Mine Your Own vieW: Self-Supervised Learning Through Across-Sample\n Prediction

2021/02/19 by Mehdi Azabou, Azabou, Mehdi, Mohammad Gheshlaghi Azar +23 · 6 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · #AI in cancer detection #Cell Image Analysis Techniques #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning and Data Classification #Multimodal Machine Learning Applications

paper · pdf · doi:10.48550/arxiv.2102.10106

openalex publication_date 2021/02/19 · openalex created_date 2023/02/18 · openalex updated_date 2026/07/28

Abstract

State-of-the-art methods for self-supervised learning (SSL) build\nrepresentations by maximizing the similarity between different transformed\n"views" of a sample. Without sufficient diversity in the transformations used\nto create views, however, it can be difficult to overcome nuisance variables in\nthe data and build rich representations. This motivates the use of the dataset\nitself to find similar, yet distinct, samples to serve as views for one\nanother. In this paper, we introduce Mine Your Own vieW (MYOW), a new approach\nfor self-supervised learning that looks within the dataset to define diverse\ntargets for prediction. The idea behind our approach is to actively mine views,\nfinding samples that are neighbors in the representation space of the network,\nand then predict, from one sample's latent representation, the representation\nof a nearby sample. After showing the promise of MYOW on benchmarks used in\ncomputer vision, we highlight the power of this idea in a novel application in\nneuroscience where SSL has yet to be applied. When tested on multi-unit neural\nrecordings, we find that MYOW outperforms other self-supervised approaches in\nall examples (in some cases by more than 10%), and often surpasses the\nsupervised baseline. With MYOW, we show that it is possible to harness the\ndiversity of the data to build rich views and leverage self-supervision in new\ndomains where augmentations are limited or unknown.\n

Citations

• Data augmentation using learned transformations for one-shot medical image segmentation
• Time-Contrastive Networks: Self-Supervised Learning from Video
• A Simple Framework for Contrastive Learning of Visual Representations
• What Makes for Good Views for Contrastive Learning?
• Momentum Contrast for Unsupervised Visual Representation Learning
• Unsupervised Learning of Visual Features by Contrasting Cluster Assignments
• Data-Efficient Reinforcement Learning with Self-Predictive\n Representations
• Demystifying Contrastive Self-Supervised Learning: Invariances, Augmentations and Dataset Biases
• Learning Multiple Layers of Features from Tiny Images

Cited by

• Know Thyself by Knowing Others: Learning Neuron Identity from Population Context
• Understand and Improve Contrastive Learning Methods for Visual Representation: A Review
• AutoSciDACT: Automated Scientific Discovery through Contrastive Embedding and Hypothesis Testing
• On Feature Decorrelation in Self-Supervised Learning
• Unsupervised Object-Level Representation Learning from Scene Images
• MIC: Model-agnostic Integrated Cross-channel Recommenders

Related

• Learning Representations by Maximizing Mutual Information Across Views
• Occam's Razor for Self Supervised Learning: What is Sufficient to Learn Good Representations?
• Beyond Instance Consistency: Investigating View Diversity in Self-supervised Learning
• Self-supervised Learning from a Multi-view Perspective
• Bootstrap your own latent: A new approach to self-supervised Learning