Research
Efficient Training and Deployment of Language and Vision Models
We develop techniques to reduce the computational and memory costs of training and deploying large-scale language and vision models. Our research focuses on efficient training methods, model compression techniques such as quantization and sparsity, KV-cache optimization, and efficient inference and serving systems.
AI-Driven Circuit Design and Hardware Optimization
We explore AI-driven approaches to automate and optimize electronic circuit design. Our research focuses on generative AI for circuit topology generation, learning-based design-space exploration, and intelligent hardware optimization to improve design efficiency and reduce development costs.
Domain-Specific AI Accelerators and Memory-Centric Architectures
We explore novel computer architectures to enable high-performance and energy-efficient AI computing. Our research focuses on domain-specific AI accelerators, memory-centric computing architectures, and hardware-software co-design to overcome computational and memory bottlenecks in modern AI workloads.
