WIP
Documents: Summary (WIP).
WIP
Documents: Summary (WIP).
This course provides an in-depth theoretical treatment of classical and modern optimization methods that are relevant in data science. During the lectures, algorithms are presented and analyzed in the following four categories: first-order methods (gradient and coordinate descent, Frank-Wolfe, subgradient and mirror descent, stochastic and incremental gradient methods); second-order methods (Newton and quasi Newton methods); non-convexity (local convergence, provable global convergence, cone programming, convex relaxations); and min-max optimization (extragradient methods). The emphasis is on the motivations and design principles behind the algorithms, on provable performance bounds, and on the mathematical tools and techniques to prove them.
Documents: Summary, Cheatsheet.
The goal of this course is to enable students to connect their mathematical background in linear algebra, analysis, probability, and optimization with their basic knowledge in machine learning and their general skill set in computer science to gain a deeper understanding of models and tools of great practical impact. Students will acquire fundamental theoretical concepts and methodologies from machine learning and how to apply these techniques to build intelligent systems.
Documents: Summary.
This course is a deep dive into deep learning algorithms and architectures with applications to a variety of perceptual and generative tasks.
Documents: Summary, Cheatsheet.
This course comprises fundamental NLP algorithms for efficiently computing probability distributions over structures. It covers backpropagation, log-linear modelling, multi-layer perceptrons, recurrent neural networks, language modelling, semirings for generalizing algorithms, part-of-speech tagging with conditional random fields and the backward/forward algorithm, transliteration with finite-state automata and Lehmann’s algorithm, constituency parsing with the CKY algorithm, dependency parsing with the Chu-Liu-Edmonds algorithm, semantic parsing with linear-indexed grammars and combinatory categorial grammars, transformers in the context of translation.
Documents: Summary, Cheatsheet.
This course comprises many computer vision applications and gives an overview of the research within that area. It covers the Harris detector, deep learning, optical flow, recognition, segmentation, object detection, tracking, projective geometry, the camera model, epipolar geometry, structure from motion, and stereo matching.
Documents: Summary.
This course consisted of solving many problems using algorithms learned during a datastructures and algorithms class, in C++. Methods covered include binary search, graph algorithms, infinite precision computing, greedy algorithms, split and list, maximum flow, minimum cost maximum flow, Delaunay triangulation, and linear programming.
Documents: Summary.
This course first covers how to represent uncertainty using Bayesian linear regression, Gaussian processes, and Bayesian neural networks, and goes over several methods for making the intractable probability distribution over outputs tractable. The covered methods are variational inference and Markov chain Monte Carlo. Secondly, it covers how to make use of this uncertainty in concepts such as active learning, Bayesian optimization, Markov decision processes, and reinforcement learning.
Documents: Summary, Cheatsheet.