arXiv subjects

Zixuan Xia

Publications and source records attributed to Zixuan Xia.

5 recordsLinked to original sources

arXiv2026

Exact Regret Frontiers and Externality Scheduling in Centralized Serial-Dictatorship Bandits

Exploration in centralized serial-dictatorship matching bandits must use complete matchings, so learning one player--arm pair can impose regret on others. We study this externality under a known common priority order and Gaussian rewards with unit variance. We show that the matching-level Graves--Lai constraints reduce to finitely many pairwise exploration quotas and, at top-choice-separated instances, yield a polynomial-size marginal linear program. At these instances, the exact attainable set of expected logarithmic regret coefficients is $G(θ)\Xset(θ)$, where $\Xset$ is the feasible matching-allocation set and $G$ maps allocations to player regret. The usual upper-closed Graves--Lai region can be strictly larger despite having the same Pareto-minimal boundary. We further show that identical exploration quotas can induce very different regret through their scheduling. Finally, we construct estimate--solve--track policies, uniformly good on the full row-strict class, that attain every fixed positively weighted optimum without assuming optimizer uniqueness. Every Pareto-minimal point is pointwise attainable, possibly through an instance-calibrated target.

cs.GT↗

arXiv2026

Diverse via bounded Agreement: Geometric Regularization for Multimodal Fusion

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 optimization does not fully determine the geometry of intermediate representations. Supervised multimodal models may still learn low-diversity modality-specific embeddings or allow paired cross-modal observations to drift excessively apart, weakening both unimodal robustness and multimodal fusion. We introduce \regName, a lightweight plug-and-play geometric regularization framework for multimodal representation learning. Rather than enforcing rigid cross-modal alignment, \regName follows a bounded-agreement principle: preserve modality-specific diversity while softly constraining only the portion of paired cross-modal drift that exceeds an admissible agreement band. Operationally, \regName combines a dispersion term that mitigates spectral concentration with an agreement-band anchoring term that controls excessive paired drift, requiring no architectural modification or inference-time overhead. Experiments across audio-visual, image-text, and RF-based benchmarks show that \regName consistently improves multimodal performance and often strengthens unimodal representations. These results suggest that explicitly regulating representation geometry is an effective complement to optimization balancing, and provide evidence that geometry-aware regularization can improve multimodal learning across diverse architectures and domains.

cs.CV↗

arXiv2026

COMPASS: Complete Multimodal Fusion via Proxy Tokens and Shared Spaces for Ubiquitous Sensing

Missing modalities in multimodal sensing cause not only information loss but also a fusion-interface mismatch: a fusion head trained on a canonical set of modality slots must operate on changing observed subsets at inference time. We propose Compass, an interface-complete fusion framework that restores this canonical slot structure before prediction. Each modality is assigned a fixed fusion slot. Observed modalities populate their slots with real representations, while absent modalities are filled with target-slot completion representations estimated from the observed sources. Multiple source-specific estimates for the same missing slot are aggregated into a single slot filler, allowing the same lightweight fusion operator to be applied under arbitrary missing-modality patterns. Training uses synthetic modality masking, slot-compatibility supervision, and representation-space stabilization to make completed slots compatible with real modality representations and useful for downstream recognition. Across XRF55, MM-Fi, and OctoNet, Compass improves robustness under diverse single- and multiple-missing settings, including controlled comparisons against imputation, distillation, and translation-style baselines. These results suggest that preserving the fusion interface is a simple and effective principle for robust multimodal sensing.

cs.CV↗

arXiv2026

K-Score: Kalman Filter as a Principled Alternative to Reward Normalization in Reinforcement Learning

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. Instead of relying on fixed heuristics, our method recursively estimates the latent reward mean, smoothing high-variance returns and adapting to non-stationary environments. This approach incurs minimal overhead and requires no modification to existing policy architectures. Experiments on \textit{LunarLander} and \textit{CartPole} demonstrate that Kalman-filtered rewards significantly accelerate convergence and reduce training variance compared to standard normalization techniques. Code is available at https://github.com/Sumxiaa/Kalman_Normalization.

cs.LG↗

arXiv2025

KOALA++: Efficient Kalman-Based Optimization with Gradient-Covariance Products

We propose KOALA++, a scalable Kalman-based optimization algorithm that explicitly models structured gradient uncertainty in neural network training. Unlike second-order methods, which rely on expensive second order gradient calculation, our method directly estimates the parameter covariance matrix by recursively updating compact gradient covariance products. This design improves upon the original KOALA framework that assumed diagonal covariance by implicitly capturing richer uncertainty structure without storing the full covariance matrix and avoiding large matrix inversions. Across diverse tasks, including image classification and language modeling, KOALA++ achieves accuracy on par or better than state-of-the-art first- and second-order optimizers while maintaining the efficiency of first-order methods.

cs.LG↗