Catastrophic Importance of Catastrophic Forgetting

TL;DR

This paper explores the potential advancements in artificial neural networks once the challenge of catastrophic forgetting is addressed, including a simple model and reinforcement learning applications.

Contribution

It introduces a simple model and applies existing methods to reinforcement learning to demonstrate progress beyond catastrophic forgetting.

Findings

Potential for neural networks to advance after overcoming catastrophic forgetting

Proposed simple model for studying neural network behavior

Application of existing methods to reinforcement learning tasks

Abstract

This paper describes some of the possibilities of artificial neural networks that open up after solving the problem of catastrophic forgetting. A simple model and reinforcement learning applications of existing methods are also proposed.