FRIDA -- Generative Feature Replay for Incremental Domain Adaptation

TL;DR

This paper introduces FRIDA, a novel framework for incremental unsupervised domain adaptation that uses generative feature replay to maintain performance across multiple domains without access to past data.

Contribution

FRIDA employs a new incremental GAN (DGAC-GAN) and an extended domain alignment method (DANN-IB) to improve stability and generalization in incremental domain adaptation.

Findings

FRIDA outperforms existing methods on Office-Home, Office-CalTech, and DomainNet datasets.

The framework effectively balances stability and plasticity in continual domain adaptation.

Experimental results show superior performance in preserving past domain accuracy while adapting to new domains.

Abstract

We tackle the novel problem of incremental unsupervised domain adaptation (IDA) in this paper. We assume that a labeled source domain and different unlabeled target domains are incrementally observed with the constraint that data corresponding to the current domain is only available at a time. The goal is to preserve the accuracies for all the past domains while generalizing well for the current domain. The IDA setup suffers due to the abrupt differences among the domains and the unavailability of past data including the source domain. Inspired by the notion of generative feature replay, we propose a novel framework called Feature Replay based Incremental Domain Adaptation (FRIDA) which leverages a new incremental generative adversarial network (GAN) called domain-generic auxiliary classification GAN (DGAC-GAN) for producing domain-specific feature representations seamlessly. For domain alignment, we propose a simple extension of the popular domain adversarial neural network (DANN) called DANN-IB which encourages discriminative domain-invariant and task-relevant feature learning. Experimental results on Office-Home, Office-CalTech, and DomainNet datasets confirm that FRIDA maintains superior stability-plasticity trade-off than the literature.