Unsupervised Learning by Predicting Noise

Abstract

Convolutional neural networks provide visual features that perform remarkably well in many computer vision applications. However, training these networks requires significant amounts of supervision. This paper introduces a generic framework to train deep networks, end-to-end, with no supervision. We propose to fix a set of target representations, called Noise As Targets (NAT), and to constrain the deep features to align to them. This domain agnostic approach avoids the standard unsupervised learning issues of trivial solutions and collapsing of features. Thanks to a stochastic batch reassignment strategy and a separable square loss function, it scales to millions of images. The proposed approach produces representations that perform on par with state-of-the-art unsupervised methods on ImageNet and Pascal VOC.

Cited by in corpus (51)

• Unsupervised Learning of Visual Features by Contrasting Cluster Assignments
• DINOv2: Learning Robust Visual Features without Supervision
• Barlow Twins: Self-Supervised Learning via Redundancy Reduction
• Contrastive Multiview Coding
• Deep Clustering for Unsupervised Learning of Visual Features
• Image Quality Assessment using Contrastive Learning
• Towards Representation Alignment and Uniformity in Collaborative Filtering
• Optimizing the Latent Space of Generative Networks
• Understanding Contrastive Representation Learning through Alignment and Uniformity on the Hypersphere
• Unsupervised Learning via Meta-Learning
• Self-Supervised Spatiotemporal Feature Learning via Video Rotation Prediction
• Improvements to context based self-supervised learning
• Cluster-guided Asymmetric Contrastive Learning for Unsupervised Person Re-Identification
• Space-Time Correspondence as a Contrastive Random Walk
• A critical analysis of self-supervision, or what we can learn from a single image
• Meta-Learning Update Rules for Unsupervised Representation Learning
• Test-Time Training with Self-Supervision for Generalization under Distribution Shifts
• Hyperspherical Prototype Networks
• Unsupervised Alignment of Embeddings with Wasserstein Procrustes
• Sinkhorn AutoEncoders
• Practical Vertical Federated Learning with Unsupervised Representation Learning
• Image Clustering using an Augmented Generative Adversarial Network and Information Maximization
• MAD: Self-Supervised Masked Anomaly Detection Task for Multivariate Time Series
• Deep Comprehensive Correlation Mining for Image Clustering
• When Does Self-supervision Improve Few-shot Learning?
• Modeling Multiple Views via Implicitly Preserving Global Consistency and Local Complementarity
• How Useful is Self-Supervised Pretraining for Visual Tasks?
• SCARF: Self-Supervised Contrastive Learning using Random Feature Corruption
• Déjà Vu: an empirical evaluation of the memorization properties of ConvNets
• Towards a Near Universal Time Series Data Mining Tool: Introducing the Matrix Profile
• flexgrid2vec: Learning Efficient Visual Representations Vectors
• Self-Adaptive Training: Bridging Supervised and Self-Supervised Learning
• SLA${}^2$P: Self-supervised Anomaly Detection with Adversarial Perturbation
• TextTopicNet - Self-Supervised Learning of Visual Features Through Embedding Images on Semantic Text Spaces
• Audio-Visual Self-Supervised Terrain Type Discovery for Mobile Platforms
• Cross-Domain Self-supervised Multi-task Feature Learning using Synthetic Imagery
• Loss Function Entropy Regularization for Diverse Decision Boundaries
• DeepDiffusion: Unsupervised Learning of Retrieval-adapted Representations via Diffusion-based Ranking on Latent Feature Manifold
• Multimodal Representation Learning via Maximization of Local Mutual Information
• Federated Learning of User Verification Models Without Sharing Embeddings
• I-SAFE: Instant Suspicious Activity identiFication at the Edge using Fuzzy Decision Making
• Unsupervised learning with sparse space-and-time autoencoders
• Information Theory-Guided Heuristic Progressive Multi-View Coding
• An Iterative Closest Points Approach to Neural Generative Models
• Towards Robust Pattern Recognition: A Review
• MC-SSL0.0: Towards Multi-Concept Self-Supervised Learning
• Representation Learning by Reconstructing Neighborhoods
• CycleCluster: Modernising Clustering Regularisation for Deep Semi-Supervised Classification
• Discriminative Clustering with Representation Learning with any Ratio of Labeled to Unlabeled Data
• Constructing Multiple Tasks for Augmentation: Improving Neural Image Classification With K-means Features
• Learning Representations on the Unit Sphere: Investigating Angular Gaussian and von Mises-Fisher Distributions for Online Continual Learning