DeepAI AI Chat
Log In Sign Up

Complementary Learning for Overcoming Catastrophic Forgetting Using Experience Replay

by   Mohammad Rostami, et al.
University of Pennsylvania
HRL Laboratories, LLC

Despite huge success, deep networks are unable to learn effectively in sequential multitask learning settings as they forget the past learned tasks after learning new tasks. Inspired from complementary learning systems theory, we address this challenge by learning a generative model that couples the current task to the past learned tasks through a discriminative embedding space. We learn an abstract level generative distribution in the embedding that allows the generation of data points to represent the experience. We sample from this distribution and utilize experience replay to avoid forgetting and simultaneously accumulate new knowledge to the abstract distribution in order to couple the current task with past experience. We demonstrate theoretically and empirically that our framework learns a distribution in the embedding that is shared across all task and as a result tackles catastrophic forgetting.


page 1

page 2

page 3

page 4


Generative Continual Concept Learning

After learning a concept, humans are also able to continually generalize...

Cognitively Inspired Learning of Incremental Drifting Concepts

Humans continually expand their learned knowledge to new domains and lea...

Meta-Learning with Sparse Experience Replay for Lifelong Language Learning

Lifelong learning requires models that can continuously learn from seque...

Prompt Conditioned VAE: Enhancing Generative Replay for Lifelong Learning in Task-Oriented Dialogue

Lifelong learning (LL) is vital for advanced task-oriented dialogue (ToD...

Generative Memory for Lifelong Reinforcement Learning

Our research is focused on understanding and applying biological memory ...

DynMat, a network that can learn after learning

To survive in the dynamically-evolving world, we accumulate knowledge an...

History Repeats: Overcoming Catastrophic Forgetting For Event-Centric Temporal Knowledge Graph Completion

Temporal knowledge graph (TKG) completion models typically rely on havin...