When specifications are incomplete, AI systems may make incorrect assumptions and leave users to discover the failures after the fact.
Projects
Research projects in self-verifying AI, self-improving data systems, cloud optimization and FinOps, scalable machine learning, database systems, and interactive analytics.
Current Research
Self-Improving Data Systems
GenRewriteDAGSmith
AI, database learning, and workload feedback for improving queries and data pipelines while preserving semantics and operational constraints.
Workload-aware methods for improving performance, resource efficiency, and cost across cloud data warehouses and infrastructure.
Selected Earlier Work
Methods for making machine learning more efficient, scalable, and robust under communication, computation, and memory constraints.
Methods for diagnosing, bounding, and reducing performance variability under changing workloads.
Database Learning: QuickSel and BlinkML
QuickSelBlinkML
Database-learning methods that improve estimation and reduce computation using workload history.
Systems for interactive analytics over massive data with explicit guarantees on accuracy and response time.
Variance-aware and contention-aware scheduling for reducing tail latency and throughput collapse under high contention.
Tools for diagnosing performance failures, predicting resource needs, and explaining workload-driven behavior in production databases.
Robust Database Design: CliffGuard
CliffGuard Code ↗
Physical database designs that remain effective when workloads and estimates change.
Archive
Streaming Data and Complex Event Processing
Stream Mill MinerK*SQLXSeq
Query languages and systems for mining streams and matching complex patterns over structured data.
Active-learning methods that reduce labeling cost by focusing human effort on the most informative tasks.