The Dutt Research Group (DRG), led by Distinguished Professor Nikil Dutt at the University of California, Irvine, conducts research at the intersection of embedded and cyber-physical systems, computer architecture, and intelligent computing. Building on our expertise in computational self-awareness and cross-layer co-design, we develop efficient, adaptive, and dependable computing systems that respond to changing workloads, resource constraints, and operating conditions.

Research Directions

Adaptive and Autonomous Systems

We investigate resource-aware perception, localization, mapping, and decision-making for autonomous and robotic systems. Our research emphasizes runtime adaptation, reliable execution, and coordination across sensing, computation, communication, and physical processes.

Efficient AI Models and Runtime Systems

We develop methods for efficient deployment and execution of machine learning models, including transformers, large language models, and vision-language models. Research spans adaptive inference, model sparsity, KV-cache management, inference scheduling, and distributed execution.

Computer Systems and Hardware–Software Co-Design

We explore algorithmic, architectural, and system-level optimizations for efficient and reliable computing. Our work includes hardware-aware AI, memory and storage systems, approximate computing, and cross-layer optimization of computational workloads.

Recent Work

Selected recent work from DRG and our collaborators includes:

• ACCESS-AV — Adaptive communication-computation codesign for sustainable autonomous vehicle localization in smart factories. ACM TECS, 2025.
• MemoGuard — Reliable episodic-memory reuse for communication-limited robot navigation. CODES / IEEE Embedded Systems Letters, 2026.
• AVERY — Intent-driven adaptive VLM split computing for disaster response. Preprint, revised 2026.
• GeoReFormer — Geometry-aware refinement for lane segment detection and topology reasoning. IROS 2026.
• BLAST — Blockwise activation sparsity for transformers. IEEE Embedded Systems Letters, 2026.
• OASIS — Optimized adaptive system for intelligent SLAM. ACM TECS, 2025.
• From Pixels to Places: A Screen Vocabulary for Screen-Aware Agents. WI-IAT 2026, to appear.
• Fairway — Load-aware KV cache reuse for LLM prefill scheduling. Preprint, 2026.
• FlexiCache — Leveraging temporal stability of attention heads for efficient KV cache management. MLSys 2026.
• DPAS — A prompt, accurate and safe I/O completion method for SSDs. USENIX FAST 2026.

Explore our publications, group members, and research projects. Our Past Research Projects archive preserves the group’s research history and historical funding acknowledgments.