Continual Domain Adaptation through Pruning-aided Domain-specific Weight Modulation
This addresses the challenge of adapting models to new domains over time without losing past knowledge, which is incremental as it builds on existing UDA and CL methods.
The paper tackles the problem of unsupervised domain adaptation in continual learning by updating models on changing domains while preventing catastrophic forgetting, achieving state-of-the-art performance with significant prevention of forgetting.
In this paper, we propose to develop a method to address unsupervised domain adaptation (UDA) in a practical setting of continual learning (CL). The goal is to update the model on continually changing domains while preserving domain-specific knowledge to prevent catastrophic forgetting of past-seen domains. To this end, we build a framework for preserving domain-specific features utilizing the inherent model capacity via pruning. We also perform effective inference using a novel batch-norm based metric to predict the final model parameters to be used accurately. Our approach achieves not only state-of-the-art performance but also prevents catastrophic forgetting of past domains significantly. Our code is made publicly available.