Machine learning (ML) is a paradigm shift across many sectors because it proposes the automatic construction of artificially intelligent systems that are human-like in ability to perform and adapt. It has already made well-publicized breakthroughs in transforming massive amounts of unstructured data into commercially-relevant products and services. Accordingly, the major internet giants have made recent acquisitions and investments in this space as they expand their capacity in AI. A recent study by Bloomberg Beta estimated over 2,500 startup companies also working in AI and ML. This course will begin by introducing the core principles of machine learning. It will then focus on Deep Learning: techniques that learn multiple layers of representation. It will review core approaches for supervised learning: deep neural networks, backpropagation, and optimization methods. It will also review unsupervised learning techniques, including recent advances in deep generative models.
A central challenge in automated visual reasoning is that of untangling the many factors of variation that explain an image or video: both nuisance factors (e.g. lighting, scale, camera angle) and variables of interest (e.g. person or object identity). Historically, practitioners relied on an engineered feature extraction pipeline, usually containing multiple stages of processing combined with simple machine learning techniques. Recently, Deep Learning methods have transformed the field producing winning entries to myriad competitions as well as industrial applications. This course will review the foundations of Deep Learning applied to vision including contemporary convolutional network architectures. Leading experts in the field will discuss the most relevant application areas, including object detection, structured prediction, large-scale classification and hardware acceleration, video, multi-modal and multi-task learning, and regression methods for localization. The course will also highlight the most frequently used practical development libraries and tools.
Applications of Natural Language Processing (NLP) systems are everywhere: web search, translation services, and recommender systems, are only a few examples. Understanding language is not just the core to so many products and services, it is a critical component of building strong AI systems. NLP is one area which has been deeply affected by the recent advances in Deep Learning. This course will review NLP fundamentals as well as cutting-edge models fueled by DL. Topics include word embeddings, recurrent neural network models, encoder-decoder architectures, attention models, architectures with external memories, and multimodal learning, including image and video captioning systems.
This course explores practical aspects of Machine Learning in industrial settings. It will explore the interaction between software and hardware, specifically Graphics Processing Units (GPUs) and other hardware accelerators which have been critical to scaling up ML. It will cover best practices for ML developers, such as model search, debugging, and visualization. Short and long-term ethics and implications of strong AI will form a key discussion point. Machine Learning Systems have a larger system-level complexity than traditional software-based systems and thus have the potential to incur massive ongoing maintenance costs. The course will also explore the challenges and best practices of building and deploying large-scale ML systems.
Mara does empirical research in the areas of Industrial Organization and Organizational Economics. At a broad level, Mara studies how companies compete and how they organize themselves for competitive advantage. She is best known for her research on loyalty programs and vertical integration. Much of her work has been focused on the airline industry and she is recognized as an expert in this area. Her work has been published in the American Economic Review, the RAND Journal of Economics and the Review of Economics and Statistics, among others. On the teaching side, Mara delivers courses on strategy, data analytics, and business problem-solving in Rotman’s MBA and Executive Education programs.
Joelle Pineau is a William Dawson Scholar and Associate Professor in the School of Computer Science at McGill University. Dr. Pineau’s research focuses on developing new models and algorithms that allow computers to learn to make good decisions in complex real-world domains, even in circumstances where there is incomplete or incorrect information. She also works on applying these algorithms to problems in robotics and health care. Dr. Pineau is a Senior Fellow of the Canadian Institute for Advanced Research, and a Member of the College of New Scholars, Artists and Scientists in the Royal Society of Canada.