Teaching

ECE5424: Advanced Machine Learning, Spring 2022

  • Description: Algorithms and principles involved in machine learning; focus on perception problems arising in computer vision and robotics; fundamentals of representing uncertainty, learning from data, supervised learning, ensemble methods, unsupervised learning, structured models, learning theory and reinforcement learning; design and analysis of machine perception systems; design and implementation of a technical project applied to real-world datasets (images, text, robotics). This is a theoretical course with focus on the foundations of modern machine learning.
  • Why take this course? We are witnessing an explosion in data from billions of images shared online to Petabytes of tweets, medical records and GPS tracks, generated by companies, users and scientific communities. Applications of machine learning and perception are increasing rapidly as more techniques are developed and implemented to address a wide range of scientific and societal problems. Students trained in a deeper understanding of machine learning techniques will be better equipped to make fundamental contributions to research in machine learning, and applied areas such as perception (vision, text, speech), robotics, bioinformatics, etc.
  • Please visit Canvas for details and updates.
  • Course reviews: 2022 (ECE, CS)

ECE5984 Special topic: Optimization Theory for ML, Spring 2021

  • Description: Fundamentals of optimization theory for applications in machine learning. Overview of convex analysis and formulations (e.g., linear, quadratic, second-order cone, and semidefinite programs, and conic programming), first-order methods (e.g., gradient descent, stochastic gradient descent and variants, cutting plane methods, mirror descent), second-order methods (e.g., interior point method), as well as a subset of the advanced topics, such as online optimization and non-convex optimization.
  • Why take this course? With the rapid advancements in the field of machine learning, it is crucial to develop a systematic understanding of modern optimization theory to keep up with the state-of-the-art methods and to even push the boundaries and design new machine learning methods. Indeed, the interplay between optimization and machine learning is one of the most fascinating aspect of modern computational science. However, these recently developments have not been well-documented in textbooks and are often dispersed in research papers, which make it difficult for students to grasp the gist in a short amount of time. This course is designed to bridge this gap and introduce to students these advanced topics in optimization and ML.
  • Please visit Canvas for details and updates.

ECE4424/CS4824: Machine learning, Fall’20, Fall’21, Fall’22

  • Description: Algorithms and principles involved in machine learning with applications to various engineering domains; fundamentals of representing uncertainty, learning from data, supervised learning, unsupervised learning, and learning theory; design and analysis of machine learning systems; design and implementation of a technical project applied to real-world datasets (energy, images, text, robotics, etc.).
  • Why take this course? We are witnessing an explosion in data - from billions of images shared online to Petabytes of tweets, medical records and sensor data, generated by companies, users and the infrastructures around us. Applications of machine learning are increasing rapidly as more techniques are developed and implemented to address a wide range of scientific and societal problems. Many universities are expanding programs in machine learning and perception, and employers are increasingly recognizing the importance of such knowledge. The course will give students a solid foundation in the basics of machine learning and an introduction to the opportunities in this rapidly maturing field.
  • Please visit Canvas for details and updates.
  • Term project page
  • Course reviews: 2021 (ECE, CS), 2020 (ECE, CS)

ECE5984 Special topic: Trustworthy reinforcement learning

  • Description: Fundamentals of control theory (including robust control, adaptive control) and learning theory with applications to the understanding and analysis of reinforcement learning algorithms. Overview of contemporary approaches to obtaining assurances of learning-enabled dynamical systems, including control-theoretic approaches for safety and stability, learning-to-learn approaches for sample efficiency, as well as a subset of the advanced topics, such as offline reinforcement learning and multi-agent reinforcement learning.
  • Why take this course? Despite the progress in the past decades, control-theoretic techniques are challenged by the gap between theory and reality widened by the growing complexity, volatility, and scale of modern control systems, from autonomous cars to power grid and smart cities. Learning-based control techniques, especially reinforcement learning (RL), have a vast potential to enhance system performance and adaptivity. However, unlike control-theoretic approaches, these methods lack the necessary mathematical framework to provide guarantees on correctness, such as stability and physical constraint satisfaction, causing concerns about safety. They also tend to be data hungry. This course will provide students with an introduction to current areas of research at the intersection of machine learning and control. Topics of study will include fundamentals of control theory (including robust control, nonlinear control, adaptive control) and learning theory (including concentration inequalities, regret bounds, generalization bounds) with applications to the analysis of reinforcement learning algorithms. We will also give overviews of contemporary approaches to obtaining assurances of learning-enabled dynamical systems, including control-theoretic approaches for safety and stability (e.g., Lyapunov function, control barrier function), learning-to-learn approaches for sample efficiency, as well as a subset of the advanced topics, such as offline reinforcement learning and multi-agent reinforcement learning. These recently developments have not been well-documented in textbooks and are often dispersed in research papers, which make it difficult for students to grasp the gist in a short amount of time. This course is designed to help students develop a systematic understanding of these interrelated fields to keep up with the state-of-the-art methods and to even push the boundaries and design new reinforcement learning methods. Indeed, the interplay between control, optimization and machine learning is one of the most fascinating aspect of modern automation technology, and hence well deserves a course that is dedicated to it.
  • Please visit Canvas for details and updates.