Fangchen Liu


I received my M.S. from UC San Diego, where I worked with Prof. Hao Su. Prior to that, I graduated from the elite computer science program in Peking University.

I aim to build machine intelligence that can actively learn from interactions, with the ability of reasoning, planning, and generalization. My current research mainly focuses on environment understanding for decision making, which involves reinforcement learning, computer vision, and embodied agents.

I will start my Ph.D. at UC Berkeley in 2020 Fall.

Contact: fliu [at] eng.ucsd.edu. My CV can be found here.

Education

University of California San Diego

Sep. 2018 - Mar. 2020

M.S. in Computer Science

Peking University

Sep. 2014 - Jul. 2018

B.S. in Computer Science (Honored Degree)

Publications

SAPIEN: a SimulAted Part-based Interactive ENvironment
Fanbo Xiang, Yuzhe Qin, Kaichun Mo, Yikuan Xia, Hao Zhu, Fangchen Liu, Minghua Liu, Hanxiao Jiang, Yifu Yuan, Li Yi, He Wang, Angel Chang, Leonidas Guibas, Hao Su.
CVPR 2020 (oral) We constructed a PhysX-based simulation environment using PartNet-Mobility dataset. My work focused on reinforcement learning-based robot manipulation in the simulator. In this part, ResNet-18/PointNet++ is used to encode the object mobility given RGB-D/point cloud of the interactive scene, which serve as the state representation for the RL agents.
State Alignment-based Imitation Learning
Fangchen Liu, Zhan Ling, Tongzhou Mu, Hao Su
ICLR 2020
Paper / Code Coming Soon
We propose a robust state-based imitation learning method and unify global and local alignment constraints into a novel RL objective. It achieves the SOTA on traditional imitation learning tasks, and can also imitate policies from heterogeneous experts.

Mapping State Space using Landmarks for Universal Goal Reaching
Zhiao Huang*, Fangchen Liu*, Hao Su (* indicates equal contribution)
NeurIPS 2019
Learning a structured model and combining it with RL algorithms are essential for planning over long horizons. We propose a sample-based method to dynamically map the visited state space and demonstrate its advantage in several challenging RL tasks.

Adversarial Defense by Stratified Convolutional Sparse Coding
Bo Sun, Nian-Tsuan Tsai, Fangchen Liu, Ronald Yu, Hao Su
CVPR 2019 An attack-agnostic defense mechanism using convolutional dictionary learning.

Revisiting the Master-Slave Architecture in Multi-Agent Deep Reinforcement Learning
Xiangyu Kong, Fangchen Liu*, Bo Xin*, Yizhou Wang (* indicates equal contribution)
Hierarchical RL Workshop in NIPS 2017
NIPS-W 2017 (oral)
A hierarchical RL framework for multi-agent coordination.

BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning.
Fisher Yu, Haofeng Chen, Xin Wang, Wenqi Xian, Yingying Chen, Fangchen Liu, Vash Madhavan, Trevor Darrell
CVPR 2020 (oral)
An autonomous driving dataset with diverse scene, tasks, and a scalable annotation system.

Work Experience

  • Facebook AI Research
    Research Intern (to be), advised by Dr. Yuandong Tian Summer 2020
  • Microsoft Research Asia
    Research Intern, Visual Computing Group Sep. 2017 - Mar. 2018
  • SenseTime AI
    Research Intern, Face Detection and Recognition Group Sep. 2016 - Apr. 2017
  • Service

    Reviewer for CVPR 2020, NeurIPS 2020, ECCV 2020

    To strive, to seek, to find, and not to yield. -- Lord Tennyson, in Ulysses