
He has 8 years of industry experience in Applied AI/ML. Previously, he worked at Flipp, Dialpad and McKinsey Solutions.

His areas of interest are using ML research applications to help build products with scalable ML solutions in NLP, Conversational AI, Computer Vision and Recommender Systems. Read more on linkedin

Talk: Solving Product Substitutions, The #1 Problem in Grocery E-Commerce – Through Self-Supervised ML

Abstract:
Background: Loblaw Companies Ltd is the largest grocery retailer in Canada. It operates multiple popular banners with Real Canadian Superstore, No Frills, and T&T being some of the most popular ones. E-commerce of grocery has become a significant part of the business accounting for more than $2 billion in sales per year.

Problem: Shopping for groceries online is an inherently different process than shopping in person. We take for granted the in-store shopper’s ability to make quick decisions on the fly when faced with the issue of product availability.

We fulfill from stores to ensure freshness which has a very dynamic inventory. This makes promises of items collected, sometimes a day or two after the order depending on the customer’s delivery date, affected by many factors – some of which we cannot control. Thus, we need a solution to substitute items that are out of stock at the time of picking to make sure the customer experience is minimally impacted. While shopping at a physical store, a customer can make a suitable choice of an alternative. In the e-commerce process of grocery shopping, either the customer has to make a selection of the substitute, or the Loblaw employee picking the order on behalf of the customer needs a relevant suggestion on the best substitute for the given item, personalized for the given customer.

Loblaw has historical data available on what selection was made by customers from the list of various possible substitute options available for a given item. Additionally, there is data available on the choices made by pickers – the employees who shop at the store to fulfill customers’ orders. This provides us an opportunity to tailor product similarities toward product substitutions that are tied to business metrics.

Solution: We explored multiple solutions to solve this problem. Our most promising solution that we wish to present leverages features extracted from text descriptions and images of products. In this talk, we will discuss how our approach evolved over time and how this cutting-edge self-supervised method is a big improvement over the traditional techniques.

What You’ll Learn: The talk covers the data curation process by which we prepared a benchmarking Products Substitutions dataset using historical human-selected substitutions data at Loblaw.

The audience will learn about self-supervised ML approaches we used to recommend product substitutions benchmarked against the above mentioned products substitutions Testset.

Track: Case Study

Technical Level: 5

Location: Waterloo, ON

Talk: Solving Product Substitutions, A Big Problem in Grocery E-Commerce – Through Self-Supervised ML

Presenter:
Jawad Ahmed, Staff Applied Scientist, Loblaw Digital

About the Speaker:
Jawad currently works as a Staff Applied Scientist at Loblaw Digital, supporting ML teams building personalization and recommender systems for different lines of business of Loblaw Companies.
He has 8 years of industry experience in Applied AI/ML. Previously, he worked at Flipp, Dialpad and McKinsey Solutions.
His areas of interest are using ML research applications to help build products with scalable ML solutions in NLP, Conversational AI, Computer Vision and Recommender Systems. Read more on linkedin.

Which talk track does this best fit into?
Case Study

Technical level of your talk?
(Technical level: 5/7)

What you’ll learn:
The talk covers the data curation process by which we prepared a benchmarking Products Substitutions dataset using historical human-selected substitutions data at Loblaw.

The audience will learn about self-supervised ML approaches we used to recommend product substitutions benchmarked against the above mentioned products substitutions Testset.

Abstract of Talk:
Background: Loblaw Companies Ltd is the largest grocery retailer in Canada. It operates multiple popular banners with Real Canadian Superstore, No Frills, and T&T being some of the most popular ones. E-commerce of grocery has become a significant part of the business accounting for more than $2 billion in sales per year.

Quoc Tien Au

Data Scientist, Manifest Climate

I am a data scientist at Manifest Climate, working on applying machine learning and natural language processing to climate disclosures. Extracting information at scale is paramount to increase transparency in financial markets, so that we can improve decision-making with data-driven climate information.

Talk: Assessing Alignment of Climate Disclosures Using NLP for the Financial Markets

Abstract: Climate-related disclosure is increasing in importance as companies and stakeholders alike aim to reduce their environmental impact and exposure to climate-induced risk. Companies primarily disclose this information in annual or other lengthy documents where climate information is not the sole focus. To assess the quality of a company’s climate-related disclosure, these documents, often hundreds of pages long, must be reviewed manually by climate experts. We propose a more efficient approach to assessing climate-related financial information. We construct a model leveraging TF-IDF, sentence transformers and multi-label k nearest neighbors (kNN). The developed model is capable of assessing alignment of climate disclosures at scale, with a level of granularity and transparency that will support decision-making in the financial markets with relevant climate information.

What You’ll Learn: How an early-stage startup runs machine learning experiments ; takes decisions balancing model performance, model explainability, resource constraints and added business value ; uses deep language models to create the most valuable business opportunities.

Track: Case Study

Technical Level: 5

Location: Toronto

Talk: Assessing Alignment of Climate Disclosures Using NLP for the Financial Markets

Presenters:
Quoc Tien Au, Data Scientist, Manifest Climate & Aysha Cotterill, Data Analyst, Manifest Climate

About the Speakers:
I am a data scientist at Manifest Climate, working on applying machine learning and natural language processing to climate disclosures. Extracting information at scale is paramount to increase transparency in financial markets, so that we can improve decision-making with data-driven climate information.

Which talk track does this best fit into?
Case Study

Technical level of your talk?
(Technical level: 5/7)

What you’ll learn:
How an early-stage startup runs machine learning experiments ; takes decisions balancing model performance, model explainability, resource constraints and added business value ; uses deep language models to create the most valuable business opportunities.

Abstract of Talk:
Climate-related disclosure is increasing in importance as companies and stakeholders alike aim to reduce their environmental impact and exposure to climate-induced risk. Companies primarily disclose this information in annual or other lengthy documents where climate information is not the sole focus. To assess the quality of a company’s climate-related disclosure, these documents, often hundreds of pages long, must be reviewed manually by climate experts. We propose a more efficient approach to assessing climate-related financial information. We construct a model leveraging TF-IDF, sentence transformers and multi-label k nearest neighbors (kNN). The developed model is capable of assessing alignment of climate disclosures at scale, with a level of granularity and transparency that will support decision-making in the financial markets with relevant climate information.

Talk: Marketplace Experimentation at SpotHero

Presenter:
Amish Popli, Data Scientist, SpotHero

About the Speaker:
Amish Popli is passionate about solving challenging business problems using data science and machine learning. He supports multiple departments at SpotHero including, but not limited to, marketing, sales, and product development. He likes data, manipulating it, making it (simulation), modelling it, visualizing it, and yes, even cleaning it. He works with different PMs and engineers in different domains and has brought many successful products from discovery to production.

Which talk track does this best fit into?
Case Study

Technical level of your talk?
(Technical level: 4/7)

Are there any industries (in particular) that are relevant for this talk?
Parking

Who is this presentation for?
Product Managers, Data Scientists/ ML Engineers

What you’ll learn:
Experimentation is a very nebulous topic. There is a lot of companies, articles and research available on google, but each company has its own unique way of running and measuring impact from experiments.

Pre-requisite Knowledge:
Knowledge of basic statistical tests

What is unique about this speech, from other speeches given on the topic?
In my knowledge there is no company in North America doing data science/ML in parking industry. The problems we are solving are present in other industries but parking adds another layer of complexity on top of it.

Abstract of Talk:
SpotHero is the biggest and fastest growing off-street reservation platform in North America. It is a two sided marketplace involving drivers and parking garage owners. The data science team at SpotHero is working on many interesting problems in the areas of dynamic pricing, marketing, ranking etc. One of the key challenges that we face is how we test our machine learning models in production and make sure that the changes we make lead to an improvement in our KPI’s. In this talk, I will focus on how SpotHero runs experiments whenever we make improvements or create a new model to generate prices for our parking spots. I will cover why the general A/B test framework will not work in our scenario, various approaches that we considered and introduce switchback experimentation as an alternative. I will discuss our experiment design and conclude the talk with a result from one of our experiments and our technical architecture.

Can you suggest 2-3 topics for post-discussion?
A/B Tests, Switchback experiments, challenges in running live expeirments

Hien Luu

Head of Machine Learning Platform, DoorDash

Hien Luu is a Sr. Engineering Manager at DoorDash, leading the Machine Learning Platform team. He is particularly passionate about the intersection between Big Data and Artificial Intelligence. He is the author of the Beginning Apache Spark 3 book. He has given presentations at various conferences such as GHC 2022, Data+AI Summit, XAI 21 Summit, MLOps World, YOW Data!, appy(), QCon (SF,NY, London).

