CRITICAL ANALYSIS OF SELF-SUPERVISION

Presented by

Maral Rasoolijaberi

Introduction

This paper evaluated the performance of state-of-the-art unsupervised learning methods on learning weights of convolutional neural networks (CNNs) to figure out whether self-supervised methods can learn deep features from only one image. In self-supervised learning, data generate ground truth labels per se by pretext tasks such as rotation estimation. The main goal of self-supervised learning is utilizing unlabeled data, e.g., a picture of a dog without the label “dog”, for training CNNs and finding generalized image representations.