Publications by Research Theme

(* = equal contribution)
(** = alphabetical order)
(† = co-last author)
(Each paper appears in a single research theme.)

The organization below follows the three research themes used on the homepage. Within each theme, papers are displayed from the most recent work to earlier contributions.

Latent Structure, Mixtures, and Identifiability · Modern AI Architectures and Conditional Computation · Geometry, Computation, and Robustness

Papers are displayed from newest to oldest within each theme, while the labels remain chronological and stable (for example, [L.1] is the earliest paper in the latent-structure theme).

Latent Structure, Mixtures, and Identifiability

Finite and infinite mixture models, hierarchical and Bayesian models, identifiability and singularity, latent-variable estimation, Bayesian inference, and related statistical foundations.

Modern AI Architectures and Conditional Computation

Mixture-of-experts, routing and gating, attention and Transformers, parameter-efficient adaptation, multimodal and medical AI, continual and federated learning, and other architecture-centered problems in modern AI.

Geometry, Computation, and Robustness

Optimal transport and Wasserstein geometry, scalable computation, optimization and statistical-computational trade-offs, distributional robustness, and geometric methods for machine learning.