Talk: Scaling & Evolving the Machine Learning Platform at DoorDash

Abstract: As DoorDash business grows, the online ML prediction volume grows exponentially to support the various Machine Learning use cases, such as the ETA predictions, the Dasher assignments, the personalized restaurants and menu items recommendations, and the ranking of the large volume of search queries.

In this session, we will share our journey of building and scaling our Machine Learning platform and particularly the prediction service, the various optimizations experimented, lessons learned, technical decisions and tradeoffs made. We will also share how we measure success and how we set goals for the future.

What You’ll Learn: The challenges and learning lessons from building an ML platform to support ML at scale

Track: Case Study

Technical Level: 5

Location: San Jose, CA

Talk: Scaling & Evolving the Machine Learning Platform at DoorDash.

Presenter:
Hien Luu, Head of Machine Learning Platform, DoorDash

About the Speaker:
Hien Luu is a Sr. Engineering Manager at DoorDash, leading the Machine Learning Platform team. He is particularly passionate about the intersection between Big Data and Artificial Intelligence. He is the author of the Beginning Apache Spark 3 book. He has given presentations at various conferences such as GHC 2022, Data+AI Summit, XAI 21 Summit, MLOps World, YOW Data!, appy(), QCon (SF,NY, London).

Which talk track does this best fit into?
Case Study

Technical level of your talk?
(Technical level: 5/7)

Who is this presentation for?
Senior Business Executives, Product Managers, Data Scientists/ ML Engineers and High-level Researchers, Data Scientists/ ML Engineers, ML Engineers

What you’ll learn:
The journey in scaling the ML platform

On a scale of 1-10 how mature is this applied AI application you plan to discuss?
10/10

Pre-requisite Knowledge:
High level understanding of microservices

What kind of DevOps tools you plan to discuss? Open source?
CI/CD, Git, MLFlow,

What are some of the languages you plan to discuss?
Python, Kotlin

What are some of the infrastructures you plan to discuss?
Feature engineering at scale, low latency and high QPS model prediction service

What is unique about this speech, from other speeches given on the topic?
This is a case study about our journey of building ML platform at DoorDash

Abstract of Talk:
As DoorDash business grows, the online ML prediction volume grows exponentially to support the various Machine Learning use cases, such as the ETA predictions, the Dasher assignments, the personalized restaurants and menu items recommendations, and the ranking of the large volume of search queries.

Can you suggest 2-3 topics for post-discussion?
Adopting MLOps

Talk: Transforming The Retail Industry with Transformers

Presenters:
Kyryl Truskovskyi, Applied Research Scientist, Georgian & Rohit Saha, Applied Research Scientist, Georgian

About the Speakers:
Kyryl has over eight years of experience in the field of Machine Learning. For the bulk of this career, he has helped build machine learning startups, from inception to a product. He has also developed expertise in choosing and implementing state-of-the-art deep learning architectures and large-scale solutions based on them.

Rohit Saha is currently an applied research scientist at Georgian’s R&D team and is assisting portfolio companies with their research endeavours. Owing to previous roles, he has experience building end-to-end machine learning pipelines. He holds a master’s degree from the University of Toronto, and his research interests include generative modelling and transfer learning for Computer Vision tasks

Which talk track does this best fit into?
Case Study

Technical level of your talk?
(Technical level: 6/7)

What you’ll learn:
The insights and findings we share in this talk are derived from using the latest ML techniques and tools for solving real-world use cases at SPINS and are not readily available on the internet. We also open the stage for Q&A at the end of this talk to address the questions from the audience.

Abstract of Talk:
In recent years, we have seen amazing results in artificial intelligence and machine learning owing to the emergence of models such as transformers and pretrained language models. Despite the astounding results published in academic papers, there remains a lot of ambiguity and challenges when it comes to deploying these models in the industry because 1) troubleshooting, training, and maintaining these models is very time and cost consuming due to their inherent large sizes and complexities 2) there is not yet enough clarity about when the advantages and challenges of these models outweigh classical ML models. These challenges are even more severe for small and mid-sized companies that do not have access to huge compute resources and infrastructure. In this talk, we discuss these challenges and share our findings and recommendations from working on real-world examples at SPINS, a data/tech company focused on the natural grocery industry. More specifically, we describe how we leverage state-of-the-art language models to seamlessly automate parts of SPINS’ data ingestion workflow and drive substantial business outcomes. We provide a walk-through of our end-to-end MLOps system and discuss how using the right tools and methods has helped to mitigate some of these challenges. We also share our findings from our experimentation and provide insights on when one should use these massive transformer models instead of classical ML models. Considering that we have a variety of challenges in our use cases from an ill-defined label space to a huge number of classes (~86,000) and massive data imbalance, we believe our findings and recommendations can be applied to most real-world settings. We hope that the learnings from this talk can help you to solve your own problems more effectively and efficiently!

Chloe Pou-Prom

Data Scientists, Unity Health Toronto

Chloé Pou-Prom is a data scientist with the Data Science and Advanced Analytics (DSAA) team at Unity Health Toronto. The DSAA team uses high quality healthcare data in innovative ways to catalyze communities of data users and decision makers in making transformative changes that improve patient outcomes and healthcare system efficiency.

Co-Presenter: Vaakesan Sundrelingam

Workshop: NLP for Healthcare: Challenges With Processing and De-Identifying Clinical Notes

Abstract: Clinical notes (e.g., admission notes, nurse notes, radiology reports) are rich with information. In this session, we discuss the challenges of working with text data from two different perspectives. First, we provide an overview of the different issues that one can encounter when working with healthcare data, with an emphasis on data processing and cleaning. Then, we focus on the challenges that arise when it comes to sharing data across hospitals, more specifically de-identifying clinical text data. Finally, we provide a demo of pydeid, a Python-based de-identification software that identifies and replaces personal health information (PHI).

What You’ll Learn:
1) Why NLP for healthcare is challenging;
2) Why sharing clinical notes across hospitals is difficult; and
3) Some tips and tools to help out with (1) and (2)

Technical Level: 3

Location: Toronto

Workshop: NLP for Healthcare: Challenges With Processing and De-Identifying Clinical Notes

Presenters:
Chloe Pou-Prom, Data Scientists, Unity Health Toronto & Vaakesan Sundrelingam, Data Scientists, Unity Health Toronto

About the Speakers:
Chloé Pou-Prom is a data scientist with the Data Science and Advanced Analytics (DSAA) team at Unity Health Toronto. The DSAA team uses high quality healthcare data in innovative ways to catalyze communities of data users and decision makers in making transformative changes that improve patient outcomes and healthcare system efficiency.

Vaakesan Sundrelingam is a data scientist with the GEMINI team at Unity Health Toronto. GEMINI is Canada’s largest hospital data & analytics study, helping physicians, health care teams, and hospitals use data to gain insights into patient care and improve patient outcomes. GEMINI uses machine learning in creative ways to prepare large amounts of data for researchers, as well as in clinical applications such as to detect particularly difficult to measure conditions for quality of care improvement initiatives.

Technical level of your talk?
(Technical Level: 3/7)

What you’ll learn:
1) Why NLP for healthcare is challenging;
2) Why sharing clinical notes across hospitals is difficult; and
3) Some tips and tools to help out with (1) and (2)

Abstract of Talk:
Clinical notes (e.g., admission notes, nurse notes, radiology reports) are rich with information. In this session, we discuss the challenges of working with text data from two different perspectives. First, we provide an overview of the different issues that one can encounter when working with healthcare data, with an emphasis on data processing and cleaning. Then, we focus on the challenges that arise when it comes to sharing data across hospitals, more specifically de-identifying clinical text data. Finally, we provide a demo of pydeid, a Python-based de-identification software that identifies and replaces personal health information (PHI).

Stefanie Molin

Software Engineer / Data Scientist, Bloomberg

Stefanie Molin is a software engineer and data scientist at Bloomberg in New York City, where she tackles tough problems in information security, particularly those revolving around data wrangling/visualization, building tools for gathering data, and knowledge sharing. She is also the author of “Hands-On Data Analysis with Pandas,” which is currently in its second edition. She holds a bachelor’s of science degree in operations research from Columbia University’s Fu Foundation School of Engineering and Applied Science. She is currently pursuing a master’s degree in computer science, with a specialization in machine learning, from Georgia Tech. In her free time, she enjoys traveling the world, inventing new recipes, and learning new languages spoken among both people and computers.

Workshop: Beyond the Basics: Data Visualization in Python

Abstract: The human brain excels at finding patterns in visual representations, which is why data visualizations are essential to any analysis. Done right, they bridge the gap between those analyzing the data and those consuming the analysis. However, learning to create impactful, aesthetically-pleasing visualizations can often be challenging. This session will equip you with the skills to make customized visualizations for your data using Python.

