Intermediate Machine Learning

S&DS 365 is a second course in machine learning at the advanced undergraduate or beginning graduate level. The course assumes familiarity with the basic ideas and techniques in machine learning, for example as covered in S&DS 265. The course treats methods together with mathematical frameworks that provide intuition and justifications for how and when the methods work. Assignments give students hands-on experience with machine learning techniques, to build the skills needed to adapt approaches to new problems. Topics include nonparametric regression and classification, kernel methods, risk bounds, nonparametric Bayesian approaches, attention and language models, generative models, sparsity and manifolds, and reinforcement learning. Programming is central to the course, and is based on the Python programming language and Jupyter notebooks.

As prerequisites, students are expected to have a background in probability and statistics, at the level of S&DS 242 (Theory of Statistics), familiarity with the core ideas from linear algebra, for example through Math 222 (Linear Algebra with Applications), and computational skills at the level of S&DS 265 (Introductory Machine Learning) or CPSC 200 (Introduction to Information Systems). Background material can be found at the Introductory Machine Learning (S&DS 265) course site.

Computing for the course uses Python in Jupyter notebooks. These can be run using Anaconda with the IML environment that includes the packages needed (click here to download) ; instructions for installing this environment are available on Yale Canvas. The notebooks can also be run in Google Colab by clicking on the icon.

Complementary readings refer to sections in the book Probabilistic Machine Learning: An Introduction, by Kevin Murphy, MIT Press, 2022. Part I “Foundations” gives a good treatment of background in probability, statistics, and linear algebra that is useful for this course. (But this part of the book also covers much more than we need.)

Assignments and quizzes are posted and due on Wednesday in a given week.

Calendar Fall 2024

Lectures: Monday/Wednesday 1:00-2:15pm
WLH 201 - William L. Harkness Hall 201

Week	Dates	Topics	Demos & Tutorials	Lecture Slides	Readings & Notes	Assignments & Exams
1	Aug 28, Aug 30	Sparse regression	Python elements Pandas and regression Lasso example	Wed: Course overview Fri: Sparse regression	PML Section 11.4 Notes on linear regression
2	Sep 4	Smoothing and kernels	Smoothing example Using different kernels	Wed: Smoothing	PML Sections 16.3, 17.1 Notes on computing the lasso	Quiz 1
3	Sep 9, 11	Density estimation and Mercer kernels	Density estimation demo Mercer kernels (1/3) Mercer kernels (2/3) Mercer kernels (3/3)	Mon: Smoothing and density estimation Wed: Mercer kernels	Risk bounds for local smoothing Notes on Mercer kernels	Assn 1 out
4	Sep 16, 18	Neural networks and overparameterized models	np-complete example (1/2) np-complete example (2/2) TensorFlow playground	Mon: Neural networks Wed: Double descent	PML Sections 13.1, 13.2 Notes on backpropagation Notes on double descent	Quiz 2
5	Sep 23, 25	Convolutional neural networks	Convolution demo CNN demo (1/2) CNN demo (2/2)	Mon: Convolutional neural networks Wed: CNNs and Gaussian Processes	PML Section 17.2 Notes on Bayesian inference Notes on nonparametric Bayes	Assn 1 in Assn 2 out
6	Sept 30, Oct 2	Gaussian processes and approximate inference	Parametric Bayes Gaussian processes Gibbs sampling for image denoising	Mon: Gaussian processes Wed: Introduction to approximate inference	Notes on simulation	Quiz 3
7	Oct 7, 9	Variational inference	Variational autoencoders	Mon: Variational inference Wed: VAEs	PML Section 20.3 Notes on variational inference	Assn 2 in Assn 3 out
8	Oct 14	Midterm			Practice midterms	Oct 14: Midterm exam
9	Oct 21, 23	Graphs and structure learning	Graphical lasso demo	Mon: Sparsity and graphs Wed: Discrete data and graph neural nets	Notes on graphs and structure learning Graph neural networks PML Section 23.4
10	Oct 28, Oct 30	Deep reinforcement learning	Q-learning demo DQN demo	Mon: Reinforcement learning Wed: Deep reinforcement learning	Sutton and Barto, Section 6.5	Nov 4: Assn 3 in Assn 4 out
11	Nov 4, 6	Policy gradient methods	Policy gradients demo Actor-critic demo	Mon: Policy gradient methods Wed: Actor-critic methods	Sutton and Barto, Section 13.1-13.3, 13.5	Quiz 4
12	Nov 11, 13	Sequential models	vanilla RNN Fakespeare GRU	Mon: HMMs and RNNs Wed: RNNs, GRUs, LSTMs, and all that	TensorFlow: Text generation Notes on HMMs and Kalman filters PML Chapter 15
13	Nov 18, 20	Sequence-to-sequence models and Transformers	GPT-4 Python API	Mon: Sequence models and attention Wed: Transformers, LLM scaling, finetuning	Notes on mixtures PML Sections 15.4, 15.5	Assn 4 in (Nov. 18) Assn 5 out (Nov. 18) Quiz 5
14	Nov 25, 27	No class, Thanksgiving break
15	Dec 2, 4	Transformers Societal issues	Transformer demo Minimal LLM decoder	Mon: LLM alignment Wed: Course wrap up		Assn 5 in
17	December 15, 2pm WALL53 AUD	Final exam			Practice exams	Registrar: final exam schedule