General Domain Adaptation Through Proportional Progressive Pseudo Labeling
This work addresses the problem of developing a more general domain adaptation technique for researchers and practitioners, aiming to overcome the limitation of existing methods that perform poorly across different data types.
This paper introduces Proportional Progressive Pseudo Labeling (PPPL), a domain adaptation technique designed to work across various input types like images, text, and time-series. PPPL progressively reduces target domain classification error by training with pseudo-labeled target samples while excluding and postponing training on samples with less reliable pseudo-labels. Experiments on 6 diverse datasets show that PPPL outperforms other baselines and generalizes better.
Domain adaptation helps transfer the knowledge gained from a labeled source domain to an unlabeled target domain. During the past few years, different domain adaptation techniques have been published. One common flaw of these approaches is that while they might work well on one input type, such as images, their performance drops when applied to others, such as text or time-series. In this paper, we introduce Proportional Progressive Pseudo Labeling (PPPL), a simple, yet effective technique that can be implemented in a few lines of code to build a more general domain adaptation technique that can be applied on several different input types. At the beginning of the training phase, PPPL progressively reduces target domain classification error, by training the model directly with pseudo-labeled target domain samples, while excluding samples with more likely wrong pseudo-labels from the training set and also postponing training on such samples. Experiments on 6 different datasets that include tasks such as anomaly detection, text sentiment analysis and image classification demonstrate that PPPL can beat other baselines and generalize better.