Education

Purdue University, West Lafayette, Indiana, USA. 2013/08 - 2018/05

Ph.D. in Electrical and Computer Engineering. GPA 3.98/4.
Advisors:Zhongming Liu, Eugenio Culurciello, Charles A. Bouman, Gregory Francis and Stanley H. Chan
Thesis: Neural Encoding and Decoding with Deep Learning for Natural Vision.

Zhejiang University, Hangzhou, China.2009/08 - 2013/06

B.S. in Information Science and Electronic Engineering. GPA 3.90/4.
Advisor: Dr. Haoji Hu
Thesis: Blind Watermarking Scheme of 3D Triangular Meshes Resilient to Cropping Attacks.

Work Experience

Senior deep learning research and development engineer. Nvidia Corporation.2018/08 - present

Apply state-of-the-art deep learning research and advanced computing hardware to create the next generation of autonomous vehicles.
Develop perception models and algorithms for detecting objects, predicting driving paths, planning, etc.
Build and maintain the AI infrastructure for research and development engineering.
Active learning for sampling diverse and effective data to improve models over time.
Reseaerch on world model and learn driving policy from human-drive data using reinforcement learning for L5 autonomy.
Research on self-supervised learning to reduce the labeling cost and improve the robustness and generalization of DNNs.

Research assistant. Purdue University. 2013/08 - 2018/05

Led a group that aimed to develop brain-inspired deep-learning models not only for artificial intelligence but also for studying the brain.
Mapped human brain networks using functional MRI and electrophysiology.

Neural encoding and decoding with deep learning for natural vision

Developed new encoding models to predict brain activities given visual stimuli by using deep neural networks (e.g. CNN, RNN, VAE).
Developed a new decoding scheme that directly reconstructed and categorized the visual stimuli for mind-reading.
Proposed an efficient learning method for transferring deep-learning-based encoding models across human brains, and developed artificial brains that can simulate brain activities by using deep neural networks.
Provided systematic evaluation of the similarity between the deep neural networks and the human brain.

Brain-inspired artificial neural networks

Developed a bidirectional and recurrent neural network, as inspired by a neuroscience theory called predictive coding. The network supports recurrent computation via feedforward, feedback, and recurrent connections.
Demonstrated that the brain-inspired network enabled a shallow network to perform similarly or better than very deep feedforward-only neural network in object recognition.

Multimodal neuroimaging on mapping functional networks in the brain

Developed an algorithm to separate oscillatory and fractal components of neural signals and it has been an increasingly used tool for neural signal processing.
Mapped the organization of functional networks in the brain (using ICA, k-means, Restricted Boltzmann Machine, sparse coding).