While there are many plotting libraries to choose from, the prolific Matplotlib library is always a great place to start. Since various Python data science libraries utilize Matplotlib under the hood, familiarity with Matplotlib itself gives you the flexibility to fine tune the resulting visualizations (e.g., add annotations, animate, etc.). This session will also introduce interactive visualizations using HoloViz, which provides a higher-level plotting API capable of using Matplotlib and Bokeh (a Python library for generating interactive, JavaScript-powered visualizations) under the hood.

What You’ll Learn: Data visualization is essential for anyone working with data, but sometimes it can be difficult to create impactful visualizations in Python. In this workshop, we will move beyond the plotting basics and explore how to make compelling static, animated, and interactive visualizations.

Technical Level: 4

Location: New York City

Workshop: Beyond the Basics: Data Visualization in Python

Presenter:
Stefanie Moliin, Software Engineer / Data Scientist, Bloomberg

About the Speaker:
Stefanie Molin is a software engineer and data scientist at Bloomberg in New York City, where she tackles tough problems in information security, particularly those revolving around data wrangling/visualization, building tools for gathering data, and knowledge sharing. She is also the author of “Hands-On Data Analysis with Pandas,” which is currently in its second edition. She holds a bachelor’s of science degree in operations research from Columbia University’s Fu Foundation School of Engineering and Applied Science. She is currently pursuing a master’s degree in computer science, with a specialization in machine learning, from Georgia Tech. In her free time, she enjoys traveling the world, inventing new recipes, and learning new languages spoken among both people and computers.

Which talk track does this best fit into?
Workshop (1.5-4 hours)

Technical level of your talk?
(Technical level: 4/7)

Who is this presentation for?
Data Scientists/ ML Engineers, ML Engineers, Researchers

What you’ll learn:
A workshop provides the attendees opportunities to ask questions to make sure they are understanding the concepts. Attendees will also have a workshop of curated examples using real-world data rather than the dummy or randomly-generated data nearly everywhere. Each of the visualizations is also created step-by-step, viewing how it changes with each command, which gives attendees a much stronger grasp of the concepts that they can apply elsewhere.

What are the main core message (learning) you want attendees to take away from this talk?
Data visualization is essential for anyone working with data, but sometimes it can be difficult to create impactful visualizations in Python. In this workshop, we will move beyond the plotting basics and explore how to make compelling static, animated, and interactive visualizations.

Pre-requisite Knowledge:
You should have basic knowledge of Python and be comfortable working in Jupyter Notebooks. Check out this notebook for a crash course in Python or work through the official Python tutorial for a more formal introduction. The environment we will use for this workshop comes with JupyterLab, which is pretty intuitive, but be sure to familiarize yourself using notebooks in JupyterLab and additional functionality in JupyterLab. In addition, a basic understanding of pandas will be beneficial, but is not required; reviewing the first section of my pandas workshop will be sufficient.

What is unique about this speech, from other speeches given on the topic?
My teaching style is very different: since the code examples I provide are carefully chosen, it’s easy to see why would take the approach I show, so I make sure that the attendees understand exactly what each line of code is doing to make that happen. I find that this gives the attendees knowledge that they can apply to other problems, rather than just knowing that the code all together has some effect — they get a deeper understanding and can use the concepts like building blocks for their own use cases. Attendees often praise the content in the slides as a detailed reference for later as well.

Abstract of Talk:
The human brain excels at finding patterns in visual representations, which is why data visualizations are essential to any analysis. Done right, they bridge the gap between those analyzing the data and those consuming the analysis. However, learning to create impactful, aesthetically-pleasing visualizations can often be challenging. This session will equip you with the skills to make customized visualizations for your data using Python.

Can you suggest 2-3 topics for post-discussion?
Anything relating to the content covered, building data tools, or writing a book/creating workshops

Patricia Thaine

Co-Founder & CEO, Private AI

Patricia Thaine is the Co-Founder & CEO of Private AI, a Microsoft-backed startup, is also a Computer Science PhD Candidate at the University of Toronto (on leave) and a Vector Institute alumna. Her R&D work is focused on privacy-preserving natural language processing, with a focus on applied cryptography and re-identification risk. She also does research on computational methods for lost language decipherment. Patricia is a recipient of the NSERC Postgraduate Scholarship, the RBC Graduate Fellowship, the Beatrice “Trixie” Worsley Graduate Scholarship in Computer Science, and the Ontario Graduate Scholarship. She has ten years of research and software development experience, including at the McGill Language Development Lab, the University of Toronto’s Computational Linguistics Lab, the University of Toronto’s Department of Linguistics, and the Public Health Agency of Canada.

Workshop: Demystifying De-Identification

Abstract: Workshop with discussion and demo. The session will begin with an overview of privacy enhancing technologies and then dive into de-identification terminology (de-identification, anonymization, redaction, pseudonymization), how these have been misunderstood, and what to think about when choosing between one of these and other privacy enhancing technologies.
The attendees should bring a sample dataset (preferably made up of unstructured text) and a use case in mind. Each attendee will receive an API key to process a data sample and we will discuss the results. Data can be in languages other than English. Please confirm with organizer that the language is supported first.

What You’ll Learn: Attendees will learn about which privacy enhancing technologies are best for their use case and understand when de-identification is right for them and how not to misuse terminology such as “anonymization”

Technical Level: 4

Location: Toronto

Workshop: Demystifying De-Identification

Presenter:
Patricia Thaine, Co-Founder & CEO, Private AI

About the Speaker:
Patricia Thaine is the Co-Founder & CEO of Private AI, a Microsoft-backed startup, is also a Computer Science PhD Candidate at the University of Toronto (on leave) and a Vector Institute alumna. Her R&D work is focused on privacy-preserving natural language processing, with a focus on applied cryptography and re-identification risk. She also does research on computational methods for lost language decipherment. Patricia is a recipient of the NSERC Postgraduate Scholarship, the RBC Graduate Fellowship, the Beatrice “Trixie” Worsley Graduate Scholarship in Computer Science, and the Ontario Graduate Scholarship. She has ten years of research and software development experience, including at the McGill Language Development Lab, the University of Toronto’s Computational Linguistics Lab, the University of Toronto’s Department of Linguistics, and the Public Health Agency of Canada.

Technical level of your talk?
(Technical Level: 4/7)

What you’ll learn:
Attendees will learn about which privacy enhancing technologies are best for their use case and understand when de-identification is right for them and how not to misuse terminology such as “anonymization”

Abstract of Talk:
Workshop with discussion and demo. The session will begin with an overview of privacy enhancing technologies and then dive into de-identification terminology (de-identification, anonymization, redaction, pseudonymization), how these have been misunderstood, and what to think about when choosing between one of these and other privacy enhancing technologies.
The attendees should bring a sample dataset (preferably made up of unstructured text) and a use case in mind. Each attendee will receive an API key to process a data sample and we will discuss the results. Data can be in languages other than English. Please confirm with organizer that the language is supported first.

Denys Linkov

ML Lead, Voiceflow

Started the ML team at Voiceflow, Kickstarted Rbc’s MLOps journey, Youngest Senior Architect at Rbc. Lead discussion groups and mentorship on MLOps and various blog posts

Workshop: Iterating on NLP Models from R&D to Production

Abstract: Research papers, blogs and products are the culmination of many hours of work, iteration and frustration. However, in these final polished formats, we often gloss over the iteration or creative process on how to get to our desired results.

In this talk, I’ll cover a series of short labs that mirror some of the challenges we’ve faced in building out our NLP models and algorithms. It will be an interactive session with a series of collaborative problem solving, and explanations of what we built and the process we took along the way.

Some of the twists and turns will include:
– Integrating a BERT based model with a message queue system
– Speeding up semantic search through vectorization
– Enabling multi lingual recommendations

Each member of the talk will have access to the code examples and will be encouraged to think beyond the challenges addressed and how they can apply some of our lessons learned to their own work.

What You’ll Learn:
– How to go from idea to product
– How to iterate on a product
– How to go to production
– How to incorporate customer feedback

Technical Level: 6

Location: Toronto

Workshop: Iterating on NLP models from R&D to Production

Presenter:
Denys Linkov, ML Lead, Voiceflow

About the Speaker:
Started the ML team at Voiceflow, Kickstarted Rbc’s MLOps journey, Youngest Senior Architect at Rbc. Lead discussion groups and mentorship on MLOps and various blog posts

Technical level of your talk?
(Technical level: 6/7)

Are there any industries (in particular) that are relevant for this talk?
Computer Software, Information Technology & Service, Any startup / large company looking at the R&D process

Who is this presentation for?
Data Scientists/ ML Engineers, ML Engineers, Researchers

What you’ll learn:
Specific examples and challenges of building NLP products

What are the main core message (learning) you want attendees to take away from this talk?
– How to go from idea to product
– How to iterate on a product
– How to go to production
– How to incorporate customer feedback

What is unique about this speech, from other speeches given on the topic?
Nothing super unique, but many startups of this size rarely share their experience building products and iterating. Many tutorials cover basics and not real business problems.

