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 we’ll need (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.

Readings marked are written at a more advanced and technical level, and give further detail and as a complement to what is presented in class; students are not responsible for the parts that are not discussed in lecture. Additional readings are listed as “Readings: PML” and refer to sections in the book Probabilistic Machine Learning: An Introduction, by Kevin Murphy, MIT Press, 2022 (forthcoming).

Calendar Spring 2022

Lectures: Monday/Wednesday 11:30-12:50

Week	Dates	Topics	Demos & Tutorials	Lecture Slides	Readings & Notes	Assignments & Exams
1	Jan 26, 28	Course overview	Python elements Pandas and regression Lasso example	Jan 26: Course overview Jan 28: Sparse regression	PML Section 11.4
2	Jan 31, Feb 2	Smoothing and kernels	Smoothing example Using different kernels Mercer kernels	Jan 31: Smoothing Feb 2: Mercer Kernels	PML Sections 16.3, 17.1 Notes on Mercer kernels
3	Feb 7, 9	Density estimation and risk bounds	Density estimation demo	Feb 7, 9: Sync up	Bias-variance tradeoff for density estimation	Feb 9: Assn1 out
4	Feb 14, 16	Neural networks for classification	TensorFlow playground Convolution demo Problem 4 warmup	Feb 9: Neural networks Feb 14: Convolutional neural networks Feb 16: CNNs continued	PML Sections 13.1, 13.2 Notes on backpropagation	Feb 16: Quiz 1
5	Feb 21, 23	Nonparametric Bayes	Parametric Bayes Gaussian processes Dirichlet processes	Feb 21: Gaussian processes Feb 23: Gaussian and Dirichlet processes	PML Section 17.2 Notes on Bayesian inference Notes on nonparametric Bayes	Feb 23: Assn 1 in; Assn2 out
6	Feb 28, Mar 2	Gibbs sampling	DP demo, ver. 2 Gibbs sampling	Feb 28: Dirichlet processes Mar 2: Gibbs sampling	Notes on Gibbs sampling	Mar 2: Quiz 2
7	Mar 7, 9	Variational inference	Variational autoencoders	Mar 7: Introduction to approximate inference Mar 9: Variational inference and VAEs	PML Section 20.3 Notes on variational inference	Mar 9: Assn 2 in
8	Mar 14, 16	Review and midterm		Mar 14: VAEs and review Mar 16: Midterm	Practice midterm	Mar 16: Midterm exam
9	Mar 28, 30	Graphs and structure learning	Graphical lasso demo Graph neural networks	Mar 28: Sparsity and graphs Mar 30: Discrete data and graph neural nets	Notes on graphs and structure learning PML Section 23.4	Mar 30: Assn3 out
10	Apr 4, 6	Deep reinforcement learning	Q-learning demo	Apr 4: Reinforcement learning Apr 6: Deep reinforcement learning	Sutton and Barto, Section 6.5	Apr 6: Quiz 3
11	Apr 11, 13	Policy gradient methods	DQN demo Policy gradients demo Actor-critic demo	Apr 11: Policy gradient methods Apr 13: Actor-critic methods	Sutton and Barto, Section 13.1-13.3, 13.5	Apr 13: Assn 3 in Assn4 out
12	Apr 18, 20	Sequential and sequence-to-sequence models	RNN demo: Fakespeare TensorFlow: Text generation	Apr 18: HMMs and RNNs Apr 20: RNNs, GRUs, LSTMs, and all that	PML Chapter 15	Apr 20: Quiz 4
13	Apr 25, 27	Attention and language models	GPT-3 demo Codex demo	Apr 25: Sequence-to-sequence models and attention Apr 27: Course wrap up	PML Sections 15.4, 15.5	Apr 27: Assn 4 in
	May 7	Final exam, 2pm in SPL 59				Practice final Registrar: final exam schedule