Parameter-Masked Decoupled Optimization for Cross-Domain Class-Incremental Learning

Cross-domain class-incremental learning (CD-CIL) requires models to continuously acquire new classes across shifting domains while retaining previously learned knowledge. Existing approaches often entangle what to update with how to update, resulting in unstable adaptation and severe forgetting under domain shifts. Inspired by the hippocampal learning mechanism that separates rapid adaptation from stable consolidation, we propose Parameter-Masked Decoupled Optimization (PMDO) that disentangles what knowledge is adapted from how learning proceeds in cross-domain class-incremental learning. We introduce a domain-aware knowledge decoupler that selectively adapts domain-relevant shared parameters, constraining incremental updates while preserving prior representations. To regulate how learning proceeds, we further design a stability-aware trajectory regulation that guides optimization along transferable and stable optimization trajectories, thereby reducing interference across domain transitions. PMDO enables effective cross-domain adaptation while mitigating catastrophic forgetting and maintaining long-term learnability. Extensive experiments across multiple benchmarks demonstrate the effectiveness of PMDO and its superiority over state-of-the-art methods.

In real-world applications, AI models often need to learn new categories from different domains over time, such as moving from natural images to medical or remote-sensing images. However, when a model learns new knowledge, it may easily overwrite what it learned before. This work proposes PMDO, a new learning framework that helps the model decide which parts should be updated and how they should be updated separately. By selectively adjusting only domain-relevant parameters and guiding the learning process along stable directions, PMDO allows the model to adapt to new domains while better preserving old knowledge. Experiments show that PMDO improves both continual learning ability and long-term knowledge retention across different benchmarks.