Abstract of Talk:
Research papers, blogs and products are the culmination of many hours of work, iteration and frustration. However, in these final polished formats, we often gloss over the iteration or creative process on how to get to our desired results.

Each member of the talk will have access to the code examples and will be encouraged to think beyond the challenges addressed and how they can apply some of our lessons learned to their own work.

Can you suggest 2-3 topics for post-discussion?
BERT based models and embeddings
Deploying models into production
NLP product development

Dr. Nasim Abdollahi

Postdoctoral Fellow, University of Toronto / Machine Learning Researcher, Cyclica

Nasim is a Postdoctoral Fellow at University of Toronto and a Machine Learning Researcher Intern at Cyclica, leading a collaborative project between Cyclica, University of Toronto and Vector Institute. She is the vice-chair of Engineering in Medicine and Biology Society of IEEE Toronto section. Nasim obtained her Ph.D. in electrical and computer engineering from University of Manitoba and has M.Sc. and B.Sc. in biomedical engineering. With her passion for developing and applying novel machine learning techniques for improving the quality of health care, she has conducted numerous research projects on enhancing biomedical imaging for breast cancer detection and monitoring. Her current research is focused on graph-based machine learning models that can predict proteins’ biological functions from their 3D atomic structures, with a promise to enhance designing novel medicines. Nasim is an advocate for women in STEM, serves as vice-chair of IEEE Canada Women in Engineering, and was recognized as a “Visionary Emerging Leader”.

Co-Presenter: Dr. Farnoosh Khodakarami

Workshop: Graph Neural Network Modeling in Drug Discovery Using PyTorch

Abstract: Graph Neural Networks (GNNs) have been among the most popular neural network architectures, and as graph is a natural representation for protein and molecule, GNNs have shown big sparks in graph-based ML modeling for drug discovery and protein science. Graph-based ML models can help us in identifying the topology of a protein structure from protein sequence, predicting protein’s biological functions from protein structure as well as identifying protein-protein and protein-drug interactions. In this workshop, we will have an introduction on Graph Neural Network (GNN) and its application in drug discovery followed by a code session on PyTorch Geometric, which is a great PyTorch library for building GNN models for structured data. We will then have a code-base session to walk you through two useful tools built with PyTorch Geometric: TorchDrug and NodeCoder.

What You’ll Learn: Audience will learn about:
– Graph Neural Network (GNN) in drug discovery
– How to build GNN with PyTorch Geometric
– TorchDrug – ML platform for drug discovery
– TorchProtein – a ML library for protein science
– NodeCoder – a graph-based ML framework for predicting proteins’ biological functions

Technical Level: 7

Location: Toronto

Workshop: Graph Neural Network Modeling in Drug Discovery Using PyTorch

Presenters:
Dr. Nasim Abdollahi, Postdoctoral Fellow at University of Toronto, Machine Learning Researcher at Cyclica & Dr. Farnoosh Khodakarami Computer Scientist & ML Researcher, Cyclica

About the Speaker:
Nasim is a Postdoctoral Fellow at University of Toronto and a Machine Learning Researcher Intern at Cyclica, leading a collaborative project between Cyclica, University of Toronto and Vector Institute. She is the vice-chair of Engineering in Medicine and Biology Society of IEEE Toronto section. Nasim obtained her Ph.D. in electrical and computer engineering from University of Manitoba and has M.Sc. and B.Sc. in biomedical engineering. With her passion for developing and applying novel machine learning techniques for improving the quality of health care, she has conducted numerous research projects on enhancing biomedical imaging for breast cancer detection and monitoring. Her current research is focused on graph-based machine learning models that can predict proteins’ biological functions from their 3D atomic structures, with a promise to enhance designing novel medicines. Nasim is an advocate for women in STEM, serves as vice-chair of IEEE Canada Women in Engineering, and was recognized as a “Visionary Emerging Leader”.

Farnoosh Khodakarami is an experienced computer scientist with a demonstrated history of working in the research industry. Skilled in application development with experience in machine learning applications. Strong research professional with a Doctor of Philosophy (Ph.D.) focused in Computer Science. Creative, self-motivated, and committed to working with a team-player attitude, great problem-solving skills, and the ability to quickly grasp new concepts.

Which talk track does this best fit into?
Workshop

Technical level of your talk?
(Technical level: 7/7)

Are there any industries (in particular) that are relevant for this talk?
Hospital & Health Care

What are the main core message (learning) you want attendees to take away from this talk?
Audience will learn about:
– Graph Neural Network (GNN) in drug discovery
– How to build GNN with PyTorch Geometric
– TorchDrug – ML platform for drug discovery
– TorchProtein – a ML library for protein science
– NodeCoder – a graph-based ML framework for predicting proteins’ biological functions

Abstract of Talk:
Graph Neural Networks (GNNs) have been among the most popular neural network architectures, and as graph is a natural representation for protein and molecule, GNNs have shown big sparks in graph-based ML modeling for drug discovery and protein science. Graph-based ML models can help us in identifying the topology of a protein structure from protein sequence, predicting protein’s biological functions from protein structure as well as identifying protein-protein and protein-drug interactions. In this workshop, we will have an introduction on Graph Neural Network (GNN) and its application in drug discovery followed by a code session on PyTorch Geometric, which is a great PyTorch library for building GNN models for structured data. We will then have a code-base session to walk you through two useful tools built with PyTorch Geometric: TorchDrug and NodeCoder.

Arthur Vitui

Senior Data Scientist Specialist Solution Architect, RedHat Canada

Arthur is a senior data scientist specialist solution architect at RedHat Canada where with the help of open source software is helping organizations develop intelligent application ecosystems and bring them into production using MLOps best practices.
He is also pursuing his Ph.D. degree in Computer Science at Concordia University, Montreal, Canada, and he is a research assistant in the Software Perfomance Analysis and Reliability (SPEAR) Lab.
His research interests are related to AIOps with focus on performance and scalability optimization.

Workshop: Open Source Intelligent Application Delivery on Kubernetes

Abstract: The recent rise in popularity of containerized workloads demanded better ways to orchestrate and manage these workloads hence the creation of the Kubernetes platform.

When it comes to running intelligent application workloads which contain built-in AI/ML software components, the requirement of a Kubernetes platform as a service extends beyond agility, portability, flexibility and scalability as it is required to also answer to the datascientist’s dilemma: getting started and getting into production.

However, as the ML code is only a small part of the entire intelligent application ecosystem, with this workshop we present a showcase for using a Kubernetes platform and a blueprint architecture that proposes an answer to many challenges related to the development, deployment and management of distributed applications.
The user stories we shall focus on in this workshop concerning the developer, data scientist and operations engineer personas are:
– As a data scientist, I want to develop ML models using Jupyter Hub (lab/notebooks) as my preferred research environment.
– As a data scientist, I want my model to be deployed quickly so that it may be used by other applications.
– As a (fullstack) developer, I want to have quick access to resources that support the business logic of my applications, including databases, storage, messaging.
– As a (fullstack) developer, I want an automated build process to support new releases/code updates as soon as they are available in a git repository.
– As an operations engineer, I want an integrated monitoring dashboard to new applications available on the (production) infrastructure.

What You’ll Learn: Open source container platforms are a great option to integrate Machine Learning with any application or service by boosting productivity while maintaining a high level of security.

Technical Level: 4

Location: Bossard

Workshop: Open Source Intelligent Application Delivery on Kubernetes

Presenter:
Arthur Vitui, Senior Data, Scientist Specialist Solution Architect, RedHat Canada

About the Speaker
Arthur is a senior data scientist specialist solution architect at RedHat Canada where with the help of open source software is helping organizations develop intelligent application ecosystems and bring them into production using MLOps best practices.
He is also pursuing his Ph.D. degree in Computer Science at Concordia University, Montreal, Canada, and he is a research assistant in the Software Perfomance Analysis and Reliability (SPEAR) Lab.
His research interests are related to AIOps with focus on performance and scalability optimization.

Which talk track does this best fit into?
Workshop

Technical level of your talk?
(Technical level: 4/7)

Are there any industries (in particular) that are relevant for this talk?
Computer Software, Hospital & Health Care, Information Technology & Service, Insurance

Who is this presentation for?
Senior Business Executives, Product Managers, Data Scientists/ ML Engineers and High-level Researchers, Product Managers, Data Scientists/ ML Engineers, ML Engineers, Researchers

What you’ll learn:
The audience will learn about what an intelligent application is, how to orchestrate its design, deployment and monitoring in a kubernetes environment. The audience will also learn about the datascientist dillema and how it may be addressed.

