MOSAIC: Modality-Specific Adaptation for Incremental Continual Learning in Parkinson's Disease Gait Assessment

2026-06-11 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

MOSAIC is a compact continual learning framework designed for Parkinson's disease gait assessment using heterogeneous sensors in a modality-incremental setting. This framework addresses three key challenges: unreliable cross-modal distillation, modality-specific statistical shifts, and reduced plasticity after preservation, which arise when new sensors are introduced or historical patient data is unavailable due to privacy. MOSAIC introduces Modality-Specific Warm-Up to stabilize new modality representations, a statistics-decoupled MSBN architecture to isolate sensor statistics while maintaining a shared semantic backbone, and a curriculum-guided repulsive objective for Plasticity Recovery. Experiments conducted on three multimodal Parkinson's gait datasets demonstrate that MOSAIC significantly improves final performance and effectively mitigates forgetting. The project code is publicly available.

Key takeaway

For Machine Learning Engineers developing clinical systems for Parkinson's disease gait assessment with evolving sensor modalities, you should consider integrating MOSAIC's continual learning strategies. This framework directly addresses challenges like unreliable cross-modal distillation and plasticity loss, ensuring your models adapt effectively to new data streams without forgetting historical patient information. Implementing its Modality-Specific Warm-Up and MSBN architecture can significantly improve model performance and mitigate forgetting in real-world deployments.

Key insights

MOSAIC enables robust continual learning for multimodal Parkinson's gait assessment by addressing modality-specific challenges.

Principles

• Stabilize new modality representations before distillation.
• Decouple sensor statistics from semantic backbone.
• Recover plasticity while preserving legacy knowledge.

Method

MOSAIC employs Modality-Specific Warm-Up, a statistics-decoupled MSBN architecture, and a curriculum-guided repulsive objective for Plasticity Recovery to manage incremental sensor data.

In practice

• Implement Modality-Specific Warm-Up to prevent "Toxic Teacher" issues.
• Utilize MSBN for shared semantic learning with decoupled statistics.
• Apply repulsive objectives for plasticity recovery in continual learning.

Topics

• Continual Learning
• Parkinson's Disease
• Gait Assessment
• Multimodal Sensors
• Modality Adaptation
• MSBN Architecture

Code references

• minlinzeng/MOSAIC_Modality-Specific-Adaptation-for-Incremental-Continual-Learning-in-PD-Gait-Assessment

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• MOSAIC: Modality-Specific Adaptation for Incremental Continual Learning in Parkinson's Disease Gait Assessment — MOSAIC resolves modality-incremental learning challenges in PD gait assessment by stabilizing new sensor representations and decoupling statistics.
• Continual Learning for fMRI-Based Brain Disorder Diagnosis via Functional Connectivity Matrices Generative Replay — A continual learning framework uses generative replay and knowledge distillation to diagnose brain disorders from sequential fMRI data.
• Treatment-Conditioned Diffusion for Forecasting Neurodegenerative Disease Progression — A treatment-conditioned diffusion framework accurately forecasts neurodegenerative disease progression by integrating imaging and pharmacological data, preserving anatomical detail.
• EvoBrain: Continual Learning of EEG Foundation Models Across Heterogeneous BCI Tasks — EvoBrain pioneers cross-task continual learning for EEG foundation models, balancing plasticity and stability for unified brain decoding.
• Beyond Single-Model Optimization: Preserving Plasticity in Continual Reinforcement Learning — Preserving diverse policy neighborhoods in a stable latent space enhances continual reinforcement learning plasticity.
• MASER: Modality-Adaptive Specialist Routing for Embodied 3D Spatial Intelligence — Dynamically selecting the best modality adapter based on question semantics improves 3D spatial reasoning for embodied agents.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.