Global Identifiability of L1-based Dictionary Learning via Matrix Volume Optimization
Establishes global identifiability guarantees for L1-based dictionary learning via a matrix-volume formulation.
About me
I am an Applied Scientist at Amazon, working on LLM-based agentic reasoning systems and adaptive decision-making. My research spans identifiable representation learning, non-convex optimization, and learning from partial feedback.
I received my Ph.D. in Computer Science from the University of Florida, advised by Prof. Kejun Huang. Before joining UF, I received my bachelor’s degree from Nanjing University.
Selected publications
All publications on Scholar ↗
Identifiable Bounded Component Analysis via Minimum Volume Enclosing Parallelotope
Introduces a minimum-volume enclosing parallelotope view for identifiable bounded component analysis.
Complex Bounded Component Analysis: Identifiability and Algorithm
Provides identifiability analysis and an efficient algorithm for complex-valued bounded component analysis.
Research interests
- Machine learning foundations. Identifiable representation learning, latent variable models, and non-convex optimization.
- Adaptive decision systems. Contextual neural bandits, learning from partial feedback, and uncertainty-aware decisions.
- LLM & agentic reasoning. Routing, orchestration, evaluation, and control for systems built on large language models.
Academic service
Conference reviewer: NeurIPS, ICML, ICLR, AISTATS, AAAI, ICASSP, MLSP, IJCNN.
Journal reviewer: IEEE Transactions on Signal Processing; Journal of Machine Learning Research.