What are the main core message (learning) you want attendees to take away from this talk?
Open source container platforms are a great option to integrate Machine Learning with any application or service by boosting productivity while maintaining a high level of security.

Pre-requisite Knowledge:
Generic SDLC and basic Kubernetes knowledge

What is unique about this speech, from other speeches given on the topic?
Bringing in an enterprise perspective and enterprise ready Kubernetes platform beyond just a proof of concept (POC) although presenting a POC showcase for an end to end intelligent application.

Abstract of Talk:
The recent rise in popularity of containerized workloads demanded better ways to orchestrate and manage these workloads hence the creation of the Kubernetes platform.

Can you suggest 2-3 topics for post-discussion?
– DataScientist Kubernetes Platform as a Service
– Automating builds and exposure of ML models inference endpoints

Jörg Schad

CTO, ArangoDB

Jörg Schad is CTO at ArangoDB and enjoys working in intersections of (Graph) databases, Cloud Architectures, and Machine Learning. In a previous life, he has worked on or built machine learning pipelines in healthcare, distributed systems at Mesosphere, and in-memory databases. He received his Ph.D. for research around distributed databases and data analytics. He’s a frequent speaker at meetups, international conferences, and lecture halls.

Workshop: Graph ML – The Next Level of Machine Learning

Abstract: This workshop focuses on why Graphs have become one of the biggest trends in Machine Learning. Graph Machine Learning based on Graph Analytic Algorithms is driving significant improvements in Fraud/Anomaly Detection, Ranking (Page Rank), Recommendation Engines (collaborative filtering), text summarization, and other NLP tasks. We will cover Graph Analytic Algorithms, their applications, and the more novel–but equally exciting–field of Graph Machine Learning, including topics such as Graph Neural Networks, Graph Embeddings, and applications of Graph Machine Learning.

The workshop will be hands-on based on Jupyter notebooks and cover sessions
– Why Graph and Graph Thinking
– Graph Algorithms
– Graph Embeddings
– Graph Neural Networks

What You’ll Learn: Graph Machine is considering relations (and neighborhood context) as first class citizens and hence can lead to more powerful and simplified Machine learning models.

Technical Level: 6

Location: Berlin/San Francisco

Workshop: Graph ML – The Next Level of Machine Learning

Presenter:
Jörg Schad, CTO, ArangoDB

About the Speaker:
Jörg Schad is CTO at ArangoDB and enjoys working in intersections of (Graph) databases, Cloud Architectures, and Machine Learning. In a previous life, he has worked on or built machine learning pipelines in healthcare, distributed systems at Mesosphere, and in-memory databases. He received his Ph.D. for research around distributed databases and data analytics. He’s a frequent speaker at meetups, international conferences, and lecture halls.

Which talk track does this best fit into?
Workshop

Technical level of your talk?
(Technical level: 5/7)

Are there any industries (in particular) that are relevant for this talk?
All industries

What are the main core message (learning) you want attendees to take away from this talk?
Machine learning is much more than just building models and the overall pipeline should be considered early on in order to result in actual business impact. Luckily there exist a number of Open-Source projects to help…

Abstract of Talk:
This workshop focuses on why Graphs have become one of the biggest trends in Machine Learning. Graph Machine Learning based on Graph Analytic Algorithms is driving significant improvements in Fraud/Anomaly Detection, Ranking (Page Rank), Recommendation Engines (collaborative filtering), text summarization, and other NLP tasks. We will cover Graph Analytic Algorithms, their applications, and the more novel–but equally exciting–field of Graph Machine Learning, including topics such as Graph Neural Networks, Graph Embeddings, and applications of Graph Machine Learning.

The workshop will be hands-on based on Jupyter notebooks and cover sessions
– Why Graph and Graph Thinking
– Graph Algorithms
– Graph Embeddings
– Graph Neural Networks

Presenter:
Jörg Schad, CTO, ArangoDB

Which talk track does this best fit into?
Workshop

Technical level of your talk?
(Technical level: 5/7)

Are there any industries (in particular) that are relevant for this talk?
All industries

Abstract of Talk:
Many Machine Learning fail to turn an initial idea and potentially even first model into the business impact, as they neglect the importance (and associated work) of building a production-grade ML pipeline. There are many great tutorials for training your deep learning models using PyTorch, TensorFlow, Keras, Spark or one of the many other frameworks. But training is only a small part in the overall deep learning pipeline. This workshop gives an overview into building a complete automated deep learning pipeline starting with exploratory analysis, over training, model storage, model serving, meta-data storage, and monitoring using available Open-Source tool.

The participants will build an end-to-end data analytics pipeline including:
– Pipeline Orchestration
– Data preparation using Apache Spark
– Jupyter Notebooks
– Distributed training with TensorFlow
– Automation & CI/CD using Jenkins and Argo
– Model and metadata storage
– Model serving and monitoring

Mahmudul Hasan

Lead Data Scientist, TELUS Business Marketing

With deep expertise in Machine Learning and AI, Mahmudul has over 10 years industry experience of building enterprise level data products to achieve digital transformation, improve customer experience, new revenue opportunity, and cost savings for companies across the globe. He is currently serving as a Lead Data Scientist in TELUS Business Marketing. Mahmudul also designed and developed NLP course content for University of Toronto School of Continuing Studies and also serving as an instructor for the same.
Mahmudul holds a Master’s degree in Management Science from University of Waterloo and a Bachelor’s in Computer Science & Engineering.

Workshop: Introduction to NLP & a Step by Step Implementation of a Real World Use Case from TELUS

Abstract: The workshop will be delivered in two part:
– Part-1: Brief introduction to NLP concepts and ideas which would include
– Basic definitions and use cases
– Why NLP is a different ball game inside AI/ML (major challenges of processing natural language etc.)
– How those challenges are overcame with ML based approach
– Major workflow of building NLP application.
– Part-2: is a detail implementation of a case study with coding details which I have implemented in TELUS. During this part-2, audience will see how a business problem is solved leveraging unstructured text data using NLP algorithms along with necessary tips and tricks which makes a unsupervised learning based project financially successful for the company.

What You’ll Learn: Audience will see how a business problem is solved leveraging unstructured text data using NLP algorithms along with necessary tips and tricks which makes a unsupervised learning based project financially beneficial for the business.

Technical Level: 6

Location: Toronto

Workshop: Introduction to NLP & a Step by Step Implementation of a Real World Use Case from TELUS

Presenter:
Mahmudul Hasan, Lead Data Scientist, TELUS Business Marketing

About the Speaker:
With deep expertise in Machine Learning and AI, Mahmudul has over 10 years industry experience of building enterprise level data products to achieve digital transformation, improve customer experience, new revenue opportunity, and cost savings for companies across the globe. He is currently serving as a Lead Data Scientist in TELUS Business Marketing. Mahmudul also designed and developed NLP course content for University of Toronto School of Continuing Studies and also serving as an instructor for the same.
Mahmudul holds a Master’s degree in Management Science from University of Waterloo and a Bachelor’s in Computer Science & Engineering.

Which talk track does this best fit into?
Workshop

Technical level of your talk?
(Technical level: 6/7)

Are there any industries (in particular) that are relevant for this talk?
Computer Software, Marketing & Advertising, Telecommunications

Who is this presentation for?
Senior Business Executives, Product Managers, Data Scientists/ ML Engineers and High-level Researchers, Data Scientists/ ML Engineers

What you’ll learn:
The audience will have a real world case study of how unsupervised NLP algorithm can be successfully create values for a business, and some tips and tricks which make this kind of project successful for a data scientist

What are the main core message (learning) you want attendees to take away from this talk?
Audience will see how a business problem is solved leveraging unstructured text data using NLP algorithms along with necessary tips and tricks which makes a unsupervised learning based project financially beneficial for the business.

Pre-requisite Knowledge:
Some basic understanding on Data Science

What is unique about this speech, from other speeches given on the topic?
The audience will get an idea of how unstructured data can be converted to generate financially impactful benefits for business. Also will share some tips on how to make this kind of unsupervised learning based project a successful for a big corporation like TELUS.

Abstract of Talk:
The workshop will be delivered in two part:
Part-1: Brief introduction to NLP concepts and ideas which would include
– Basic definitions and use cases
– Why NLP is a different ball game inside AI/ML (major challenges of processing natural language etc.)
– How those challenges are overcame with ML based approach
– Major workflow of building NLP application.
Part-2: is a detail implementation of a case study with coding details which I have implemented in TELUS. During this part-2, audience will see how a business problem is solved leveraging unstructured text data using NLP algorithms along with necessary tips and tricks which makes a unsupervised learning based project financially successful for the company.

