Research Overview
I am generally interested in embodied AI, which enables learning through
interaction with environments, and mainly focus on reinforcement learning (RL).
Specifically, I am keen on:
- model-free RL algorithms,
- multi-agent and multi-task deep RL, and
- RL scalability, reproducibility, and replicability.
Besides the above mentioned, my curiosities naturally extend to game theory, neuroscience, epistemology, and so on. I believe that the RL community will evolve together with these fields.
Toward the overall goal of AI to create intelligent machines, I have studied,
researched, and done work in RL, computer vision, and the intersection of
machine
learning, engineering, and physics. Currently, I am a research intern in deep RL
group at ByteDance AI
Lab. Our group focuses on game AI and I research how
advanced deep RL algorithms can be employed in multi-agent games to tackle complex tasks
with partial observability.
During Summer 2020, I
was
glad to collaborate with Jim Fan
from Stanford Vision and
Learning
Lab. We worked on visual control and distributed RL with particular
focuses on sim-to-sim and sim-to-real transfer. Before that, I worked as an
undergraduate
researcher supervised by Dr. Chang Liu in Agile
Tomography Group at IDCOM. We developed line-of-sight laser
imaging systems empowered by deep learning.
Selected Projects
CSTNet: A Dual-Branch Convolutional Network for Imaging of Reactive Flows using
Chemical Species Tomography
Yunfan Jiang, Jingjing Si, Rui
Zhang,
Godwin
Enemali, Bin
Zhou,
Hugh McCann, Chang
Liu
Submitted to IEEE Transactions on Neural Networks and Learning Systems
arxiv / bibtex
This paper subsumes my bachelor thesis.
This is the first time, to the best of our knowledge, that a deep learning-based
algorithm for Chemical Species Tomography (CST) has been experimentally
validated for simultaneous imaging of
multiple critical parameters in reactive flows using a low-complexity optical
sensor with severely limited number of laser beams.

Wavelength Modulation Spectroscopy-Based Tomography Image Reconstruction
using
Deep Learning
Undergraduate thesis
(Ewart Farvis
Project Prize)
thesis / bibtex
This thesis is subsumed by our CSTNet paper.
To the best of my knowledge, it is the first time to verify and demonstrate the
feasibility and effectiveness of applying deep learning-based laser tomographic
methods on
experimental imaging of flame temperature distributions.
Industrial Experience
ByteDance Ltd., Beijing,
China
September 2020 - Present
Research intern in Deep Reinforcement Learning group.
Publications & Submitted Manuscripts
Journal Articles
Credit
The earliest design followed that of Jon Barron's
website. His website is awesome!