My research interests lie in the intersection of Machine Learning (ML), Formal Methods (Verification and Synthesis) and Robotics.
[2021/06/25] New preprint of “POLAR: A Polynomial Arithmetic Framework for Verifying Neural-Network Controlled Systems” is available. [code]
[2021/05/21] Honored to receive second place award at ACM SIGBED Student Research Competition 2021!
[2021/03/01] New preprint of “DRIBO: Robust Deep Reinforcement Learning via Multi-View Information Bottleneck” is available. [code]
[2020/12/02] Paper for efficiently training verifiably robust deep neural networks: “Adversarial Training and Provable Robustness: A Tale of Two objectives” is accepted to AAAI'21.
[2020/08/13] Arxiv draft of “Adversarial Training and Provable Robustness: A Tale of Two objectives” is available.
[2020/07/07] Our paper “Divide and Slide: Layer-Wise Refinement for Output Range Analysis of Deep Neural Networks” has been accepted to EMSOFT'20.
[2020/06/23] Our new tool paper "ReachNN*: A Tool for Reachability Analysis ofNeural-Network Controlled Systems" will appear at ATVA'20.
[2019/08/9] Our new paper "Towards Verification-Aware Knowledge Distillation for Neural-Network Controlled Systems" will appear at ICCAD'19.
[2019/07/10] Our paper “ReachNN: Reachability Analysis of Neural-Network Controlled Systems” has been accepted to appear at EMSOFT’19.
[2019/03/20] Our Paper "Safety-Guided Deep Reinforcement Learning via Online Gaussian Process Estimation" has been accepted at ICLR’19 Workshop on SafeML.
Jiameng Fan and Wenchao Li
Adversarial Training and Provable Robustness: A Tale of Two Objectives
AAAI Conference on Artificial Intelligence (AAAI), February 2021
A joint training scheme to efficiently train verifiably robust deep neural networks that achieves state-of-art verified robustness in many settings.
Jiameng Fan, Chao Huang, Wenchao Li, Xin Chen and Qi Zhu
Towards Verification-Aware Knowledge Distillation for Neural-Network Controlled Systems
International Conference on Computer-Aided Design (ICCAD), November 2019
A novel verification-aware knowledge distillation framework that transfers the knowledge of a trained network to a new and easier-to-verify network.
Chao Huang, Jiameng Fan, Wenchao Li, Xin Chen and Qi Zhu
ReachNN: Reachability Analysis of Neural-Network Controlled Systems
International Conference on Embedded Software (EMSOFT), October 2019
A new reachability analysis approach based on Bernstein polynomials that can verify neural-network controlled systems with a more general form of activation functions.
Jiameng Fan and Wenchao Li
Safety-Guided Deep Reinforcement Learning via Online Gaussian Process Estimation
International Conference on Learning Representation (ICLR), Workshop on Safe Machine Learning: Specification, Robustness, and Assurance, May 2019
Formulate the state-action value function of a notion of trajectory-based safety as a candidate Lyapunov function and extend control-theoretic results to approximate its derivative using online Gaussian Process estimation.