Disentangling Channel Semantics in Vision Transformers via Token Decorrelation and Composition-Aware Modulation

Daeun Kim, Hyejin Park, Hyesong Choi, Dongbo Min

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

Deep Barycentric Regression for Optimal Transport Map Estimation and its Statistical Optimality

BROT estimates optimal transport maps via barycentric regression with deep networks, achieving minimax optimal convergence rates under Lipschitz conditions with stable training.

Kunwoong Kim, Insung Kong, Yongdai Kim

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

A Composite Activation Function for Learning Stable Binary Representations

HTAF smoothly approximates Heaviside via a sigmoid-tanh composite to enable stable gradient-based training of binary neural networks, yielding interpretable ICBMs with comparable or superior accuracy.

Seokhun Park, Choeun Kim, Kwanho Lee, Sehyun Park and 2 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

LUMOS: Tracing Parametric Knowledge from Training Data to Behavioral Outputs in LLMs

LUMOS traces LLM knowledge from verified training exposure to outputs, revealing high encoding but lower expression of rare facts, self-reflection failures on unseen content, and chain-of-thought overconfidence.

Seoyeon Ye, Gayoung Kim, Jiyoung Hong, Soo Kyung Kim and 1 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026