5 papers
Understanding Confabulation and Rethinking Reconstruction in Activation Explanations
Gert Lek, Zixuan Xia, Pin-Yu Chen +1
Natural Language Autoencoders (NLAs) produce unsupervised text explanations of a model's activations: a verbalizer describes an activation and a reconstructor learns to recover it…
K-Score: Kalman Filter as a Principled Alternative to Reward Normalization in Reinforcement Learning
Zixuan Xia, Quanxi Li
We propose a simple yet effective alternative to reward normalization in policy gradient reinforcement learning by integrating a 1D Kalman filter for online reward estimation. Inst…
COMPASS: Fusion-Matched Supervision for Missing-Modality Human Sensing
Hao Wang, Yanyu Qian, Pengcheng Weng +4
Multimodal human activity recognition (HAR) and human pose estimation (HPE) must cope with modalities missing at inference. Completions generated for a fixed fusion model must pres…
Diverse via bounded Agreement: Geometric Regularization for Multimodal Fusion
Zixuan Xia, Hao Wang, Pengcheng Weng +4
Multimodal fusion is often treated as an optimization-balancing problem, where training signals are adjusted to prevent one modality from dominating the others. However, balanced o…
KOALA++: Efficient Kalman-Based Optimization with Gradient-Covariance Products
Zixuan Xia, Aram Davtyan, Paolo Favaro
We propose KOALA++, a scalable Kalman-based optimization algorithm that explicitly models structured gradient uncertainty in neural network training. Unlike second-order methods, w…