Continual Learning

More Resources:

Continual Learning Repository@ Mila

Other resources

Surveys:

• Embracing Change: Continual Learning in Deep Neural Networks (2020)

• Towards Continual Reinforcement Learning: A Review and Perspectives (2020)

• A wholistic view of continual learning with deep neural networks: Forgotten lessons and the bridge to active and open world learning (2020)

• Matthias De Lange et al. "Continual learning: A comparative study on how to defy forgetting in classification tasks" (2019) - focus on task-incremental supervised continual learning, where tasks arrive sequentially and are delineated by clear boundaries

• Parisi, German I., et al. "Continual lifelong learning with neural networks: A review." Neural Networks (2019).

• Book: Lifelong Machine Learning (2nd ed. 2018)

• ICML 2019 Tutorial on Never-Ending Learning by Tom Mitchell and Partha Talukdar

Classics:

Recent:

MER paper (ICL2019)

Model-agnostic meta-learning for fast adaptation of deep networks

Meta-Learning Representations for Continual Learning

*Understanding the Role of Training Regimes in Continual Learning

Learning Causal Models Online (IRM + CL)

Continual Invariant Risk Minimization (IRM + CL)

Improved Schemes for Episodic Memory based Lifelong Learning Algorithm

Organizing recurrent network dynamics by task-computation to enable continual learning

Learning with Long-term Remembering: Following the Lead of Mixed Stochastic Gradient

Continual learning with direction-constrained optimization

Continual Learning in Low-rank Orthogonal Subspaces

Bilevel Continual Learning

https://arxiv.org/abs/2007.15553 Another: Bilevel CL

*A Neural Dirichlet Process Mixture Model for Task-Free Continual Learning (ICLR2020)

CL theory:

*Generalisation guarantees for continual learning with orthogonal gradient descent

A Theoretical Analysis of Catastrophic Forgetting through the NTK Overlap Matrix

Orthogonal gradient descent for continual learning

CL and generalization:

*How neural networks find generalizable solutions: Self-tuned annealing in deep learning

CL and neuro-inspired approaches:

Brain-inspired replay for continual learning with artificial neural networks

Continual Learning Through Synaptic Intelligence

Alleviating catastrophic forgetting using context-dependent gating and synaptic stabilization