CS291A - Special Topics in Foundation Models home | schedule

Announcements

March 31: The course time has changed to Monday/Friday 1:00-2:50pm,

Abstract: This is a graduate-level research course on Transformer and Large Language Models. Over the duration of this course, we will delve into the latest publications within the expansive domain of Large Language Models (LLMs), intelligent agents, and multimodal learning. Each student is entrusted with the following responsibilities: Conducting critical analyses and authoring paper reviews, programming with LLMs, delivering comprehensive paper presentations, and undertaking a substantial, research-quality course project. The intent behind this course is to foster a deep understanding of LLMs and their pivotal role in today's technological landscape.

Each student is expected to read papers before lecture, write paper reviews or program with LLMs, present papers,  and complete a research-quality course project (e.g., implement an existing algorithm or solve a new problem creatively using deep learning. One team could have two students. ).  Projects that simply apply Transformers/LLMs are not encouraged.

Prerequisites: Neural network building experience or successfully finished an introductory deep learning course.

Instructor: Prof. Xifeng Yan

Time: Monday/Friday 1:00- 2:50pm, Location: PHELP 3526   Office Hour:  Wed 10:00-11:00pm, Henley Hall 2017

TA: N/A. 

Grading: Your grade will be derived from paper review or programming (15%),  paper presentation (15%), exam (15%), project presentation (15%), and project (40%)

Paper Reading: You are required to read every paper carefully before lecture.

Paper Review or Programming:  Option 1: Find two Transformer/LLMs related papers that are not presented in lecture but highly related to your project, and write a detailed 1 page review per paper.  Your review shall include your understanding of the work, discuss its strength and weakness, and mostly important, propose new ideas to improve it.  A review will be graded by the quality of the paper (20%) and the quality of your review (80%).  Option 2: programming with systems like LangChain, AutoGen, CrewAI, Mica, etc.  A programming task will be graded by the quality of the task (40%) and the completion and (2 pages) report of your task (60%).

Text Books (not required)