aiqbench analysis
MOSAIC addresses the critical challenge of incremental continual learning in Parkinson's disease gait assessment by enabling AI models to adapt to new sensor modalities without forgetting prior knowledge. The framework introduces specific mechanisms like Modality-Specific Warm-Up and a statistics-decoupled architecture to handle data heterogeneity common in clinical deployments where historical datasets are unavailable. This research advances the practicality of deploying robust, evolving AI systems for medical diagnostics under real-world constraints.
MOSAIC: Modality-Specific Adaptation for Incremental Continual Learning in Parkinson's Disease Gait Assessment
arXiv:2606.13258v2 Announce Type: replace Abstract: Gait-based Parkinson's disease assessment increasingly relies on heterogeneous sensors, but clinical systems rarely collect all modalities simultaneously. New sensors may arrive through device upgrades, protocol changes, or multi-center deployment, while historical patient data are often unavailable because of privacy and storage constraints. This modality-incremental setting faces three challenges: unreliable cross-modal distillation, modality-specific statistical shifts, and reduced plasticity after preservation. We propose MOSAIC, a compact continual learning framework. First, we identify the Toxic Teacher phenomenon and introduce Modality-Specific Warm-Up to stabilize newly learned modality representations before distillation. Second, we propose a statistics-decoupled MSBN architecture that isolates sensor statistics while maintaining a shared semantic backbone. Third, we design a curriculum-guided repulsive objective for Plasticity Recovery, preserving legacy knowledge while recovering modality-specific capacity. Experiments on three multimodal Parkinson's gait datasets show that MOSAIC improves final performance and mitigates forgetting. Project code is available at: https://github.com/minlinzeng/MOSAIC_Modality-Specific-Adaptation-for-Incremental-Continual-Learning-in-PD-Gait-Assessment.git
Abstract:Gait-based Parkinson's disease assessment increasingly relies on heterogeneous sensors, but clinical systems rarely collect all modalities simultaneously. New sensors may arrive through device upgrades, protocol changes, or multi-center deployment, while historical patient data are often unavailable because of privacy and storage constraints. This modality-incremental setting faces three challenges: unreliable cross-modal distillation, modality-specific statistical shifts, and reduced plasticity after preservation. We propose MOSAIC, a compact continual learning framework. First, we identify the Toxic Teacher phenomenon and introduce Modality-Specific Warm-Up to stabilize newly learned modality representations before distillation. Second, we propose a statistics-decoupled MSBN architecture that isolates sensor statistics while maintaining a shared semantic backbone. Third, we design a curriculum-guided repulsive objective for Plasticity Recovery, preserving legacy knowledge while recovering modality-specific capacity. Experiments on three multimodal Parkinson's gait datasets show that MOSAIC improves final performance and mitigates forgetting. Project code is available at: this https URL
Artificial Intelligence (cs.AI)
arXiv:2606.13258 [cs.AI]
(or arXiv:2606.13258v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2606.13258 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 236 words
Related coverage
MOSAIC: Modality-Specific Adaptation for Incremental Continual Learning in Parkinson's Disease Gait Assessment
arXiv · Jun 12, 2026
DynaTokens: Controlling Token Dynamics for Continual Video-Language Understanding
arXiv · Sep 3, 2026
FreKoo++: Learning Continuous Spectral Dynamics for Temporal Domain Generalization
arXiv · Aug 25, 2026
Dyna-2: General Purpose Physical Agents with Hour-Long Autonomy — Jason Ma, Dyna Robotics
Latent Space · Oct 13, 2026