Machine Learning is a science of getting machines to learn, more specifically, designing algorithms that allow computers to learn from empirical data. In the past decade, Machine Learning has successfully made computers to recognize speeches and hand-written characters, to convert spoken words to text, to effectively understanding and generating human language. In this class, you will learn the most important and up-to-date machine learning techniques, not only the theoretical foundations of these techniques, but also the practice implementation of them. The main topics cover the key questions such as what could be learned by an ML model? what are the factors affecting machine learning model? how to address these training challenges? how have LLMs extended the capabilities of machine learning models and how have they been trained under the schema of reinforcement learning, as well as the future of machine learning in advancing Artificial General Intelligence (AGI).
Class time and location: Monday/Wednesday/Friday, 11:30am-12:20pm, Geddes Hall B034
Office Hours: Prof. Xiangliang Zhang, xzhang33@nd.edu, 354 Fitzpatrick Hall. Mondays from 1:00 PM to 2:00 PM, or by appointment. TA: Yue Huang, yhuang37@nd.edu Wednesdays 1:30 PM to 3:30 PM, CSE Commons (151/170 Fitzpatrick), or by appointment.
Assignment and Grading:
| Homework | Reading | Mid-term exam | Final exam |
| 55% | 15% | 15% | 15% |
Readings will be distributed in a document annotation application called Perusall. Perusall is accessible via canvas and creates a collaborative setting for reading: pose a question, answer a question, challenge an assumption, extend a point, etc. The implementation codes and reports (.ipynb file) should be submitted to Canvas prior to the due date. There are no late submissions, and no partial credit will be given if the submission is after the due date. Any extensions must require extenuating circumstances or a priori negotiation. In principle, each one can at most have an extension to only one submission. More details can be found in the course syllabus.
Course Schedule:
| Classes | Topic | Assignment |
| Aug 28 (Wed) | Welcome & Introduction | Reading and Replying on Persuall (due 11am on Sep 4) |
| Unit 1 What could be learned by an ML model? | ||
| Aug 30 (Fri) | The revolution of machine learning | Reading of representation learning (due 11am on Sep 9) Reading of generative models (due 11am on Sep 17) HW Unit1 (due 11:59pm on Sep 25) |
| Sep 2 (Mon) | Learn for prediction | |
| Sep 4 (Wed) | Learn for representation | |
| Sep 6 (Fri) | ||
| Sep 9 (Mon) | Learn for generation | |
| Sep 11 (Wed) | ||
| Sep 16 (Mon) | Learn for decision-making | |
| Unit 2 What are the factors affecting machine learning model? | ||
| Sep 18 (Wed) | Data quality | Reading of overfitting (11am on Oct 2) HW Unit 2 (11:59pm on Oct 8, Tuesday) |
| Sep 20 (Fri) | Over-fitting issue | |
| Sep 23 (Mon) | Model complexity | |
| Sep 25 (Wed) | Hyperparameter | |
| Sep 27 (Fri) | Loss functions | |
| Unit 3 How to address these training challenges? | ||
| Sep 30 (Mon) | Regularization | Reading of Transfer Learning Reading of Auto-ML (11am on Oct 14) HW Unit 3(11:59pm on Oct 31, Thursday) |
| Oct 2 (Wed) | Pre-training + fine-tuning | |
| Oct 4 (Fri) | Transfer learning | |
| Oct 7 (Mon) | Auto Machine Learning | |
| Oct 9 (Wed) | Auto Machine Learning | |
| Oct 11 (Fri) | Midterm Review | |
| Oct 14 (Mon) | Midterm Exam (our class room) | |
| Unit 4 Training ML models with constraints? | ||
| Oct 16 (Wed) | Active learning | Reading of active learning(11am on Oct 28) |
| Oct 18 (Fri) | Semi-supervised learning | |
| Oct 19-27 Mid-term break (no class) | ||
| Unit 5 How have LLMs extended the capabilities of machine learning models? | ||
| Oct 28-Nov 1 (Mon-Fri) | In-context learning | Reading of LLMs’ capabilities (Nov 4, 11am) |
| Nov 4 (Mon) | LLMs as agents and using tools | HW Unit 5 (Nov 19, 11:59pm) |
| Unit 6 How have LLMs been trained? | ||
| Nov 6 (Wed) | Word Vectors | Reading of GPT4 system card (Nov 20, 11am) HW Unit 6 |
| Nov 8 (Fri) | LLMs’ Transformer | |
| Nov 11 (Mon) | ||
| Nov 13 (Wed) | LLMs’ training paradigms | |
| Nov 15 (Fri) | LLMs’ ethical concerns (discussion class) | |
| Unit 7 Reinforcement Learning | ||
| Nov 18 (Mon) | Q-learning | Reading of Reinforcement learning HW Unit 7 |
| Nov 20 (Wed) | Deep Q-learning | |
| Nov 22 (Fri) | Policy gradient | |
| Nov 25 (Mon) | RL and LLMs | |
| Nov 27-Dec 1 Thanksgiving (no class) | ||
| Unit 8 Foundation Models and Beyond | ||
| Dec 2 (Mon) | Foundation models | Reading of foundation models |
| Dec 4 (Wed) | Multi-modal LLMs training | |
| Dec 6 (Fri) | Artificial General Intelligence (AGI) | |
| Dec 9 (Mon) | Foundation applications | |
| Dec 11 (Wed) | Review class | |
| Final exam: Monday, December 16 4:15 PM – 5:30 PM B034 Geddes Hal | ||