Multi-level Collaborative Distillation Meets Global Workspace Model: A Unified Framework for OCIL
Researchers have developed a novel framework for Online Class-Incremental Learning (OCIL) that addresses the challenge of balancing stability and plasticity in models learning from non-i.i.d. data streams. The proposed method, inspired by Global Workspace Theory (GWT), utilizes a Global Workspace Model (GWM) as a shared, implicit memory to guide multiple student models. This GWM is formed by fusing student parameters and is periodically redistributed to stabilize learning and promote cross-task consistency. Additionally, a multi-level collaborative distillation mechanism ensures peer-to-peer consistency and preserves historical knowledge by aligning students with the GWM, leading to significant performance improvements across various OCIL models and memory budgets. AI