Zexin Li

Zexin Li

Ph.D. Student of ECE

The University of California, Riverside

Biography

I am a Ph.D. student at the University of California, Riverside (UCR). I am fortunate to be advised by Dr. Cong Liu and working with Dr. Yinglun Zhu . I received a bachelor’s degree from the Southern University of Science and Technology (SUSTech) under the advice of Dr. Yuqun Zhang in July 2020. My research interests lie in interdisciplinary fields of real-time embedded systems and on-device machine learning.

I am actively looking for cooperation in the following topics: (1) deploying machine learning models on real-time embedded devices, (2) system-application co-optimization of machine learning systems, and (3) improving performance robustness in machine learning systems.

Feel free to contact me if we share common research interests.

Download my Curriculum Vitae .

Interests
  • Real-time Embedded System
  • On-device Machine Learning
  • Efficient Machine Learning
Education

  • Ph.D. Candidate, 2022 - Now

    The University of California, Riverside

  • Ph.D. Candidate, 2020 - 2022

    The University of Texas at Dallas

  • B.S., 2016 - 2020

    Southern University of Science and Technology

News

04/2024 - One paper is accepted by FSE'24 .
03/2024 - Pass the qualification exam of Ph.D. at UCR .
07/2023 - Three papers are accepted by RTSS'23 .
06/2023 - One paper is accepted by IROS'23 .
05/2023 - Two papers are accepted by ACL'23 .
02/2023 - One paper is accepted by CVPR'23 .
09/2022 - Honored to receive the Dean’s Distinguished Fellowship, UCR.
09/2022 - Join UCRiverside and Dr. Cong Liu ’s Autonomous Embedded Systems Lab in 2022 Fall.
01/2022 - Pass the qualification exam of Ph.D. at UTDallas .
01/2022 - Join UTDallas and Dr. Cong Liu ’s Real-Time System Lab in 2022 Spring.
05/2021 - Begin the research intern in Tecent Youtu Lab , Shanghai, China.
04/2021 - Admitted into Tencent Rhino Bird Elite Talent Training Program .
03/2021 - One paper is accepted by JSA , great thanks to Dr. Yuqun Zhang .
10/2019 - Begin the research intern in Kuaishou , Shenzhen, China.

Publications

Simin Chen, Zexin Li, Wei Yang, Cong Liu (2024). DeciX: Explain Deep Learning Based Code Generation Applications. In FSE'24.

Cite

Zexin Li, Soroush Bateni, Cong Liu (2024). Genie: Smart ROS-based Caching for Connected Autonomous Robots. In Arxiv.

PDF Cite

Yufei Li, Zexin Li, Wei Yang, Cong Liu (2023). RT-LM: Uncertainty-Aware Resource Management for Real-Time On-Device Language Models. In RTSS'23.

PDF Cite

Zexin Li, Aritra Samanta, Yufei Li, Andrea Soltoggio, Hyoseung Kim, Cong Liu (2023). R^3: On-device Real-Time Deep Reinforcement Learning for Autonomous Robotics. In RTSS'23.

PDF Cite

Zexin Li, Tao Ren, Xiaoxi He, Cong Liu (2023). RED: A Systematic Real-Time Scheduling Approach for Robotic Environmental Dynamics. In RTSS'23.

PDF Cite

Zexin Li, Xiaoxi He, Yufei Li, Wei Yang, Lothar Thiele, Cong Liu (2023). MIMONet: Multi-Input Multi-Output On-Device Deep Learning. In Arxiv.

PDF Cite

Shahab Nikkhoo, Zexin Li, Aritra Samanta, Yufei Li, Cong Liu (2023). PIMbot: Policy and Incentive Manipulation for Multi-Robot Reinforcement Learning in Social Dilemmas. In IROS'23.

PDF Cite Video

Yiming Chen, Simin Chen, Zexin Li, Wei Yang, Cong Liu, Robby Tan, Haizhou Li (2023). Dynamic Transformers Provide a False Sense of Efficiency. In ACL'23.

PDF Cite Code

Yufei Li, Zexin Li, Yingfan Gao, Cong Liu (2023). White-Box Multi-Objective Adversarial Attack on Dialogue Generation. In ACL'23.

PDF Cite Code

Zexin Li*, Bangjie Yin, Taiping Yao, Junfeng Guo, Shouhong Ding, Simin Chen, Cong Liu (2023). Sibling-Attack: Rethinking Transferable Adversarial Attacks against Face Recognition. In CVPR'23.

PDF Cite Code Video DOI

Zexin Li, Yuqun Zhang*, Ao Ding, Husheng Zhou, Cong Liu (2021). Efficient algorithms for task mapping on heterogeneous CPU/GPU platforms for fast completion time. In JSA.

PDF Cite Code DOI

Projects

*
All Deep Learning System Robotics
R^3: On-device Real-Time Deep Reinforcement Learning for Autonomous Robotics
This project presents R^3, a holistic solution for managing timing, memory, and algorithm performance in on-device real-time DRL training. R^3 employs (i) a deadline-driven feedback loop with dynamic batch sizing for optimizing timing, (ii) efficient memory management to reduce memory footprint and allow larger replay buffer sizes, and (iii) a runtime coordinator guided by heuristic analysis and a runtime profiler for dynamically adjusting memory resource reservations.
R^3: On-device Real-Time Deep Reinforcement Learning for Autonomous Robotics
RED: A Systematic Real-Time Scheduling Approach for Robotic Environmental Dynamics
In this project, we introduce RED, a systematic real-time scheduling approach designed to support multi-task deep neural network workloads in resource-limited robotic systems. It is designed to adaptively manage the Robotic Environmental Dynamics (RED) while adhering to real-time constraints.
RED: A Systematic Real-Time Scheduling Approach for Robotic Environmental Dynamics
PIMbot: Policy and Incentive Manipulation for Multi-Robot Reinforcement Learning in Social Dilemmas
We present PIMbot, an approach to manipulate the reward function in multi-robot collaboration through two distinct forms: policy and incentive manipulation. PIMBot introduces a new angle for manipulating multi-agent RL social dilemmas through a unique reward function for incentivization.
PIMbot: Policy and Incentive Manipulation for Multi-Robot Reinforcement Learning in Social Dilemmas
MIMONet: Multi-Input Multi-Output On-Device Deep Learning
This projects aims to generate a neural network compression techniques for on-device deep learning, and applying to real-world robots. We consider a two-phase deep compression approach consist of reducing both intra-model and inter-model redundancy.
MIMONet: Multi-Input Multi-Output On-Device Deep Learning
Sibling-Attack: Rethinking Transferable Adversarial Attacks against Face Recognition
The core objective of this project is to enhance attacking transferability on face recogntion systems. This approach is able to significantly improve the attacking success rate on several publicly available datasets, and break some state-of-the-art commercial platforms corresponding to face recognition models.
Sibling-Attack: Rethinking Transferable Adversarial Attacks against Face Recognition
Efficient algorithms for task mapping on heterogeneous CPU/GPU platforms for fast completion time
This projects aims to design a fine-grain mapping framework that explores a set of critical factors is needed for heterogeneous embedded systems. The source code is published at https://github.com/asteriaaaa/smartCoopScheduler.
Efficient algorithms for task mapping on heterogeneous CPU/GPU platforms for fast completion time
Robot Path Planning
This projects aims to generate proper path for autonomatic driving. Since path can not be simplified as skelton in this program, we need to do 2D approximately convex decomposition (ACD). After applying ACD, we use Self-organizing Map (SOM) alogrithm to generate path for Multiple Travelling salesman (MTSP) problem.
Robot Path Planning

Contact