Fan Lai
Assistant Professor
Studies machine learning systems, including efficient LLM serving and federated learning infrastructure.
Fan Lai's lab at UIUC works on efficient training & inference, distributed, cloud & edge systems and trustworthy & safe ML. 26 papers since 2023, 19 at top venues, mostly NeurIPS and ICLR. 5 PhD students and 11 master's students. 14 alumni, most often to ByteDance and Google. Applications close .
At a glance
- papers, 2023–2026
- 7/ year
- at T1 venues
- 73%
- PhD students
- 5
- MS students
- 11
- alumni listed
- 14
Papers by yearfrom the lab homepage, 2023–2026
- 9 papers2026 · T1 7 · collecting
- 9 papers2025 · T1 6
- 2 papers2024 · T1 2
- 6 papers2023 · T1 4
T1 venuesOtherVenues: NeurIPS, ICLR, ICML, MLSys, NSDI
Research focuspaper titles that use each topic's words
- Efficient training & inference11 papers
- serving6 papers
- Distributed, cloud & edge systems4 papers
- efficient llm1 paper
- Trustworthy & safe ML0 papers
Paper list
From the lab homepage.
20269 papers
- VLDB (T1)Compass: SLO-aware Query Planner for Compound AI Serving at Scale
- ICML (T1)CORRECT: COndensed eRror RECognition via knowledge Transfer in multi-agent systems
- SIGCOMM (T1)EMA: Efficient Model Adaptation for Learning-based Systems
- COLMIntrospective Diffusion Language Models
- NSDI (T1)JITServe: SLO-aware LLM Serving with Imprecise Request Information
- ICDMMCP: From Context Compression to Mining for Efficient RAG-based LLM Serving
- ICLR (T1)OPPO: Accelerating PPO-based RLHF via Pipeline Overlap
- NeurIPS (T1)The Last Human-Written Paper: Agent-Native Research Artifacts
- ICML (T1)XRPO: Pushing the limits of GRPO with Targeted Exploration and Exploitation
20259 papers
- NeurIPS (T1)Act Only When It Pays: Efficient Reinforcement Learning for LLM Reasoning via Selective Rollouts
- ACL FindingDiSCo: Device-Server Collaborative LLM-Based Text Streaming Services
- IEEE Transactions on ComputersFIITED: Fine-Grained Embedding Dimension Optimization During Training for Recommender Systems
- NeurIPS (T1)HyGen: Efficient LLM Serving via Elastic Online-Offline Request Co-location
- SOSP (T1)IC-Cache: Efficient Large Language Model Serving via In-context Caching
- NeurIPS (T1)Inv-Entropy: A Fully Probabilistic Framework for Uncertainty Quantification in Language Models
- MICRO (T1)NetZIP: Algorithm/Hardware Co-design of In-network Lossless Compression for Distributed Large Model Training
- APSysTowards Fully Disaggregated Recommendation Model Serving
- MLSys (T1)Venn: Resource Management Across Federated Learning Jobs
20242 papers
- MLSys (T1)FedTrans: Efficient Federated Learning via Multi-Model Transformation
- NeurIPS (T1)Learn To be Efficient: Build Structured Sparsity in Large Language Models (opens in a new tab)
20236 papers
- OSDI (T1)AdaEmbed: Adaptive Embedding for Large-Scale Recommendation Models (opens in a new tab)
- SoCCAuxo: Heterogeneity-Mitigating Federated Learning via Scalable Client Clustering (opens in a new tab)
- ICLR (T1)Coverage-centric Coreset Selection for High Pruning Rates (opens in a new tab)
- EuroSys (T1)Egeria: Efficient DNN Training with Knowledge-Guided Layer Freezing (opens in a new tab)
- IEEE Transactions on ASEFed-ensemble: Ensemble Models in Federated Learning for Improved Generalization and Uncertainty Quantification (opens in a new tab)
- NSDI (T1)ModelKeeper: Accelerating DNN Training via Automated Training Warmup (opens in a new tab)
Where alumni went14 listed, 25% to industry
Industry 3Academia 9
Most often to
- ByteDance
- NVIDIA
Also to
- UIUC
- Carnegie Mellon University
- MIT
Apply to UIUCFall 2027
UIUC Siebel Computer Science PhD
- Round
- Single round
- Opens
- —
- GRE
- optional
- English
- TOEFL iBT / IELTS
- Fee
- USD 90
Not yet confirmed for this intake.
Similar labsnearest by research summary
- CHETH Zurich
Yannis Chronis
- Accelerators, chips & architecture
- SGNTU
Yonggang Wen
- Efficient training & inference
- CNPeking
Zhen Xiao
- Distributed, cloud & edge systems
- USUIUC
Daniel Kang
- Efficient training & inference
- CAToronto
Baochun Li
- Distributed, cloud & edge systems
- HKCUHK
Bei Yu
- Accelerators, chips & architecture
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.
UIUCcloses