Continual Domain Adaptation Through Pruning-Aided Domain-Specific Weight Modulation

Prasanna B, Sunandini Sanyal, R. Venkatesh Babu; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 2457-2463

Abstract


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.

Related Material


[pdf] [arXiv]
[bibtex]
@InProceedings{B_2023_CVPR, author = {B, Prasanna and Sanyal, Sunandini and Babu, R. Venkatesh}, title = {Continual Domain Adaptation Through Pruning-Aided Domain-Specific Weight Modulation}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2023}, pages = {2457-2463} }