The Taub Faculty of Computer Science Events and Talks
Evgenii Zheltonozhskii (M.Sc. Thesis Seminar)
Tuesday, 26.10.2021, 14:30
Advisor: Prof. A. Mendelson, Prof. A. Bronstein, Dr. C. Baskin
While deep neural networks (DNNs) have shown tremendous success across various computer vision tasks, including image classification, object detection, and semantic segmentation, requirements for a large number of high-quality labels obstruct the adoption of DNNs in real-life problems. Lately, researchers have proposed multiple approaches for reducing requirements to the amount or quality of these labels or even working in a fully unsupervised way. In a series of works, we study different approaches to supervision reduction in visual recognition tasks: self-supervised learning, learning with noisy labels, and semi-supervised learning. For self-supervised learning, we show that dimensionality reduction followed by simple k-nearest neighbors clustering is a very strong baseline for fully unsupervised large-scale classification (ImageNet). Additionally, we present a learning with noisy labels framework comprising two stages: self-supervised pre-training and robust fine-tuning. The framework, dubbed "Contrast to Divide" (C2D), significantly outperforms prior art on synthetic and real-life noise, showing state-of-the-art performance with different methods and pre-training approaches. Furthermore, since self-supervised pre-training is unaffected by label noise, C2D is especially efficient in a high noise regime. Finally, for semi-supervised learning, we propose a simple weighting scheme that reduces confirmation bias among unlabeled samples and, as a result, outperforms existing methods on different datasets and a wide range of labeled sample fractions. The talk will be given in English.