FRIDA -- Generative Feature Replay for Incremental Domain Adaptation
This work addresses a novel problem for machine learning practitioners dealing with sequential domain shifts, but it is incremental as it builds on existing methods like GANs and DANN.
The paper tackles the problem of incremental unsupervised domain adaptation, where labeled source and unlabeled target domains are observed sequentially, aiming to maintain accuracy for past domains while adapting to new ones. The proposed FRIDA framework achieves superior stability-plasticity trade-off, as confirmed by experiments on datasets like Office-Home, Office-CalTech, and DomainNet.
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.