Purify-then-Align: Towards Robust Human Sensing under Modality Missing with Knowledge Distillation from Noisy Multimodal Teacher

2026-04-07

AI Summary

This work addresses the representation gap and performance degradation caused by missing or low-quality modalities in multimodal human sensing. To tackle these challenges, the authors propose a “purify-then-align” framework that first suppresses the influence of noisy modalities through a meta-learning-driven dynamic modality weighting mechanism, followed by diffused knowledge distillation to transfer the purified multimodal teacher knowledge to a unimodal student model. This approach uniquely integrates meta-learning with diffused knowledge distillation, effectively decoupling the causal dependency between modality corruption and representation disparity. Experimental results on the MM-Fi and XRF55 datasets demonstrate that the method significantly enhances the robustness and performance of unimodal models across diverse modality-missing scenarios.

Abstract

Robust multimodal human sensing must overcome the critical challenge of missing modalities. Two principal barriers are the Representation Gap between heterogeneous data and the Contamination Effect from low-quality modalities. These barriers are causally linked, as the corruption introduced by contamination fundamentally impedes the reduction of representation disparities. In this paper, we propose PTA, a novel "Purify-then-Align" framework that solves this causal dependency through a synergistic integration of meta-learning and knowledge diffusion. To purify the knowledge source, PTA first employs a meta-learning-driven weighting mechanism that dynamically learns to down-weight the influence of noisy, low-contributing modalities. Subsequently, to align different modalities, PTA introduces a diffusion-based knowledge distillation paradigm in which an information-rich clean teacher, formed from this purified consensus, refines the features of each student modality. The ultimate payoff of this "Purify-then-Align" strategy is the creation of exceptionally powerful single-modality encoders imbued with cross-modal knowledge. Comprehensive experiments on the large-scale MM-Fi and XRF55 datasets, under pronounced Representation Gap and Contamination Effect, demonstrate that PTA achieves state-of-the-art performance and significantly improves the robustness of single-modality models in diverse missing-modality scenarios.

Problem

• modality missing
• representation gap
• contamination effect
• robust human sensing
• multimodal learning

Innovation

• Purify-then-Align
• knowledge distillation
• meta-learning
• multimodal robustness
• modality missing