Self-supervised Neural Architecture Search
This addresses the limitation of NAS for researchers and practitioners in scenarios with scarce labeled data, representing an incremental advancement by combining self-supervised learning with NAS.
The paper tackles the problem of Neural Architecture Search (NAS) requiring large labeled datasets by proposing a self-supervised NAS method that uses unlabeled data, achieving comparable results to supervised NAS and improving self-supervised learning performance, especially with limited labels.
Neural Architecture Search (NAS) has been used recently to achieve improved performance in various tasks and most prominently in image classification. Yet, current search strategies rely on large labeled datasets, which limit their usage in the case where only a smaller fraction of the data is annotated. Self-supervised learning has shown great promise in training neural networks using unlabeled data. In this work, we propose a self-supervised neural architecture search (SSNAS) that allows finding novel network models without the need for labeled data. We show that such a search leads to comparable results to supervised training with a "fully labeled" NAS and that it can improve the performance of self-supervised learning. Moreover, we demonstrate the advantage of the proposed approach when the number of labels in the search is relatively small.