Can you suggest 2-3 topics for post-discussion?
1. What are the challenges of implementing a data science project in business?
2. how can you make your AIML project impactful for the business?

Bhaskarjit Sarmah

Senior Data Scientist, BlackRock

Bhaskarjit is a data scientist and has solved business problems in many domains including Retail, FMCG, Banking, Media & Entertainment etc. using machine learning. Currently he is working as a data scientist BlackRock where he builds predictive models for financial markets. His research interests are Network Science, AI Interpretability, Uncertainty, NLP etc.

Workshop: Learning Embedded Representation of the Stock Correlation Matrix using Graph Machine Learning

Abstract: Understanding non-linear relationships among financial instruments has various applications in investment processes ranging from risk management, portfolio construction and trading strategies. Here, we focus on interconnectedness among stocks based on their correlation matrix which we represent as a network with the nodes representing individual stocks and the weighted links between pairs of nodes representing the corresponding pair-wise correlation coefficients. The traditional network science techniques, which are extensively utilized in financial literature, require handcrafted features such as centrality measures to understand such correlation networks.

However, manually enlisting all such handcrafted features may quickly turn out to be a daunting task. Instead, we propose a new approach for studying nuances and relationships within the correlation network in an algorithmic way using a graph machine learning algorithm called Node2Vec.

In particular, the algorithm compresses the network into a lower dimensional continuous space, called an embedding, where pairs of nodes that are identified as similar by the algorithm are placed closer to each other. By using log returns of S&P 500 stock data, we show that our proposed algorithm can learn such an embedding from its correlation network. We define various domain specific quantitative (and objective) and qualitative metrics that are inspired by metrics used in the field of Natural Language Processing (NLP) to evaluate the embeddings in order to identify the optimal one. Further, we discuss various applications of the embeddings in investment management.

What You’ll Learn: In this paper we have shown how to create stock embedding representation from stock correlation matrix. And evaluated the learnt embeddings using a quantitative way

Pre-requiste Knowledge: Network Science, Machine Learning, Word Embeddings

Technical Level: 5

Location: Delhi

Workshop: Learning Embedded Representation of the Stock Correlation Matrix using Graph Machine Learning

Presenter:
Bhaskarjit Sarmah, Senior Data Scientist, BlackRock

About the Speaker:
Bhaskarjit is a data scientist and has solved business problems in many domains including Retail, FMCG, Banking, Media & Entertainment etc. using machine learning. Currently he is working as a data scientist BlackRock where he builds predictive models for financial markets. His research interests are Network Science, AI Interpretability, Uncertainty, NLP etc.

Which talk track does this best fit into?
Research: Advanced Technical.

Technical level of your talk?
(Technical level: 4 /7)

Are there any industries (in particular) that are relevant for this talk?
Banking & Financial Services, Information Technology & Service, Insurance, Marketing & Advertising

What you’ll learn:
In this paper we have shown how to create stock embedding representation from stock correlation matrix. And evaluated the learnt embeddings using a quantitative way.

What are the main core message (learning) you want attendees to take away from this talk?
How to represent financial securities in form of embeddings using graph machine learning

Pre-requisite Knowledge:
Network Science, Machine Learning, Word Embeddings

What is unique about this speech, from other speeches given on the topic?
This speech is centered around feature extraction from networks. In this speech, will first introduce the traditional hand crafted feature extraction technique from networks. And then will explain how we can use graph machine learning for automatic feature extraction in the form embeddings. And how to evaluate those embeddings in quantitative way.

Abstract of Talk:
Understanding non-linear relationships among financial instruments has various applications in investment processes ranging from risk management, portfolio construction and trading strategies. Here, we focus on interconnectedness among stocks based on their correlation matrix which we represent as a network with the nodes representing individual stocks and the weighted links between pairs of nodes representing the corresponding pair-wise correlation coefficients. The traditional network science techniques, which are extensively utilized in financial literature, require handcrafted features such as centrality measures to understand such correlation networks. However, manually enlisting all such handcrafted features may quickly turn out to be a daunting task. Instead, we propose a new approach for studying nuances and relationships within the correlation network in an algorithmic way using a graph machine learning algorithm called Node2Vec. In particular, the algorithm compresses the network into a lower dimensional continuous space, called an embedding, where pairs of nodes that are identified as similar by the algorithm are placed closer to each other. By using log returns of S&P 500 stock data, we show that our proposed algorithm can learn such an embedding from its correlation network. We define various domain specific quantitative (and objective) and qualitative metrics that are inspired by metrics used in the field of Natural Language Processing (NLP) to evaluate the embeddings

Can you suggest 2-3 topics for post-discussion?
Node2Vec, Stock Embeddings, Network Science

Workshop: Building a Fraud Detection Model with Feature Stores (Includes Bonus Case Study: How Shopify uses Feast to Manage its ML Features)

Presenters:
Danny Chiao, Tech Lead, Feast & Eddie Esquivel, Sr. Solutions Architect, Tecton & Abhin Chhabra, ML Platform Tech Lead, Shopify

About the Speakers:
Danny Chiao is an engineering lead at Tecton/Feast Inc working on building a next-generation feature store. Previously, Danny was a technical lead at Google working on end to end machine learning problems within Google Workspace, helping build privacy-aware ML platforms / data pipelines and working with research and product teams to deliver large-scale ML powered enterprise functionality. Danny holds a Bachelor’s degree in Computer Science from MIT. |

Eddie Esquivel is a Solutions Architect at Tecton, where he helps customers implement feature stores as part of their stack for Operational ML. Prior to Tecton, Eddie was a Solutions Architect at AWS.

Abhin leads the feature store team for Shopify’s ML Platform.

Which talk track does this best fit into?
Workshop

Technical level of your talk?
(Technical level: 4/7)

Who is this presentation for?
Product Managers, Data Scientists/ ML Engineers, ML Engineers

What you’ll learn:
You will learn how to:
– Build new features
– Automate the transformation of batch data
– Automate the transformation of streaming and real-time data
– Create training datasets
– Serve data online using DynamoDB or Redis
– Build fraud detection system using Tecton and Feast

Pre-requisite Knowledge:
Attendees should have functional knowledge of Python, SQL and Spark, as well as familiarity with the challenges of data engineering for ML.

What is unique about this speech, from other speeches given on the topic?
Danny and Eddie are core members of the Feast and Tecton Engineering and Solutions Architect teams. They have deep expertise in working with dozens of end-users to build real-time recommendation systems using feature stores. They also have a lot of experience working on ML infrastructure at Google, AWS, and Tecton.

Abstract of Talk:
In this workshop, we’ll show how to build a real-time fraud detection system using some of the latest tooling for managing ML data pipelines. We’ll walk through the process of building, deploying, and serving real-time data pipelines, highlighting the differences between a traditional feature store (using Feast, the open source feature store) and a feature platform (using Tecton).

We’ll present common architectural patterns and walk you through building a model in three stages:
– Batch, daily computed predictions
– Online predictions using batch features
– Online predictions using real-time features

Can you suggest 2-3 topics for post-discussion?
– Best practices for ML recommendation systems
– Building streaming and real-time data pipelines for ML
– Feature Stores: have you implemented one? Let’s share learnings

Annie En-Shiun Lee

Assistant Professor, University of Toronto

Annie En-Shiun Lee is an Assistant Professor (Teaching Stream) for the Computer Science Department at the University of Toronto. She received her PhD from the University of Waterloo in 2014 under the supervision of Professor Andrew K. C. Wong and Daniel Stashuk from the Centre of Pattern Intelligence and Machine Intelligence. She has also been a visiting researcher at the Fields Institute (invited by Nancy Reid) and CUHK (invited by K. S. Leung and M. H. Wong) as well as a research scientist at VerticalScope and Stradigi AI.

Workshop: Pre-Trained Multilingual Sequence-to-Sequence Models for NMT: Tips, Tricks and Challenges

Abstract: Neural Machine Translation (NMT) has seen a tremendous spurt of growth in less than ten years, and has already entered a mature phase. Pre-trained multilingual sequence-to-sequence (PMSS) models, such as mBART and mT5, are pre-trained on large general data, then fine-tuned to deliver impressive results for natural language inference, question answering, text simplification and neural machine translation. This tutorial presents 1) An Introduction to Sequence-to-Sequence Pre-trained Models, 2) How to adapt pre-trained models for NMT, 3) Tips and Tricks for NMT training and evaluation, 4) Challenges/Problems faced when using these models. This tutorial will be useful for those interested in NMT, from a research as well as industry point of view.

