Re-evaluating Continual Learning Scenarios: A Categorization and Case for Strong Baselines

Abstract

Continual learning has received a great deal of attention recently with several approaches being proposed. However, evaluations involve a diverse set of scenarios making meaningful comparison difficult. This work provides a systematic categorization of the scenarios and evaluates them within a consistent framework including strong baselines and state-of-the-art methods. The results provide an understanding of the relative difficulty of the scenarios and that simple baselines (Adagrad, L2 regularization, and naive rehearsal strategies) can surprisingly achieve similar performance to current mainstream methods. We conclude with several suggestions for creating harder evaluation scenarios and future research directions. The code is available at https://github.com/GT-RIPL/Continual-Learning-Benchmark

Continual Learning Workshop, 32nd Conference on Neural Information Processing Systems (NIPS 2018)

References in corpus (2)

• Measuring Catastrophic Forgetting in Neural Networks
• Generative replay with feedback connections as a general strategy for continual learning

Cited by in corpus (73)

• LoRA: Low-Rank Adaptation of Large Language Models
• A continual learning survey: Defying forgetting in classification tasks
• Three scenarios for continual learning
• Implicit Generation and Generalization in Energy-Based Models
• A Neural Dirichlet Process Mixture Model for Task-Free Continual Learning
• Dark Experience for General Continual Learning: a Strong, Simple Baseline
• SpaceNet: Make Free Space For Continual Learning
• Domain-Incremental Continual Learning for Mitigating Bias in Facial Expression and Action Unit Recognition
• Orthogonal Gradient Descent for Continual Learning
• Class-incremental Learning with Pre-allocated Fixed Classifiers
• An Investigation of Replay-based Approaches for Continual Learning
• Meta-Consolidation for Continual Learning
• Gradient Projection Memory for Continual Learning
• Balanced softmax cross-entropy for incremental learning with and without memory
• An Adaptive Random Path Selection Approach for Incremental Learning
• Visualizing the PHATE of Neural Networks
• Distilling Causal Effect of Data in Class-Incremental Learning
• Task Agnostic Continual Learning Using Online Variational Bayes with Fixed-Point Updates
• Continual Learning on the Edge with TensorFlow Lite
• A multifidelity approach to continual learning for physical systems
• Continual Learning in Neural Networks
• Rethinking Architecture Design for Tackling Data Heterogeneity in Federated Learning
• Linear Mode Connectivity in Multitask and Continual Learning
• Hierarchical Indian Buffet Neural Networks for Bayesian Continual Learning
• Training Networks in Null Space of Feature Covariance for Continual Learning
• Encoders and Ensembles for Task-Free Continual Learning
• CORA: Benchmarks, Baselines, and Metrics as a Platform for Continual Reinforcement Learning Agents
• Federated Transfer Learning with Dynamic Gradient Aggregation
• Continual Learning using a Bayesian Nonparametric Dictionary of Weight Factors
• Meta-learnt priors slow down catastrophic forgetting in neural networks
• Self-Supervised Learning Aided Class-Incremental Lifelong Learning
• Tackling Dynamics in Federated Incremental Learning with Variational Embedding Rehearsal
• Autoencoder-Based Incremental Class Learning without Retraining on Old Data
• Memory-Efficient Semi-Supervised Continual Learning: The World is its Own Replay Buffer
• Powerpropagation: A sparsity inducing weight reparameterisation
• Towards Fair Affective Robotics: Continual Learning for Mitigating Bias in Facial Expression and Action Unit Recognition
• Formalizing the Generalization-Forgetting Trade-off in Continual Learning
• Unifying Regularisation Methods for Continual Learning
• Localizing Catastrophic Forgetting in Neural Networks
• Learning Invariant Representation for Continual Learning
• Layerwise Optimization by Gradient Decomposition for Continual Learning
• CoReS: Compatible Representations via Stationarity
• Wide Neural Networks Forget Less Catastrophically
• A Combinatorial Perspective on Transfer Learning
• Visually Grounded Continual Learning of Compositional Phrases
• Rethinking Experience Replay: a Bag of Tricks for Continual Learning
• Neuromodulated Neural Architectures with Local Error Signals for Memory-Constrained Online Continual Learning
• Online Continual Learning For Visual Food Classification
• Continual Learning of New Sound Classes using Generative Replay
• Task-agnostic Continual Learning with Hybrid Probabilistic Models
• Continual Learning in the Presence of Repetition
• Neurocoder: Learning General-Purpose Computation Using Stored Neural Programs
• Graph-Based Continual Learning
• AlterSGD: Finding Flat Minima for Continual Learning by Alternative Training
• Learn Faster and Forget Slower via Fast and Stable Task Adaptation
• Energy Aligning for Biased Models
• Domain-Aware Augmentations for Unsupervised Online General Continual Learning
• Regularization-based Continual Learning for Fault Prediction in Lithium-Ion Batteries
• DIODE: Dilatable Incremental Object Detection
• Generative Feature Replay with Orthogonal Weight Modification for Continual Learning
• Unsupervised Continual Learning Via Pseudo Labels
• Mixture-of-Variational-Experts for Continual Learning
• Schematic Memory Persistence and Transience for Efficient and Robust Continual Learning
• TAG: Task-based Accumulated Gradients for Lifelong learning
• Learning Representations on the Unit Sphere: Investigating Angular Gaussian and von Mises-Fisher Distributions for Online Continual Learning
• Leveraging Semantics for Incremental Learning in Multi-Relational Embeddings
• Dynamic Gradient Aggregation for Federated Domain Adaptation
• Principal Gradient Direction and Confidence Reservoir Sampling for Continual Learning
• Better Knowledge Retention through Metric Learning
• Condensed Composite Memory Continual Learning
• Compression-aware Continual Learning using Singular Value Decomposition
• Continual Learning for Image-Based Camera Localization
• Defeating Catastrophic Forgetting via Enhanced Orthogonal Weights Modification