About

Table of contents

1. About the Class
2. Prerequisites
3. Assignments / Grading
   1. Participation and Discussion (15%)
   2. Presentations (25%)
   3. Course project (60%; group of 1~3 students)

About the Class

This is a discussion-based research-oriented class. This is not a lecture-based class (if you prefer lectures, take CS 7650). In each class, students will present and lead discussion of recent research papers on large language papers. This class is designed for students who want to do research, and students will conduct a self-directed research project in this class.

This class will have a heavy course load, highly challenging, intended for PhD students and MS students who have enough time, motivations, and capabilities (i.e., data analytics, creativity, programming, technical writing, critical thinking) to comprehend and conduct research at the top NLP research level.

Bonus point: some students may continue on working in the NLP X Lab at the Georgia Tech after the class and publish their projects at top ML/NLP conferences!

Prerequisites

This is an advanced-level research-oriented class. The class will require a good understanding of machine learning algorithms (CS 4641/7641), deep learning models (CS 4644/7643), and NLP techniques (CS 4650/7650). For most of the classes, one or two students will present an assigned research paper in-depth and lead the discussion in class.

Generally speaking, if you want to learn about NLP and want lectures given by an instructor, you should take CS 7650.

Yet, if you have enough background, we do encourage you to take this 8803-LLM class in your first year of your graduate school – especially if you want to complete a paper for publication and/or if you plan to pursue a Ph.D. later.

Below are some example papers which students will be asked to present and discuss in the classes:

• “Direct Preference Optimization - Your Language Model is Secretly a Reward Model”
• “What do BPE tokenizers reveal about their training data?”

Assignments / Grading

Participation and Discussion (15%)

This class will require 100% in-person attendance for every class. We will take attendance in the class.

For most of the classes, we will discuss one research paper (i.e., analyze its strengths and weaknesses). Every student will read and write a critique about the paper before the class, and engage in discussions in class.

Presentations (25%)

Students will give multiple presentations of recent research papers (with varied length of time, e.g., 30 minutes or 2 minutes).

Course project (60%; group of 1~3 students)

Students will work in group of 1~3 to propose, conduct, give presentations, and write written reports of a self-directed research project.

Deliverables:

• Literature Review
• Project Proposal
• Midway Report
• Final Report

Rubics:

• Clarity: For the reasonably well-prepared reader, is it clear what was done and why? Is the report well-written and well structured?
• Originality / Innovativeness: How original is the approach? Does this project break new ground in topic, methodology, or content? How exciting and innovative is the work that it describes?
• Soundness / Correctness: First, is the technical approach sound and well-chosen? Second, can one trust the claims of the report – are they supported by proper experiments, proofs, or other argumentation?
• Meaningful Comparison: Does the author make clear where the problems and methods sit with respect to existing literature? Are any experimental results meaningfully compared with the best prior approaches?
• Substance: Does this project have enough substance, or would it benefit from more ideas or results?  Note that this question mainly concerns the amount of work; its quality is evaluated in other categories.