About
I am a second-year PhD student in the Department of Mathematics at The University of Hong Kong (expected 2029), advised by Professor Guangyue Han. I obtained my BSc degree in Mathematics from Jilin University, TAQ Honors Program in Science, in 2025.
My current research and primary research interests include:
- 01 Information Theory for Learning and Inference
- 02 High-Dimensional Feature Selection and Representation Learning
- 03 Optimization in Deep Learning and Reinforcement Learning
- 04 Foundational Frameworks for AI-driven Scientific Discovery
News
-
On the Cardinality of Optimal Representations in the Binary-Source Information Bottleneck
Uploaded to arXiv. We proved Chen's conjecture that, for the information bottleneck with a binary source, a binary representation is always optimal, so the tight cardinality bound is exactly 2 rather than the classical bound of 3.
-
Mutual Information Constrained Chernoff Bottleneck
Uploaded to arXiv. We proposed a generalized information bottleneck that replaces the relevance term with the Chernoff error exponent of the downstream binary test, established its curve shape and cardinality bound, and developed an alternating algorithm combining a generalized Blahut–Arimoto algorithm and a nonlinear equation solver.
-
Sharp Regret Bounds and a Task-Covariance Correction for Spectral Representation Learning
Uploaded to arXiv. We show that spectral features transfer by aligning with the directions tasks use, not by isotropy, and that only the task covariance matters. A spectral gap gives a criterion for when regret is bounded quadratically rather than linearly in anisotropy, and a task-covariance correction fixes poorly aligned features even in a frozen encoder without retraining.
Publications
- On the Cardinality of Optimal Representations in the Binary-Source Information Bottleneck
- Universal Distributed Feature Extraction Complementary to Side Information via H-Score
- Mutual Information Constrained Chernoff Bottleneck
- Calibrated α-Barycenters for Multiobjective f-Divergence Optimization
- Sharp Regret Bounds and a Task-Covariance Correction for Spectral Representation Learning
- Universal Feature Selection with Noisy Observations and Weak Symmetry Conditions
Presentations
- 2026-06-29 Universal Feature Selection with Noisy Observations and Weak Symmetry Conditions
Teaching
- 2026 · Sem 2 · HKU MATH 6224 / 7224: Topics in Advanced Probability Theory
- 2026 · Sem 1 · HKU MATH 1011: University Mathematics I
Service & Membership
Academic Service
- Journal Reviewer — IEEE Transactions on Information Theory (TIT)
- Conference Reviewer — ISIT (2026)
Professional Membership
- IEEE Graduate Student Member
- SIAM Student Member