Current research areas

My group works at the intersection of computer systems and machine learning, in both directions. On one side, we work across the stack to make modern ML workloads faster, cheaper, and more reliable: from efficient ML model architecture design to GPU energy modeling. On the other, we use machine learning inside core systems, in areas such as memory tiering and cloud configuration tuning.

• Improving LLM inference: Reducing the cost of running large language models through inference-efficient model architectures, memory management, and speculative decoding.
• GPU variability & power management: Understanding how variability across GPUs affects cluster performance through variability-aware scheduling and high-fidelity GPU energy modeling.
• Vector search: New indexing and query methods to make vector similarity search faster and more scalable, including adaptive indexes and vector database deployment on HPC platforms.
• Integrating ML into systems: Using machine learning to improve core system components, including memory tiering and tuning unstable and noisy cloud applications.

Group

• Rutwik Jain — co-advised w/ Matt Sinclair
• Brandon Tran — co-advised w/ Matt Sinclair
• Minghao Yan
• Johannes Freischuetz
• Tzu-Tao Chang
• Fanchao Chen
• Tareq Mahmood
• Seth Ockerman

Selected recent publications

Please see Google Scholar for a complete list.