What You’ll Learn: This tutorial will give an overview of Pre-trained Sequence-to-Sequence Multilingual Models, tips, tricks and frameworks that can be used to adapt these models for NMT especially for low resource languages and the challenges faced while using these models and how to overcome them.

Technical Level: 5

Location: Toronto

Workshop: Pre-Trained Multilingual Sequence-to-Sequence Models for NMT: Tips, Tricks and Challenges

Presenter:
Annie En-Shiun Lee, Assistant Professor, University of Toronto

About the Speaker:
Annie En-Shiun Lee is an Assistant Professor (Teaching Stream) for the Computer Science Department at the University of Toronto. She received her PhD from the University of Waterloo in 2014 under the supervision of Professor Andrew K. C. Wong and Daniel Stashuk from the Centre of Pattern Intelligence and Machine Intelligence. She has also been a visiting researcher at the Fields Institute (invited by Nancy Reid) and CUHK (invited by K. S. Leung and M. H. Wong) as well as a research scientist at VerticalScope and Stradigi AI.

Which talk track does this best fit into?
Workshop

Technical level of your talk?
(Technical Level: 5/7)

What you’ll learn:
This tutorial will give an overview of Pre-trained Sequence-to-Sequence Multilingual Models, tips, tricks and frameworks that can be used to adapt these models for NMT especially for low resource languages and the challenges faced while using these models and how to overcome them.

Abstract of Talk:
Neural Machine Translation (NMT) has seen a tremendous spurt of growth in less than ten years, and has already entered a mature phase. Pre-trained multilingual sequence-to-sequence (PMSS) models, such as mBART and mT5, are pre-trained on large general data, then fine-tuned to deliver impressive results for natural language inference, question answering, text simplification and neural machine translation. This tutorial presents 1) An Introduction to Sequence-to-Sequence Pre-trained Models, 2) How to adapt pre-trained models for NMT, 3) Tips and Tricks for NMT training and evaluation, 4) Challenges/Problems faced when using these models. This tutorial will be useful for those interested in NMT, from a research as well as industry point of view.

Workshop: An Introduction to Drift Detection

Presenters:
Ed Shee, Head of Developer Relations, Seldon & Ashley Scillitoe, Data Science Research Engineer, Seldon

About the Speakers:
With a background in cloud computing and a passion for machine learning, Ed has combined those skills and now works in the MLOps field where he heads up Developer Relations at Seldon. Organizer of Tech Ethics London and MLOps London, Ed is heavily involved in lots of developer communities and, thankfully, loves both beer and pizza.

Ashley is a data science research engineer at Seldon, where he works on developing production-ready tools for drift, adversarial and outlier detection. Prior to joining Seldon, he spent a number of years as a Research Fellow at The Alan Turing Institute. Here, he explored the use of machine learning for tackling aerospace engineering problems, with a focus on explainability and uncertainty quantification. Ashley also completed a PhD at the University of Cambridge, and is a keen proponent of open-source software.

Which talk track does this best fit into?
Workshop

What you’ll learn:
What drift detection is, why it’s important and how to get started.

Pre-requisite Knowledge:
No prior knowledge or understanding of drift detection is required (we’ll be covering that) but a basic knowledge of machine learning and some experience with Python will be helpful.

Abstract of Talk:
Although powerful, modern machine learning models can be sensitive. Seemingly subtle changes in a data distribution can destroy the performance of otherwise state-of-the art models, which can be especially problematic when ML models are deployed in production. In this workshop, we will give a hands-on overview to drift detection, the discipline focused on detecting such changes. We will start by building an understanding of the ways in which drift can occur, and why it pays to detect it. We’ll then explore the anatomy of a drift detector, and learn how they can be used to detect drift in a principled manner.

You will work through a real-world example using Alibi Detect, an open-source Python library offering powerful algorithms for adversarial, outlier and drift detection.You’ll learn how to set-up drift detectors, and deduce what type of drift is occurring. Since data can take many forms, such as image, text or tabular data, you’ll explore how to use existing ML models to pre-process your data into a form suitable for drift detectors. Then, to gain further insights into the causes of drift, you’ll employ state-of-the art detectors which are able to perform fine-grained attribution to instances and features. To assess whether model performance has been affected by drift, you’ll experiment with using model uncertainty based detectors. Finally, you’ll use a novel context-aware drift detector. This takes in context (or conditioning) variables, allowing you to test for drift conditional on context that is permitted to change. We’ll discuss how this functionality can be crucial in many real-world drift detection scenarios.

This hands-on workshop is targeted at a beginner-intermediate level. No prior knowledge or understanding of drift detection is required (we’ll be covering that) but a basic knowledge of machine learning and some experience with Python will be helpful.

The workshop will be hands-on based on Jupyter notebooks and cover sessions
– Why Graph and Graph Thinking
– Graph Algorithms
– Graph Embeddings
– Graph Neural Networks

Workshop: Cancer Image Segmentation

Presenters:
Moderator (Roxana Sultan, Chief Data Officer and VP, Health, Vector Institute)
Dr. Benjamin Haibe-Kains, Senior Scientist, University Health Network
Team Fight Tumour (Jun Ma, Postdoctoral Fellow, Vector Institute / Ronald Xie, PhD Candidate, Vector Institute / Rex Ma, PhD Candidate, Vector Institute)

About the Speakers:
Roxana Sultan: Roxana Sultan is the Chief Data Officer and Vice President, Health at the Vector Institute. She leads Vector’s data strategy and its contributions to Ontario’s and Canada’s health sector. Along with our health team and partners, Roxana drives applications of AI to life sciences, fostering research, health sector and industrial sponsor projects, and initiatives to advance the health space, contributing to short-, medium-, and long-term impact achievements within the Ontario health ecosystem.
Roxana is the former Executive Director of the Provincial Council for Maternal and Child Health, where she led the implementation of evidence-based clinical quality improvement and access initiatives in obstetric, neonatal, and pediatric health services across Ontario. Her career includes leadership roles with The Hincks-Dellcrest Centre (now “SickKids Centre for Community Mental Health”), the Princess Margaret Cancer Centre in the University Health Network, the Canadian Institutes of Health Research (CIHR), Cancer Care Ontario, and the Hospital for Sick Children.
As an Adjunct Lecturer with the Institute of Health Policy, Management, and Evaluation (IHPME) at the University of Toronto (U of T), Roxana teaches a graduate-level course on intelligent medicine, machine learning, and knowledge representation. She also serves as the Vice Chair of the Board of the Canadian Cancer Society – Ontario Division.
Roxana completed her graduate education with the Department of Molecular and Medical Genetics at U of T, and holds a Masters of Health Science from IHPME.

Dr. Benjamin Haibe-Kains: Dr. Benjamin Haibe-Kains is a Senior Scientist at the Princess Margaret Cancer Centre: University Health Network, Associate Professor in the Medical Biophysics department of the University of Toronto, and Faculty Affiliate at the Vector Institute. Dr. Haibe-Kains earned his PhD in Bioinformatics at the Université Libre de Bruxelles (Belgium). Supported by a Fulbright Award, he did his postdoctoral fellowship at the Dana-Farber Cancer Institute and Harvard School of Public Health (USA). Dr. Haibe-Kains’ research focuses on the integration of high-throughput data from various sources to simultaneously analyze multiple facets of carcinogenesis. Dr. Haibe-Kains’ team is analyzing large-scale radiological and (pharmaco)genomic datasets to develop new prognostic and predictive models to improve cancer care.

Jun Ma is a Postdoctoral Fellow in the Department of Laboratory Medicine & Pathobiology at the University of Toronto. His research interests focus on the interdisciplinary areas of deep learning and medical image analysis, aiming to develop accurate, fast, and generalizable algorithms to improve healthcare. He has published seven first-author papers on top journals, such as TPAMI, TMI, and MedIA. He is the lead organizer of the MICCAI 2021-2022 FLARE Challenge.

Ronald Xie received his BSc in Microbiology and Immunology at the University of British Columbia in 2018. He then received his MPhil in Computational Biology at the Department of Applied Mathematics and Theoretical Physics at the University of Cambridge in 2019. Ronald is currently a PhD candidate in Computational Biology and Molecular Genetics (CBMG) at the Faculty of Medicine at University of Toronto. His research interests lie in deep learning applications in electron microscopy and single cell omics.

Rex Ma is currently a Computer Science Ph.D. student at the University of Toronto. He is interested in AI in healthcare and computation biology in general, with research focusing on multi-omics integration using machine learning.

