Emerging Generalization Settings

Program

Modern Paradigms in Generalization

Location

Calvin Lab Auditorium

Date

Monday, Sept. 9 – Friday, Sept. 13, 2024

Back to calendar

8:30 – 8:45 a.m.

Coffee and Check-In

8:45 – 9 a.m.

Welcome and Introductions

9 – 10 a.m.

Propositions as Parameters

Jason Eisner (Johns Hopkins University)

10 – 10:15 a.m.

Break

10:15 – 11:15 a.m.

LLM Generalization in Social Context

Diyi Yang (Stanford University)

11:15 a.m. – 1 p.m.

Lunch (on your own)

1 – 2 p.m.

In-Context Principle Learning from Mistakes

Uri Alon (Google DeepMind)

2 – 2:15 p.m.

Break

2:15 – 3:15 p.m.

LLM Metacognition: Understanding and leveraging "Thinking about Thinking"

Sanjeev Arora (Princeton University)

3:15 – 3:30 p.m.

Break

3:30 – 4:15 p.m.

Panel: Generalization in NLP

4:15 – 5:15 p.m.

Reception

8:30 – 9 a.m.

Coffee and Check-In

9 – 10 a.m.

Out-of-Distribution Generalization as Reasoning: Are LLMs Competitive?

Les Valiant (Harvard University)

10 – 10:15 a.m.

Break

10:15 – 11:15 a.m.

When calibration goes awry: hallucination in language models

Adam Kalai (OpenAI)

11:15 a.m. – 12:45 p.m.

Lunch (on your own)

12:45 – 1:45 p.m.

Generalization in Diffusion Models Arises from Geometry-Adaptive Harmonic Representations

Zahra Kadkhodaie (New York University)

1:45 – 2 p.m.

Break

2 – 3 p.m.

How well does diffusion model generate? - A quantification that accounts for both training and sampling

Molei Tao (Georgia Tech),
Yuqing Wang (Georgia Institute of Technology)

3 – 3:10 p.m.

Break

3:10 – 3:55 p.m.

Panel: New Theoretical Frameworks for Generalization

8:30 – 9 a.m.

Coffee and Check-In

9 – 10 a.m.

Robust Generalization in the Era of LLMs: Jailbreaking Attacks and Defenses

Hamed Hassani (University of Pennsylvania)

10 – 10:15 a.m.

Break

10:15 – 11:15 a.m.

The Pitfalls of Next-token Prediction

Vaishnavh Nagarajan (Google)

11:15 a.m. – 1 p.m.

Lunch (on your own)

1 – 2 p.m.

Predictability and Surprise in Language Model Benchmarks

Sanmi Koyejo (Stanford University)

2 – 2:15 p.m.

Break

2:15 – 3:15 p.m.

Human and machine inductive biases for compositional linguistic generalization

Najoung Kim (Boston University)

3:15 – 3:30 p.m.

Break

3:30 – 4:30 p.m.

Long context modeling and generalization: two perspectives

Amanda Bertsch (Carnegie Mellon University)

8:30 – 9 a.m.

Coffee and Check-In

9 – 10 a.m.

Generalization from the behavioral perspective

Max Raginsky (University of Illinois at Urbana-Champaign)

10 – 10:15 a.m.

Break

10:15 – 11:15 a.m.

Controlling distribution shifts in language models: a data-centric approach.

Tatsunori Hashimoto (Stanford University)

11:15 a.m. – 1 p.m.

Lunch (on your own)

1 – 2 p.m.

On the Curses of Future and History in Off-policy Evaluation in non-Markov Environments

Nan Jiang (University of Illinois Urbana-Champaign)

2 – 2:15 p.m.

Break

2:15 – 3:15 p.m.

Iterative preference learning methods for large language model post training

Wei Xiong (UIUC)

3:15 – 3:30 p.m.

Break

3:30 – 4:30 p.m.

Language Model Alignment: Theory & Algorithms

Ahmad Beirami (Google)

8:30 – 9 a.m.

Coffee and Check-In

9 – 10 a.m.

Unexpected Test Losses from Generalization Theory?

Frederic Koehler (University of Chicago)

10 – 10:15 a.m.

Break

10:15 – 11:15 a.m.

When do dependencies in your data help?

Ankur Moitra (Massachusetts Institute of Technology)

11:15 a.m. – 1 p.m.

Lunch (on your own)

1 – 2 p.m.

Know When You Know: Handling Adversarial Data by Abstaining

Surbhi Goel (University of Pennsylvania)

2 – 2:15 p.m.

Break

2:15 – 3:15 p.m.

Neural Networks meet Nonparametric Regression: Generalization by Weight Decay and Large Learning Rate

Yu-Xiang Wang (UC San Diego)

3:15 – 3:30 p.m.

Break

3:30 – 4:30 p.m.

Generalization in Robotic Behavior Cloning: Pitfalls and Promises

Max Simchowitz (Carnegie Mellon University)