Home
Publications
Group
alt text 

Peng Cheng
Partner Research Manager and Senior Director
Microsoft Research Asia - Vancouver


E-mail: pengc at microsoft dot com
[Google Scholar][GitHub]

I'm currently a partner research manager at Microsoft Research Asia (MSRA). I received my Ph.D. in Computer Science and Technology from Tsinghua University in 2015 and B.S. degrees in Software Engineering from Beihang University in 2010. I was as a visiting student in UCLA from September 2013 to September 2014. In 2015, I joined MSRA.

Please feel free to send me an email if you are interested in a full-time research position, internship, or visiting researcher role in MSRA Vancouver lab.

My research interests are in the broad areas of systems and networking, cloud and AI infrastructure, and AI. My recent work is mainly focusing on:

  • Artificial Specialized Intelligence

  • AI for Systems and Networking

  • AI for Cloud and AI Infrastructure

News

Selected Papers (Full List)

  • ICML’26 Pull Requests as a Training Signal for Repo-Level Code Editing

  • ICML’26 Beyond Sunk Costs: Boosting LLM Pre-training Efficiency via Orthogonal Growth of Mixture-of-Experts

  • SoCC’26 SlimeMold: Scaling Distributed Hardware Load Balancers via a Unified Giant Connection Table

  • ACL’26 Data Mixing Agent: Learning to Re-weight Domains for Continual Pre-training

  • SIGMOD’26 CoddSpeed: Hardware Accelerated Query Processing in Microsoft Fabric

  • MLSys’26 Virtual Machine NUMA Placement at Scale: Learning the Norm, Shielding the Tail

  • FSE’26 TSGuard: Automated User-Centric Incident Diagnosis for AI Workloads in the Cloud

  • NSDI’26 SmartNIC-Enabled Live Migration for Storage-Optimized VMs with PYROCUMULUS

  • ASPLOS’26 MSCCL++: Rethinking GPU Communication Abstractions for AI Inference

  • ICML’25 Optimizing Large Language Model Training Using FP4 Quantization

  • ICLR’25 Integrative Decoding: Improve Factuality via Implicit Self-consistency

  • ICLR’25 Automated Proof Generation for Rust Code via Self-Evolution

  • AAAI’25 Enhancing Large Language Model Performance with Gradient-Based Parameter Selection

  • MICRO’24 NeoMem: Hardware/Software Co-Design for CXL-Native Memory Tiering

  • ATC’24 SuperBench: Improving Cloud AI Infrastructure Reliability with Proactive Validation

  • SOSP’23 SPFresh: Incremental In-Place Update for Billion-Scale Vector Search

  • MLSys’23 Tutel: Adaptive Mixture-of-Experts at Scale

  • NSDI’23 ARK: GPU-driven Code Execution for Distributed Deep Learning

  • ASPLOS’23 ElasticFlow: An Elastic Serverless Training Platform for Distributed Deep Learning

  • CoNEXT’22 PipeDevice: A Hardware-Software Co-Design Approach to Intra-Host Container Communication

  • NeurIPS’22 An Adaptive Deep RL Method for Non-Stationary Environments with Piecewise Stable Context

  • ICWS’22 RuleCache: Accelerating Web Application Firewalls by On-line Learning Traffic Patterns

  • ATC’22 PilotFish: Harvesting Free Cycles of Cloud Gaming with Deep Learning Training

  • INFOCOM’21 NFD: Using Behavior Models to Develop Cross-Platform Network Functions

  • ATC’20 NetKernel: Making Network Stack Part of the Virtualized Infrastructure

  • NSDI’19 Direct Universal Access: Making Data Center Resources Available to FPGA

  • ICNP’18 Micro-Burst in Data Centers: Observations, Analysis, and Mitigations

  • NSDI’18 Multi-Path Transport for RDMA in Datacenters

  • CoNEXT’17 Tagger: Practical PFC Deadlock Prevention in Data Center Networks

  • SIGCOMM’16 ClickNP: Highly flexible and High-performance Network Processing with Reconfigurable Hardware

  • ATC’16 Fast and Cautious: Leveraging Multi-path Diversity for Transport Loss Recovery in Data Centers

  • EuroSys’16 TFC: Token Flow Control in Data Center Networks

  • NSDI’14 Catch the Whole Lot in an Action: Rapid Precise Packet Loss Notification in Data Center

Services