Technical level of your talk?
(Technical Level: 5/7)

Pre-requisite Knowledge:
Basic ML

Abstract of Talk:
Radiation therapy planning for head and neck cancer is a time-consuming and complex task for radiologists. AI-based tools have tremendous potential for segmenting regions of interest and optimizing therapy planning. The Vector Institute and the Cancer Digital Intelligence Program (CDI) from the Princess Margaret Cancer Centre launched a Machine Learning Challenge in June 2022 focused on cancer image segmentation.

Building on foundational work from the lab of scientist, Dr. Benjamin Haibe-Kains, ten teams from Vector and UHN participated in the Challenge leveraging RADCURE, the largest head-and-neck cancer treatment dataset of its kind, containing the imaging, treatment, demographic and clinical data of 2745 head and neck cancer patients.

In this presentation, moderated by Vector’s Roxana Sultan, Dr Haibe Kains will provide an overview of his preliminary work and the winning Challenge team, Fight Tumour, will describe their winning submission.

Workshop: Observability is Critical to MLOps

Presenter:
Marcelo Litovsky Director of Sales Engineering, Aporia

About the Speaker:
Marcelo Litovsky is an experienced Information Technology professional with 30 years of diverse background in Enterprise Architecture, AI, Systems and Database Management, and Programming. He has worked in multiple industries: Financial Services, Entertainment, and Information Technology in his career. Today, he serves as Director of Sales Engineering at Aporia, bringing his expertise to help Data Scientists, Machine Learning Engineers, and Business Users work together to unlock and promote the business value of their machine learning models. You can find him at the gym, preparing healthy vegan meals when he is not talking to customers or writing Python code.

Which talk track does this best fit into?
Workshop

Technical level of your talk?
(Technical level: 4/7)

What you’ll learn:
This session will explore the steps you can take to prepare for ML observability. We will also discuss how observability helps data scientists and MLOps practitioners showcase the business value of the applications they deploy and get recognition for their hard work.

Abstract of Talk:
MLOps is bringing a lot of attention to the business impact of Machine Learning. It also introduces new challenges that cannot be efficiently addressed with DevOps. What are these challenges, and what makes MLOps so different from DevOps? They both deal with the life cycle of an application, so what is the difference? Most software applications have a pre-defined behavior. We know the data going in, and we know the data going out. Anything not matching a predefined format or schema is a problem. Machine Learning models follow the same pattern to operate, but their value diminishes as the content of the data changes. We are looking at the schema, format, and patterns describing a change in the data. This is the big difference between DevOps and MLOps, observing the data.

Most organizations have focused on the simplification, automation, and scalability of Machine Learning applications. Observability has taken a back seat. This session will explore the steps you can take to prepare for ML observability. We will also discuss how observability helps data scientists and MLOps practitioners showcase the business value of the applications they deploy and get recognition for their hard work.

Talk: AI & Sustainability: A $50 Trillion Opportunity

Presenter:
Andrea Ruotolo, Global Head, Sustainability / ESG, Rockwell Automation

About the Speaker:
Andrea is the Global Head of Sustainability/ESG at Rockwell Automation, the world’s largest industrial automation company, with responsibility for advancing innovation in sustainability for Rockwell’s customers, which include Fortune 100 companies in energy and manufacturing, representing millions of employees and hundreds of billions of dollars in annual revenues. Andrea is a passionate evangelist for the role AI/ML can play in dramatically improving the sustainability of the industrial sector.

Across her nearly two decades of experience in leading technology innovation, in a career spanning Europe, Asia, and the Americas, Andrea has held multiple senior executive roles focused on applying advanced technologies to solve the sustainability challenge. She has served as co-founder and entrepreneur in smart grid consulting, global lead in the world’s largest engineering services firm in the energy sector, and senior director at a major utility.

As well as her Fulbright Doctorate in an ESG analysis of sustainable energy systems, Andrea holds a B.A. from the University of la Plata in Argentina, a M.Sci. in Aeronautical and Aerospace Engineering from Madrid Polytechnic, and certification in Digital Business Strategy and AI from MIT Sloan School of Management.

Which talk track does this best fit into?
Lightning Ignite Talk

Abstract of Talk:
The financial and business community have already caught on to the essential importance of sustainability. Investors now call for better practices and reporting on Environment, Social, and Governance, or ESG performance metrics. According to Bloomberg Intelligence, growth in ESG investing is fast becoming the new norm, with ESG investments projected to exceed USD 50 trillion by 2025 – more than 1/3 of all global assets under management. This movement of funds to ESG represents a massive, once-in-a-generation transition to an entirely new economy.

Sustainability is incredibly complex, involving billions of moving parts and decisions. It starts at the edge, where exabytes of data are flowing from real-time sensors and controls in factories and power plants, which aggregate up to the top-level decision makers in companies, which aggregate up to the massive funds that hold portfolios of those companies, and to government regulators and policymakers. AI is critical in analyzing those exabytes of data and enables closed-loop optimization to reduce energy, water, and waste.

In this session, we’ll explore the top 3 needs and opportunities for AI to catalyze change toward more sustainable companies, economies, and societies.

Toronto Machine Learning Summit

Meet Past Speakers

Advanced Technical / Research

Business Strategy

Case Studies

Hands-on Workshops

Virtual Workshops

Lightning Talks

Tickets

This event has ended

Agenda

This event has ended

TMLS

Ihab Ilyas

Talk: Saga: Continuous Construction and Serving of Large Scale Knowledge Graphs

Chip Huyen

Talk: Real-time Machine Learning: Architecture and Challenges

Anne Martel

Talk: Artificial Intelligence And Digital Pathology: Making The Most of Limited Annotated Data

Varun Raj Kompella

Talk: Outracing Champion Gran Turismo Drivers With Deep Reinforcement Learning

Bo Chang

Talk: Latent User Intent Modeling in Recommender Systems

Jinen Setpal

Talk: Interpretability Tools are Feedback Loops

Jesse Cresswell

Talk: Navigating the Tradeoff Between Privacy and Fairness in ML

Sophia Yang

Talk: PyScript for Data Science

Talk: Where's the Road? The Challenge of Autonomous Driving Perception in Winter

Talk: 2022 AI Index Report Briefing

Talk: Tips & Tricks for Intentional Text-to-Image Generation

Lightning Talks w/ Open Group Discussion - Text Generation Using Large Language Models

Talk: Declarative Machine Learning Systems: Ludwig & Predibase

Talk: The Framework for Great ML Products

Danielle Goldfarb

Talk: The Risks of Excluding the Disengaged From your Dataset

Danielle Goldfarb

Talk: Managing AI and the Associated Risks in a Regulatory Environment

Panel: Bias and Relevance in Application of ML in Healthcare

Talk: Data FAILS

Muhammad Mamdan

Talk: Saving Lives with ML: Applications and Learnings

Nikita Medvedev

Talk: The Application of Mobile Location Data for Vending Machine Site Selection and Revenue Optimization

Shiming Ren

Talk: From Silo to Collaboration - Building Tooling to Support Distributed ML Teams at Twitch

Valerii Podymov

Talk: Building a Fully Automated ML Platform Using Kubeflow and Declarative Approach to Development of End-to-End ML Pipelines

Eric Hart

Talk: Optimal Beer Pricing: An Optimization Layer for Price Elasticities

Jawad Ahmed

Talk: Solving Product Substitutions, A Big Problem in Grocery E-Commerce – Through Self-Supervised ML

Quoc Tien Au

Talk: Assessing Alignment of Climate Disclosures Using NLP for the Financial Markets

Talk: Marketplace Experimentation at SpotHero

Hien Luu

Talk: Scaling & Evolving the Machine Learning Platform at DoorDash.

Talk: Sensitivity and Interpretability of AI-Models

Talk: Transforming The Retail Industry with Transformers

Talk: Scaling Advanced Analytics in the Worst Crisis in the Industry Area

Talk: The Role of Alternative Data in Investing

Chloe Pou-Prom

Workshop: NLP for Healthcare: Challenges With Processing and De-Identifying Clinical Notes

Stefanie Molin

Workshop: Beyond the Basics: Data Visualization in Python

Patricia Thaine

Workshop: Demystifying De-Identification

Denys Linkov

Workshop: Iterating on NLP models from R&D to Production

Dr. Nasim Abdollahi

Workshop: Graph Neural Network Modeling in Drug Discovery Using PyTorch

Arthur Vitui

Workshop: Open Source Intelligent Application Delivery on Kubernetes

Jörg Schad

Workshop: Graph ML – The Next Level of Machine Learning

Mahmudul Hasan

Workshop: Introduction to NLP & a Step by Step Implementation of a Real World Use Case from TELUS

Workshop: Train your Models Faster by Learning How to Profile and Apply System-Level Optimizations

Workshop: Distributed Training with PyTorch