|
Yibo JiangGoogle Scholar / Twitter / Email I am a Computer Science PhD student at the University of Chicago, advised by Prof. Victor Veitch. I also work very closely with Prof. Bryon Aragam. Previously, I received a MS in Computer Science from Columbia University, and graduated, as a Bronze Tablet recipient, from the University of Illinois Urbana-Champaign with double degrees in Electrical Engineering and Math. Before starting my PhD, I was also a research fellow at Harvard, where I worked with Prof. Cengiz Pehlevan in the Theoretical Neuroscience Group. I have also interned at ByteDance (AML) and (a couple of times) at Nvidia (AI Dev Tech). |
ResearchI am broadly interested in robust/trustworthy machine learning, representation learning, causality, and more recently, the interpretability of large language models (LLMs). One speficic reseach goal of mine is to study the emergent structures and dynamics within representations, as well as complex networks arising from the training of modern machine learning models. My aim is to develop theories and methods for interpreting and ultimately utilizing these insights for alignment purposes. I like collaborations; If you have a cool problem, don't hesitate to reach out – let's explore it together! |
Publications
The Geometry of Categorical and Hierarchical Concepts in Large Language ModelsKiho Park, Yo Joong Choe, Yibo Jiang, Victor Veitch ICML Mechanistic Interpretability Workshop, 2024 (Best Paper Award) arxiv |
Do LLMs Dream of Elephants (When Told Not to)? Latent Concept Association and Associative Memory in TransformersYibo Jiang, Goutham Rajendran, Pradeep Ravikumar, Bryon Aragam Advances in Neural Information Processing Systems (NeurIPS), 2024 arxiv |
On the Origins of Linear Representations in Large Language ModelsYibo Jiang*, Goutham Rajendran*, Pradeep Ravikumar, Bryon Aragam, Victor Veitch International Conference on Machine Learning (ICML), 2024 arxiv |
Beyond Reverse KL: Generalizing Direct Preference Optimization with Diverse Divergence ConstraintsChaoqi Wang, Yibo Jiang, Chenghao Yang, Han Liu, Yuxin Chen International Conference on Learning Representations (ICLR), 2024 (Spotlight, 5%) arxiv / code |
Uncovering Meanings of Embeddings via Partial OrthogonalityYibo Jiang, Bryon Aragam, Victor Veitch Advances in Neural Information Processing Systems (NeurIPS), 2023 arxiv / video |
Learning Nonparametric Latent Causal Graphs with Unknown InterventionsYibo Jiang, Bryon Aragam Advances in Neural Information Processing Systems (NeurIPS), 2023 arxiv / video |
Invariant and Transportable Representations for Anti-Causal Domain ShiftsYibo Jiang, Victor Veitch Advances in Neural Information Processing Systems (NeurIPS), 2022 arxiv / video / code |
Associative Memory in Iterated Overparameterized Sigmoid AutoencodersYibo Jiang, Cengiz Pehlevan International Conference on Machine Learning (ICML), 2020 arxiv / video / code |
Meta-Learning to ClusterYibo Jiang, Nakul Verma arxiv Preprint, 2019 arxiv |
Model-Agnostic Meta-Learning using Runge-Kutta MethodsDaniel Jiwoong Im, Yibo Jiang, Nakul Verma arxiv Preprint, 2019 arxiv |
Academic ServicesConference Reviewer: NeurIPS, ICML, ICLR, AISTATS, KDD, AAAI |
|
Profile picture taken by Qian Sheng. Last updated Sept. 2024. |