News

• 📢 2024/06 – In collaboration with Huawei, we published the white paper on AI Phone (in Chinese).
• 📢 2024/06 – Successfully hosted the 1st workshop on edge foundation models (EdgeFM 2024) in Tokyo! See you next year!
• 📢 2024/05 – Paper “SwapMoE: Serving Off-the-shelf MoE-based Large Language Models with Tunable Memory Budget” accepted to ACL 2024 (main track). Congrats to Rui Kong and others!
• 📢 2024/03 – Our group won the First Prize in AIR Winter Camp 2024. Congrats to winning team members Shanhui Zhao, Hao Wen, Wenjie Du and Cheng Liang!
• 📢 2024/03 – One paper got accepted to ACM MobiSys 2024: “Empowering In-Browser Deep Learning Inference on Edge Through Just-In-Time Kernel Optimization”. Congrats to all collaborators!
• 📢 2024/02 – We are hosting a workshop on edge foundation models (EdgeFM 2024) in Tokyo, Japan, Co-located with MobiSys 2024. Please consider submitting your work and joining us in person!
• 📢 2024/01 – Our position & survey paper on mobile LLM agents “Personal LLM Agents: Insights and Survey about the Capability, Efficiency and Security” was released. Feeling excited to work in this new area! [arXiv], [GitHub], [机器之心报道]
• 📢 2023/12 – Our paper “Retrieval-based Battery Degradation Prediction for Battery Energy Storage System Operations” got a Best Paper Award in IEEE GreenCom 2023. Thanks and congratulations to our awesome intern Yixuan Li and partners at GDS!
• 📢 2023/11 – Two papers “AutoDroid: LLM-powered Task Automation in Android”, “FlexNN: Efficient and Adaptive DNN Inference on Memory-Constrained Edge Devices” are conditionally accepted to MobiCom 2024! The former is our first work on LLM-powered mobile agent. Kudos to Hao Wen, Xiangyu Li, and the team!
• 📢 2023/09 – Our group won the Second Prize in AIR Summer Camp 2023. Congrats to winners Guohong Liu, Shanhui Zhao and Tao Yu!
• 📢 2023/07 – Paper “PatchBackdoor: Backdoor Attack against Deep Neural Networks without Model Modification” is accepted to ACM MM‘23! This is the first backdoor attack against DNN models without model modification. Congratulations to Yizhen Yuan and Rui Kong!
• 📢 2023/07 – Paper “HiMoDepth: Efficient Training-free High-resolution On-device Depth Perception” is accepted to the IEEE Transactions on Mobile Computing (TMC), Paper “Testing and Enhancing the Robustness of Federated Learning System against Realistic Data Corruption” and “ReSPlay: Improving Cross-Platform Record-and-Replay with GUI Sequence Matching” are accepted to ISSRE 2023. Congratulations to Jinrui Zhang, Chen Yang, Shaokun Zhang, and coauthors!
• 📢 2023/03 – Our paper “ConvReLU++: Reference-based Lossless Acceleration of Conv-ReLU Operations on Mobile CPU” is conditionally accepted to MobiSys 2023! Congratulations to Rui Kong and Yizhen Yuan for their first top-conference paper!
• 📢 2023/02 – Our paper “PatchCensor: Patch Robustness Certification for Transformers via Exhaustive Testing” is accepted to TOSEM! Thank Yuheng Huang for the hard work and insightful discussions!
• 📢 2022/11 – Our paper “AdaptiveNet: Post-deployment Neural Architecture Adaptation for Diverse Edge Environments” is accepted to MobiCom 2023! Congratulations to Wen Hao for his first (top-conference) paper. Much thanks to our collaborators @Zunshuai Zhang, @Shiqi Jiang, @Yunxin Liu and industry partners!
• 📢 2022/10 – My grant application on “Neurosymbolic Programming Approaches to Reliable Edge AI” is approved and funded by NSFC. Very excited to start this cool direction.
• 📢 2022/06 – Our paper “MobiDepth: Real-Time Depth Estimation Using On-Device Dual Cameras” is accepted to MobiCom 2022! Congratulations to Jinrui Zhang and other great coauthors.
• 📢 2022/05 – Our paper “FedBalancer: Data and Pace Control for Efficient Federated Learning on Heterogeneous Clients” got accepted to MobiSys 2022! We propose to accelerate FL training through fine-grained client data selection and deadline control. Thanks to awesome collaborators Jaemin and Prof. Sung-Ju Lee at KAIST.
• 📢 2022/03 – Our paper “WhyGen: Explaining ML-powered Code Generation by Referring to Training Examples” is accepted to the Demo track of ICSE 2022. Congratulations to Weixiang for his first publication! This paper introduces an interesting idea to address the widely-concerned recitation problem in DL-based code generators.
• 📢 2022/03 – Our paper “Representational Continuity for Unsupervised Continual Learning” got accepted to ICLR 2022 (Oral)!
• 📢 2021/12 – Our paper “ReMoS: Reducing Defect Inheritance in Transfer Learning via Relevant Model Slicing” is accepted to ICSE 2022. It discusses the defect inheritance problem during transfer learning. We believe the security of pretrained models is an important problem since pretraining is increaingly popular.
• 📢 2021/11 – I recently left MSRA and joined the Institute for AI Industry Research (AIR) at Tsinghua University. Many thanks to my mentors and colleagues at MSRA.