Matei Zaharia
Associate Professor
Builds data and AI systems, including Apache Spark, MLflow and ColBERT, and LLM programming frameworks like DSPy.
Matei Zaharia's lab at UC Berkeley works on efficient training & inference. 67 papers since 2023, 32 at top venues, mostly ICLR and VLDB. 16 alumni. Applications close .
At a glance
- papers, 2023–2026
- 17/ year
- at T1 venues
- 48%
- PhD students
- 0
- MS students
- 0
- alumni listed
- 16
Papers by yearfrom the lab homepage, 2023–2026
- 13 papers2026 · T1 6 · collecting
- 28 papers2025 · T1 13
- 14 papers2024 · T1 6
- 12 papers2023 · T1 7
T1 venuesOtherVenues: ICLR, VLDB, SIGMOD, Findings of EMNLP, MLSys
Research focuspaper titles that use each topic's words
- Efficient training & inference6 papers
- llm3 papers
- frameworks0 papers
- builds data0 papers
- programming0 papers
Paper list
From the lab homepage.
202613 papers
- SIGIR (T1)Can QPP Choose the Right Query Variant? Evaluating Query Variant Selection for RAG Pipelines (opens in a new tab)
- ACM CAIS .Composing Policy Gradients and Prompt Optimization for Language Model Programs (opens in a new tab)
- ICLR (T1)EXP-Bench: Can AI Conduct AI Research Experiments? (opens in a new tab)
- ICLR . Oral.GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning (opens in a new tab)
- ICLR WorkshopGRAID: Enhancing Spatial Reasoning of VLMs through High-Fidelity Data Generation (opens in a new tab)
- MLSys . Best Paper Award.LEANN: A Low-Storage Vector Index (opens in a new tab)
- ICML . Oral.Measuring Agents in Production (opens in a new tab)
- ACM CAIS .optimize_anything: Unified Text Optimization can Outperform Specialized Systems (opens in a new tab)
- ICLR (T1)Reliable Fine-Grained Evaluation of Natural Language Math Proofs (opens in a new tab)
- CIDR .Supporting Our AI Overlords: Redesigning Data Systems to be Agent-First (opens in a new tab)
- ICLR (T1)vAttention: Verified Sparse Attention via Sampling (opens in a new tab)
- ICLR (T1)vCache: Verified Semantic Prompt Caching (opens in a new tab)
202528 papers
- SIGMOD (T1)Blink Twice - Automatic Workload Pinning and Regression Detection for Versionless Apache Spark using Retries (opens in a new tab)
- ECIRColBERT-Serve: Efficient Multi-stage Memory-Mapped Scoring (opens in a new tab)
- OSDI (T1)Compass: Encrypted Semantic Search with High Accuracy (opens in a new tab)
- Journal of Data-centric Machine Learning Research, .Data Acquisition: A New Frontier in Data-centric AI (opens in a new tab)
- SIGMOD (T1)Databricks Lakeguard: Supporting Fine-grained Access Control and Multi-user Capabilities for Apache Spark Workloads (opens in a new tab)
- VLDB Journal (T1)DBOS: Three Years Later (opens in a new tab)
- NeurIPS WorkshopDeepScholar-Bench: A Live Benchmark and Automated Evaluation for Generative Research Synthesis (opens in a new tab)
- VLDB (T1)Delta Sharing: An Open Protocol for Cross-Platform Data Sharing (opens in a new tab)
- ICLR (T1)ElasticTok: Adaptive Tokenization for Image and Video (opens in a new tab)
- NeurIPS (T1)Establishing Best Practices in Building Rigorous Agentic Benchmarks (opens in a new tab)
- ICML (T1)HashAttention: Semantic Sparsity for Faster Inference (opens in a new tab)
- npj Digital Medicine .Identification of Cardiac Wall Motion Abnormalities in Diverse Populations by Deep Learning of the Electrocardiogram (opens in a new tab)
202414 papers
- SIGMOD (T1)ACORN: Performant and Predicate-Agnostic Search Over Vector Embeddings and Structured Data (opens in a new tab)
- VLDB (T1)Adaptive and Robust Query Execution for Lakehouses at Scale (opens in a new tab)
- NeurIPS (T1)Are More LLM Calls All You Need? Towards the Scaling Properties of Compound AI Systems (opens in a new tab)
- NAACL (T1)ARES: An Automated Evaluation Framework for Retrieval-augmented Generation Systems (opens in a new tab)
- ACM/JMS Journal of Data Science, .Data Management for ML-based Analytics and Beyond (opens in a new tab)
- ICLR . Spotlight.DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines (opens in a new tab)
- ASE Workshops .Everything Everywhere All At Once: Efficient Cross-Service Program Analysis with OverSeer (opens in a new tab)
- IEEE Security and Privacy Workshops .Exploiting Programmatic Behavior of LLMs: Dual-use Through Standard Security Attacks (opens in a new tab)
- Harvard Data Science Review, .How Is ChatGPT’s Behavior Changing Over Time? (opens in a new tab)
- Digital Discovery, .Image and Data Mining in Reticular Chemistry Powered by GPT-4V (opens in a new tab)
- EMNLP (T1)Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs (opens in a new tab)
- COLM .RAFT: Adapting Language Model to Domain Specific RAG (opens in a new tab)
202312 papers
- CIDR .Analyzing and Comparing Lakehouse Storage Systems (opens in a new tab)
- NeurIPS R0-FoMo Workshop.Analyzing ChatGPT’s Behavior Shifts Over Time (opens in a new tab)
- SOSP (T1)Cornflakes: Zero-Copy Serialization for Microsecond-Scale Networking (opens in a new tab)
- VLDB (T1)Epoxy: ACID Transactions Across Diverse Data Stores (opens in a new tab)
- AAAI (T1)HAPI Explorer: Comprehension, Discovery, and Explanation on History of ML APIs (demo) (opens in a new tab)
- ICML WorkshopImplementing Block-sparse Matrix Multiplication Kernels Using Triton (opens in a new tab)
- MLSys (T1)MegaBlocks: Efficient Sparse Training with Mixture-of-Experts (opens in a new tab)
- Findings of ACLMoving Beyond Downstream Task Accuracy for Information Retrieval Benchmarking (opens in a new tab)
- VLDB (T1)Optimizing Video Analytics with Declarative Model Relationships (opens in a new tab)
- VLDB (T1)Parallelism-Optimizing Data Placement for Faster Data-Parallel Computations (opens in a new tab)
- VLDB (T1)R3: Record-Replay-Retroaction for Database-Backed Applications (opens in a new tab)
- CIDR .Transactions Make Debugging Easy (opens in a new tab)
Where alumni went16 listed
Apply to UC BerkeleyFall 2027
UC Berkeley EECS Computer Science PhD
- Round
- Single round
- Opens
- —
- GRE
- not required
- English
- TOEFL/IELTS (non-English-country degrees)
- Fee
- See program page
Confirmed on the program page.
Similar labsnearest by research summary
- CNZhejiang
Haibin Shen
- Accelerators, chips & architecture
- CHETH Zurich
Christian Holz
- Interfaces & interaction techniques
- CNSJTU
Chen Chen
- Efficient training & inference
- CNSJTU
Sheng'an Zheng
- Efficient training & inference
- USCMU
Tim Dettmers
- Efficient training & inference
- CNSJTU
Erhu Feng
- Efficient training & inference
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.
UC Berkeleycloses