Meetings are held every Wednesday in CoDa E201 at 4pm. Please sign up to be a discussant here! Goal(s) of the discussant group is to:

  • Do a single “deep dive” per week about one subject (can be multiple papers)
  • We have suggested several papers for each week, more than one can cover thoroughly in a week. Pick a small, focused set of papers and read them thoroughly
  • Prepare a 20-30 minute presentation, accessible to a second year PhD student, focusing on (a) seeding discussion and (b) identifying gaps and connections, and (c) formulating open problems

Signing up is a great way to (1) force yourself to engage with the content of the paper (2) get to know your co-discussant(s) and (3) ensure the success of the reading group!

Upcoming Sessions

Date Topic Resources
2024-10-16 Introduction Slides
2024-10-23 Scaling Laws 1 (Training Compute-Optimal Language Models) Paper Slides
2024-10-30 Scaling Laws 2 (Explaining Neural Scaling Laws) Paper Slides
2024-11-06 Data Selection 1 (Perplexity Correlations, Scaling Laws + Data Filtering) Paper 1 Paper 2 Slides
2024-11-13 Data Selection 2 (DsDm, LESS) Paper 1 Paper 2 Tutorial Slides (DsDm) Slides (LESS)
2024-11-20 Data Selection 3 (Statistical Theory) Paper Slides
2024-11-20 Data Selection 3 (Pruning, Prediction) Paper 1 Paper 2
2025-01-22 Post-training 1 (RLHF, AlpacaFarm) Paper 1 Paper 2 Slides
2025-01-29 No meeting (ICML Deadline)  
2025-02-05 Post-training 2 (Direct methods & Offline RL) Paper 1 Paper 2 Paper 3 Paper 4 Slides 1 Slides 2
2025-02-12 Post-training 3 (DeepSeek) Paper Slides
2025-02-19 Post-training 4 (Synthetic Data) Slides
2025-02-26 Post-training 4 (Synthetic Data & Self-Improvement) Paper 1 Paper 2 Slides
2025-03-04 Post-training 5 (Simplicity) Paper 1 Paper 2 Slides
2025-03-11 Post-training 5 (In-Context Learning) Slides