Minjia Zhang
Assistant Professor
Studies efficient machine learning systems, including model compression and training and inference on parallel hardware.
Minjia Zhang's lab at UIUC works on efficient training & inference, accelerators, chips & architecture and data management & big data. 52 papers since 2023, 32 at top venues, mostly ICLR and EMNLP. 9 PhD students and 6 master's students. 10 alumni, most often to ByteDance and Adobe. Applications close .
At a glance
- papers, 2023–2026
- 14/ year
- at T1 venues
- 62%
- PhD students
- 9
- MS students
- 6
- alumni listed
- 10
Papers by yearfrom the lab homepage, 2023–2026
- 21 papers2026 · T1 13 · collecting
- 16 papers2025 · T1 9
- 7 papers2024 · T1 6
- 8 papers2023 · T1 4
T1 venuesOtherVenues: ICLR, EMNLP, Findings of ACL, SIGMOD, AAAI
Research focuspaper titles that use each topic's words
- Efficient training & inference26 papers
- Data management & big data6 papers
- Accelerators, chips & architecture0 papers
- efficient machine0 papers
- model compression0 papers
Paper list
From the lab homepage.
202621 papers
- ACL (T1)A Goal Without a Plan Is Just a Wish: Efficient and Effective Global Planner Training for Long-Horizon Agent Task
- CoLMA Self-Pruning Transformer: Extreme KV-Cache Compression with Universal Attention
- ICLR (T1)AutoSP: Unlocking Long-Context LLM Training Via Compiler-Based Sequence Parallelism
- Findings of ACLFaithLens: Detecting and Explaining Faithfulness Hallucination
- ICLR (T1)From Collapse to Control: Understanding and Extending Context Length in Emerging Hybrid Models via Universal Position Interpolation
- CVPR FindingsGated Differential Linear Attention: A Linear-Time Decoder for High-Fidelity Medical Segmentation
- Findings of ACLHidden States as Early Signals: Step-level Trace Evaluation and Pruning for Efficient Test-Time Scaling
- ISCMegaFold: Efficient Training of Next-Generation 3D-Attention Protein Models on Cross-Platform GPUs
- Findings of ACLMENTOR: Efficient Autoregressive Image Generation with Balanced Multimodal Control
- EMNLP (T1)MMGR: Multi-Modal Generative Reasoning Benchmark and Evaluation
- ICML (T1)PuzzleMoE: Efficient Compression of Large Mixture-of-Experts Models via Sparse Expert Merging and Bit-packed inference
- EMNLP (T1)Quantization Inflates Reasoning: Token Inflation as a Hidden Cost of Low-Bit Reasoning Models
202516 papers
- HPCA (T1)Buffalo: Enabling Large-Scale GNN Training via Memory-Efficient Bucketization
- EMNLP (T1)Cache-of-Thought: Master-Apprentice Framework for Cost-Effective Vision Language Model Inference
- IEEE BigDataDegree-Based Scheduling and Memory Management for Large-Scale Exact Online GNN Inference
- ACM MMFlexGaussian: Flexible and Cost-Effective Training-Free Compression for 3D Gaussian Splatting
- PSBImproving Retrieval-Augmented Generation in Medicine with Iterative Follow-up Questions
- ICCV (T1)InstantEdit: Text-Guided Few-Step Image Editing with Piecewise Rectified Flow
- ICSE (T1)Large Language Models as Configuration Validators
- EMNLP (T1)Looking Beyond Text: Reducing Language bias in Large Vision-Language Models via Multimodal Dual-Attention and Soft-Image Guidance
- ACL FindingMedCite: Can Language Models Generate Verifiable Text for Medicine?
- MLSys (T1)MiLo: Efficient Quantized MoE Inference with Mixture of Low-Rank Compensators
- ACL FindingMiniKV: Pushing the Limits of 2-Bit KV Cache via Compression and System Co-Design for Efficient Long Context Inference
- MICRO (T1)NetZIP: Algorithm/Hardware Co-design of In-network Lossless Compression for Distributed Large Model Training
20247 papers
- AAAI (T1)DeepSpeed Data Efficiency: Improving Deep Learning Model Quality and Training Efficiency via Efficient Data Sampling and Routing
- PODCDeepSpeed-Ulysses: System Optimizations for Enabling Training of Extreme Long Sequence Transformer Models
- ICLR (T1)Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs
- Nature Methods (T1)OpenFold: Retraining AlphaFold2 yields new insights into its learning mechanisms and capacity for generalization
- NSDI (T1)Parcae: Proactive, Liveput-Optimized DNN Training on Preemptible Instances
- NeurIPS (T1)UltraEdit: Instruction-based Fine-Grained Image Editing at Scale
- SIGMOD (T1)Vexless: A Serverless Vector Data Management System Using Cloud Functions
20238 papers
- NSDI (T1)Bamboo: Making Preemptible Instances Resilient for Affordable Training of Large DNNs
- ASPLOS (T1)Betty: Enabling Large-Scale GNN Training with Batch-Level Graph Partitioning
- MobiCom (T1)Cost-effective On-device Continual Learning over Memory Hierarchy with Miro
- NeurIPS AI4Science WorkshopDeepSpeed4Science Initiative: Enabling Large-Scale Scientific Discovery through Sophisticated AI System Technologies
- IEEE Data Engineering BulletinExploiting Modern Hardware Architectures for High-Dimensional Vector Search at Speed and Scale
- PPoPPiQAN: Fast and Accurate Vector Search with Efficient Intra-Query Parallelism on Multi-Core Architectures
- ICLR (T1)Maximizing Communication Efficiency for Large-scale Training via 0/1 Adam
- ECAIRevisiting the Efficiency-Accuracy Tradeoff in Adapting Transformer Models via Adversarial Fine-Tuning
Where alumni went10 listed, 88% to industry
Industry 7Academia 1
Most often to
- ByteDance
- Adobe
- Amazon
Also to
- Carnegie Mellon University
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
- USUIUC
Naresh R Shanbhag
- Accelerators, chips & architecture
- CHEPFL
David Atienza Alonso
- Accelerators, chips & architecture
- CHEPFL
Mahsa Shoaran
- Accelerators, chips & architecture
- USUIUC
Nishil Talati
- Accelerators, chips & architecture
- CNZhejiang
De Ma
- Accelerators, chips & architecture
- USUT Austin
Aditya Akella
- 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