Research Interets

Efficient AI

Efficient AI lies at the intersection of algorithm design, systems optimization, and theoretical understanding. My research interests focus on improving the efficiency of modern deep learning systems through principled operator design and optimization-aware modeling. In particular, I am interested in developing high-efficiency operators and inference frameworks that reduce computational and memory overhead while preserving model expressiveness and stability.

Large Language Models

Large language models present new challenges that go beyond traditional deep learning settings, especially in long-context modeling and scalable attention mechanisms. My work explores how to redesign attention and sequence modeling architectures to better handle long-range dependencies, reduce quadratic complexity, and improve robustness under limited computational budgets. I am particularly interested in understanding the inductive biases and optimization dynamics underlying long-context attention in LLMs.

Generative AI

Generative AI represents a paradigm shift from discriminative modeling to learning and synthesizing complex data distributions. My interests in this area focus on the generative modeling paradigm itself, including how representation, memory, and conditioning mechanisms shape generation quality and consistency. I am especially drawn to exploring new generative frameworks that emphasize efficiency, structural consistency, and principled modeling of long-term dependencies.