Fan Feng

Email: ffeng1017 [at] gmail [dot] com

I am a postdoctoral researcher at UCSD and the CMU/MBZUAI CLeaR group, working with Kun Zhang and Biwei Huang. Previously, I also worked closely with Sara Magliacane at the University of Amsterdam. I completed my PhD at City University of Hong Kong.

Feel free to contact me for research collaborations or other engagements.

Research Interests

My long-term research goal is to build agents that not only imagine the world, but also understand it, act in it, discover goals within it, and continually refine their internal models in an open-ended, self-improving loop. To achieve this, I study world models, RL, and robotics. Specifically, I focus on three closely related questions:

Generalization. How can we learn world models from large-scale interactive data that are sufficient for control and planning under broad shifts in environments, tasks, and data distributions? Works include: continual adaptation to unseen domains: CSR, adaptation to non-stationary changes: Factored-RL, multi-domain and multi-task adaptation: AdaRL, and general adaptation for control and planning: Ada-Diffuser.
Compositionality. How can agents compositionally reuse learned structure across new tasks and environments in world models? Works include: interactive world models with hierarchical factorization: FIOC-WM and compositional world models: DAFT-WM.
Agentic Exploration and Self-Improvement. How can agents actively explore in a purposeful and open-ended way to discover goals, achieve them, and simultaneously improve their world models? Works include: discovering informative interactions for world model learning: ECL, improving world models through targeted interaction: CIP, reasoning and goal achievement with interactive memory: HInt & DreamSAC, and self-improving world models: WAV.

Publications (* marks equal contribution.)

🌟 Selected 📚 Topic: World Model & RL | General ML | Robotics and Vision

Service

Workshop Organization: ICLR 2025 ICBINB Workshop; RLC 2024 RLBRew Workshop; NeurIPS 2022-2023 ICBINB Workshop
Area Chair: NeurIPS 2026, KDD 2025-2026 ADS Track
Proposal Reviewer: ICML/NeurIPS/ICLR Workshop Proposals 2023-2025
Reviewer: ICML, NeurIPS, ICLR, UAI, AISTATS, CLeaR, CVPR, ECCV, ICCV, WACV, AAAI, IJCAI, ICRA, JMLR, TMLR, TPAMI, TNNLS, etc.

Talks

[05/2026] Learning Causal Latent-Action World Models, at JHU Workshop on Measurement Errors & Latent Variables Series 2026 .
[01/2026] Towards Causal and Structued World Models, at Basis Research Institute .
[11/2025] Causal World Models for Generalizable and Adaptive Decision-Making, at JHU Workshop on Measurement Errors & Latent Variables .
[08/2025] Learning Causal Representation for Efficient and Adaptive Decision-Making Agents in the Physical World, at NSF IAIFI Summer Workshop, Harvard University .
[05/2025] Learning and Using Causal Representation for Reinforcement Learning, at JHU Workshop on Measurement Errors & Latent Variables .
[04/2024] Structured and causal modeling for adaptive RL agents, at SCALAR Lab@UMass Amherst .
[03/2023] Causal Reinforcement Learning, at Swarma Club .
[05/2022] Factored Adaptation in heterogeneous and non-stationary RL, at Dr. Herke van Hoof's group in AMLab UvA. [Slides]
[03/2022] Factored Adaptation in non-stationary RL, at INDELab in UvA. [Slides]

Misc.

I am also actively engaged in community building. I co-organize the “I Can’t Believe It’s Not Better” (ICBINB) workshop series, which brings together researchers to discuss the practical challenges, failure modes, and lessons learned in deploying ML systems, to promote a culture of open, constructive discourse around what does not work and how we can collectively build more robust and reliable learning systems:)

Fan Feng

Research Interests

Publications (* marks equal contribution.)

Learning Task-Sufficient World Models by Synergizing Agentic Exploration and Structured Modeling

From Generalist to Specialist Representation

On Information Self-Locking in Reinforcement Learning for Active Reasoning of LLM Agents

Factored Causal Representation Learning for Robust Reward Modeling in RLHF

Ada-Diffuser: Latent-Aware Adaptive Diffusion for Decision-Making

Self-Improving World Models via Asymmetric Forward-Inverse Consistency

DreamSAC: Learning Hamiltonian World Models via Symmetry Exploration

Learning Interactive World Model for Object-Centric Reinforcement Learning

Online Time Series Forecasting with Theoretical Guarantees

Null Counterfactual Factor Interactions for Goal-Conditioned Reinforcement Learning

Towards Empowerment Gain through Causal Structure Learning in Model-Based Reinforcement Learning

Causal Information Prioritization for Efficient Reinforcement Learning

Towards Generalizable Reinforcement Learning via Causality-Guided Self-Adaptive Representations

Learning Dynamic Attribute-factored World Models for Efficient Multi-object Reinforcement Learning

Factored Adaptation for Non-Stationary Reinforcement Learning

AdaRL: What, Where, and How to Adapt in Transfer Reinforcement Learning

Community Channel-Net: Efficient Channel-wise Interactions via Community Graph Topology

Power Law in Deep Neural Networks: Sparse Network Generation and Continual Learning with Preferential Attachment

Towards Lifelong Object Recognition: A Dataset and Benchmark

OpenLORIS-Object: A Robotic Vision Dataset and Benchmark for Lifelong Deep Learning

Service

Talks

Misc.