Xuanzhe Liu
Professor
Researches system software and distributed systems, including efficient deep learning and large-model serving on mobile and cloud platforms.
Xuanzhe Liu's lab at Peking works on distributed, cloud & edge systems and efficient training & inference. 54 papers since 2023, 24 at top venues, mostly NSDI and SIGCOMM. 6 PhD students. 18 alumni. Applications close .
At a glance
- papers, 2023–2026
- 14/ year
- at T1 venues
- 44%
- PhD students
- 6
- MS students
- 0
- alumni listed
- 18
Papers by yearfrom the lab homepage, 2023–2026
- 10 papers2026 · T1 6 · collecting
- 9 papers2025 · T1 4
- 21 papers2024 · T1 6
- 14 papers2023 · T1 8
T1 venuesOtherVenues: NSDI, SIGCOMM, SOSP, OSDI, EuroSys
Research focuspaper titles that use each topic's words
- Efficient training & inference23 papers
- Distributed, cloud & edge systems10 papers
- system software0 papers
- platforms0 papers
- efficient deep0 papers
Paper list
From the lab homepage.
202610 papers
- ACM Transactions on Software Engineering and Methodology. Accepted to appearA First Look at Bugs in LLM Inference Engines
- Proceedings of the ACM International Conference on Mobile Systems, Applications, and Services (MobiSys), June 21-25, , Cambridge, UK.A Sparse Attention Module for NPU-Centric On-Device LLM Inference
- NSDI (T1)DistRS: Disaggregated Reward Service for RLVR with Batch-level Constraint
- NSDI (T1)FastServe: Iteration-Level Preemptive Scheduling for Large Language Model Inference
- NSDI (T1)HydraServe: Minimizing Cold Start Latency for Serverless LLM Serving in Public Clouds
- NSDI (T1)KUBEDIRECT: Unleashing the Full Power of the Cluster Manager for Serverless Computing
- Proceedings of the International Conference on Embedded Artificial Intelligence and Sensing Systems (SenSys ), May 11-14, , Saint-Malo, France.LLM as a System Service on Mobile Devices
- EuroSys (T1)MegaScale-MoE: Large-Scale Communication-Efficient Training of Mixture-of-Experts Models in Production
- ACM Transactions on Software Engineering and Methodology. Accepted to appearPromptware Engineering: Software Engineering for Prompt-Enabled Systems
- EuroSys (T1)Serverless Replication of Object Storage across Multi-Vendor Clouds and Regions
20259 papers
- SOSP (T1)Aegaeon: Effective GPU Pooling for Concurrent LLM Serving on the Market
- NSDI (T1)Automatic End-to-End Testing for Service Mesh Traffic Management
- ACM Transactions on Software Engineering and Methodology. Accepted to appearDrWASI: LLM-assisted Differential Testing for WebAssembly System Interface Implementations
- MobiCom (T1)Elastic On-Device LLM Service
- Proceedings of the ACM International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOSFast On-device LLM Inference with NPUs
- SIGCOMM (T1)MegaScale-Infer: Efficient Mixture-of-Experts Model Serving with Disaggregated Expert Parallelism
- ACM Transactions on Computer Systems.RAGCache: Efficient Knowledge Caching for Retrieval-Augmented Generation
- ACM Transactions on Software Engineering and Methodology. Accepted to appearResearch on WebAssembly Runtimes: A Survey
- ACM Transactions on Software Engineering and Methodology. Accepted to appearSCOPE: Performance Testing for Serverless Computing
202421 papers
- IEEE Transactions on Mobile Computing.A Comprehensive Deep Learning Library Benchmark and Optimal Library Selection
- ACM Transactions on Software Engineering and Methodology. Accepted to appear.Anatomizing Deep Learning Inference in Web Browsers
- ACM Transactions on Software Engineering and Methodology. Accepted to appearBias Behind the Wheel: Fairness Testing of Autonomous Driving Systems
- OSDI (T1)Burstable Cloud Block Storage with Data Processing Units
- ACM/IEEE Transactions on NetworkingDistMind: Efficient Resource Disaggregation for Deep Learning Workloads
- OSDI (T1)DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving
- OSDI (T1)dLoRA: Dynamically Orchestrating Requests and Adapters for LoRA LLM Serving
- ACM Transactions on Computer Systems.Efficient Fault Tolerance for Stateful Serverless Computing with Asymmetric Logging
- IEEE Transactions on Mobile Computing. Volume: 23, IssueEfficient, Scalable, and Sustainable DNN Training on SoC-Clustered Edge Servers
- NSDI (T1)Fast Vector Query Processing for Large Datasets Beyond GPU Memory with Reordered Pipelining
- IEEE Transactions on Mobile Computing.FLASH: Heterogeneity-Aware Federated Learning at Scale
- IEEE Transactions on Parallel and Distributed Systems. Accepted to appear.GreenFlow: A Carbon-Efficient Scheduler for Deep Learning Workloads
202314 papers
- ACM Transactions on Software Engineering and MethodologyAdonis: Practical and Efficient Control Flow Recovery through OS-level Traces
- SOSP (T1)Automated Verification of an In-Production DNS Authoritative Engine
- ACM Transactions on Software Engineering and Methodology. Accepted to appearCharacterizing and Detecting WebAssembly Runtime Bugs
- Proceedings of ACM International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS , March 25-29, , Vancouver, Canada.Disaggegated RAID Storage in Modern Datacenters
- SIGCOMM (T1)Ditto: Efficient Serverless Analytics with Elastic Parallelism
- Proceedings of ACM International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOSElasticFlow: An Elastic Serverless Training Platform for Distributed Deep Learning
- ACM Transactions on Software Engineering and Methodology.FaaSLight: General Application-level Cold-start Latency Optimization for Function-as-a-Service in Serverless Computing
- NSDI (T1)Fast, Approximate Vector Queries on Very Large Unstructured Datasets
- SOSP (T1)Halfmoon: Log-Optimal Fault-Tolerant Stateful Serverless Computing
- SIGCOMM (T1)Klotski: Efficient and Safe Network Migration of Large Production Datacenters
- NSDI (T1)POLYCORN: Data-driven Cross-layer Multipath Networking for High-speed Railway through Composable Schedulerlets
- ACM Transactions on Software Engineering and Methodology.Rise of Distributed Deep Learning Training in the Big Model Era: A Software Engineering Perspective
Where alumni went18 listed
Apply to PekingFall 2027
Computer Science PhD (international applicants)
- Round
- Main
- Opens
- —
- GRE
- See program page
- English
- See program page
- Fee
- See program page
Not yet confirmed for this intake.
Similar labsnearest by research summary
- CNSJTU
Han Zhao
- Efficient training & inference
- HKCUHK
He Chen
- Distributed, cloud & edge systems
- CNSJTU
Qinya Li
- Distributed, cloud & edge systems
- HKCUHK
Zhenyu Yan
- Distributed, cloud & edge systems
- CHETH Zurich
Ana Klimovic
- Efficient training & inference
- SGNTU
Chen Wang
- Distributed, cloud & edge systems
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
Pekingcloses