Jubayer Ibn Hamid

I am an incoming Ph.D. student, currently doing research at Stanford where I am advised by Chelsea Finn and Dorsa Sadigh. I studied Mathematical Physics (B.S) and Computer Science (M.S), both at Stanford University.

I work in artificial intelligence with a focus on the intersection of reinforcement learning, generative models and representation learning. I am also interested in pure mathematics such as abstract algebra, category theory and algebraic geometry.

CV  /  Scholar  /  Email  /  Twitter    

My research interests in building embodied agents are in online reinforcement learning fine-tuning, test-time decoding methods from complex multimodal policies and long-context policies. I am also interested in understanding what kinds of data we should scale for better generalization and why. Recently, I have become interested in sample-efficient methods for preference fine-tuning generative models.

(*) denotes co-first authorship

• Bidirectional Decoding: Improving Action Chunking via Guided Test-Time Sampling. Yuejiang Liu*, Jubayer Ibn Hamid*, Annie Xie, Yoonho Lee, Max Du, Chelsea Finn. International Conference on Learning Representations (ICLR), 2025. (Website)

• Tripod: Three Complementary Inductive Biases for Disentangled Representation Learning. Kyle Hsu*, Jubayer Ibn Hamid*, Kaylee Burns, Chelsea Finn, Jiajun Wu. International Conference on Machine Learning (ICML), 2024.

• What Makes Pre-trained Visual Representations Successful For Robust Manipulation? Kaylee Burns, Zach Witzel, Jubayer Ibn Hamid, Tianhe Yu, Chelsea Finn, Karol Hausman. Conference on Robot Learning (CoRL), 2024.

Here are some introductory notes on various topics that have fascinated me. These are not meant to be in-depth. Rather, they are meant to cover some of the basic constructions that show up periodically and are also interesting in and of themselves.

• Category Theory and Algebraic Geometry. Notes on introductory category theory and foundational constructions in algebraic geometry (work in progress).

• Algebraic Topology. Foundational constructions – fundamental group, homology and cohomology. (Incomplete, will typeset later).

• Whitney's Embedding Theorem and Immersion Theorem. Weak versions of Whitney's Embedding and Immersion theorem, which are often sufficient.

• Policy Gradient Methods. Building blocks of policy gradient methods in online reinforcement learning.

• Trust Region Optimization Methods. Introduction to trust region methods for optimization.

Talks

Bidirectional Decoding. OpenAI. 25th February, 2025.

Teaching

CS 224R - Deep Reinforcement Learning : Head CA. Spring, 2025.

CS 229 - Machine Learning : CA. Winter, 2025.