Continual Unsupervised Representation Learning

Abstract

Continual learning aims to improve the ability of modern learning systems to deal with non-stationary distributions, typically by attempting to learn a series of tasks sequentially. Prior art in the field has largely considered supervised or reinforcement learning tasks, and often assumes full knowledge of task labels and boundaries. In this work, we propose an approach (CURL) to tackle a more general problem that we will refer to as unsupervised continual learning. The focus is on learning representations without any knowledge about task identity, and we explore scenarios when there are abrupt changes between tasks, smooth transitions from one task to another, or even when the data is shuffled. The proposed approach performs task inference directly within the model, is able to dynamically expand to capture new concepts over its lifetime, and incorporates additional rehearsal-based techniques to deal with catastrophic forgetting. We demonstrate the efficacy of CURL in an unsupervised learning setting with MNIST and Omniglot, where the lack of labels ensures no information is leaked about the task. Further, we demonstrate strong performance compared to prior art in an i.i.d setting, or when adapting the technique to supervised tasks such as incremental class learning.

NeurIPS 2019

Cited by in corpus (42)

• Lifelong Teacher-Student Network Learning
• DualNet: Continual Learning, Fast and Slow
• Class-incremental Learning with Pre-allocated Fixed Classifiers
• From Machine Learning to Robotics: Challenges and Opportunities for Embodied Intelligence
• Preserving Earlier Knowledge in Continual Learning with the Help of All Previous Feature Extractors
• Continual World: A Robotic Benchmark For Continual Reinforcement Learning
• Task Agnostic Continual Learning Using Online Variational Bayes with Fixed-Point Updates
• Quantum Continual Learning Overcoming Catastrophic Forgetting
• Hierarchically Organized Latent Modules for Exploratory Search in Morphogenetic Systems
• Linear Mode Connectivity in Multitask and Continual Learning
• Encoders and Ensembles for Task-Free Continual Learning
• Hierarchical Indian Buffet Neural Networks for Bayesian Continual Learning
• Hypernetworks for Continual Semi-Supervised Learning
• Learning latent representations across multiple data domains using Lifelong VAEGAN
• Self-Supervised Learning Aided Class-Incremental Lifelong Learning
• Continual Representation Learning for Biometric Identification
• Pseudo-Rehearsal for Continual Learning with Normalizing Flows
• Few-Shot Unsupervised Continual Learning through Meta-Examples
• Lifelong Learning using Eigentasks: Task Separation, Skill Acquisition, and Selective Transfer
• Bayesian Structure Adaptation for Continual Learning
• Modelling Latent Skills for Multitask Language Generation
• Meta-Reinforcement Learning in Broad and Non-Parametric Environments
• Continual Learning from the Perspective of Compression
• Task-agnostic Continual Learning with Hybrid Probabilistic Models
• Detecting and Adapting to Irregular Distribution Shifts in Bayesian Online Learning
• On Catastrophic Interference in Atari 2600 Games
• Semi-Supervised Class Discovery
• Graph-Based Continual Learning
• Streaming Self-Training via Domain-Agnostic Unlabeled Images
• Deep Bayesian Unsupervised Lifelong Learning
• Unsupervised Class-Incremental Learning Through Confusion
• Towards Recognizing New Semantic Concepts in New Visual Domains
• Discriminative Distillation to Reduce Class Confusion in Continual Learning
• A Conditional Generative Matching Model for Multi-lingual Reply Suggestion
• Online Unsupervised Learning of Visual Representations and Categories
• POLA: Online Time Series Prediction by Adaptive Learning Rates
• Lifelong Twin Generative Adversarial Networks
• Meta-Learning with Variational Bayes
• Online Continual Learning in Image Classification: An Empirical Survey
• BooVAE: Boosting Approach for Continual Learning of VAE
• TAG: Task-based Accumulated Gradients for Lifelong learning
• Lifelong Mixture of Variational Autoencoders