Lecture Times: Tuesdays and Thursdays, 11:30am-12:45pm Eastern
Lecture:11:30am-12:45pm Tuesdays and Thursdays
Zoom link: https://umass-amherst.zoom.us/j/94525449604
The course provides an introduction to machine learning algorithms and applications. Machine learning algorithms answer the question: "How can a computer improve its performance based on data and from its own experience?" The course is roughly divided into thirds: supervised learning (learning from labeled data), reinforcement learning (learning via trial and error), and real-world considerations like ethics, safety, and fairness. Specific topics include linear and non-linear regression, (stochastic) gradient descent, neural networks, backpropagation, classification, Markov decision processes, state-value and action-value functions, temporal difference learning, actor-critic algorithms, the reward prediction error hypothesis for dopamine, connectionism for philosophy of mind, and ethics, safety, and fairness considerations when applying machine learning to real-world problems.
The TAs for this course are Cooper Sigrist (csigrist@umass.edu) and Scott Jordan (sjordan@cs.umass.edu). Scott will primarily be handling assignments and grading, and so you should ask him questions related to grading. Cooper will primarily be holding office hours.
Office hours will be at the following times:
Day | Time | Person | Link |
---|---|---|---|
Monday | 8:00am-9:45am | Cooper Sigrist | link |
Tuesday | 1:00pm-3:00pm | Cooper Sigrist | link |
Wednesday | 4:00pm-6:00pm | Philip Thomas | link |
Thursday | 8:00am-9:45am | Cooper Sigrist | link |
Friday | 1:00pm-3:00pm | Cooper Sigrist | link |
Office hours will follow the UMass Academic Calendar [link]. For example, Monday March 1 will follow a Wednesday schedule, and so Philip Thomas will be holding office hours and Cooper Sigrist will not. Office hours will run up to and including the last day of classes, May 4.
Lecture | Topic | Reading | Whiteboard |
---|---|---|---|
1 | Introduction | Chapter 1 (course notes) | link |
2 | Regression, k-Nearest Neighbors, Linear Regression I | Chapter 2 (course notes) | link |
3 | Linear Regression II | Chapter 3 | link |
4 | Linear Regression III, Gradient Descent | Chapter 4 | link |
5 | Gradient Descent (continued) | Chapter 5 | link |
6 | Basis functions, feature normalization, perceptrons | Chapter 6 | link |
7 | Perceptrons | Chapter 7 | link |
8 | Artificial Neural Networks | Chapter 8 | link |
9 | Backpropagation | Chapter 9 | link |
10 | Supervised Learning - Other Topics | Chapter 10 | link |
Lecture | Topic | Reading | Slides |
---|---|---|---|
11 | Introduction | Chapter 11 | link |
12 | MENACE, Notation, and Problem Formulation | Chapter 12 | link |
13 | Episodes and Policy Representations | Chapter 13 | link |
14 | Midterm Solutions and Linear Softmax Policies | Chapter 13 (linear softmax content added) | link |
15 | MENACE-like RL Algorithm | Chapter 14 | link |
16 | Value functions and TD error | Chapter 15 | link |
17 | Review | No readings | No whiteboard |
18 | Actor-Critics, Options, and Off-Policy Evaluation | Chapter 16 | link |
Lecture | Topic | Reading | Slides |
---|---|---|---|
Connections to psychology and neuroscience | Sutton and Barto Chapters 14 and 15 [link] A Neural Substrate of Prediction and Reward [link] Uncertainty-based competition between prefrontal and dorsolateral striatal systems for behavioral control [link] Gero Miesenboeck TED Talk [link] |
link | |
Fairness, Accountability, and Transparency | Slides posted on Moodle | No whiteboard | |
Philosophy of Mind | Slides posted on Moodle | No whiteboard | |
Ethics | See Moodle for Google Docs | No whiteboard | |
Ethics and Safety | See Moodle for Google Docs | link | |
Final Exam Review | See Moodle for Google Docs | link |