Continual Backprop: Stochastic Gradient Descent with Persistent Randomness

TL;DR

This paper identifies performance degradation in traditional Backprop during continual learning and introduces Continual Backprop, an algorithm that maintains learning ability over time by injecting persistent randomness.

Contribution

The paper presents Continual Backprop, a novel extension of Backprop that sustains continual learning by injecting persistent randomness, addressing degradation issues.

Findings

Backprop performance degrades over time in continual learning.

Continual Backprop maintains adaptability in supervised and reinforcement learning.

Continual Backprop has similar computational complexity to standard Backprop.

Abstract

The Backprop algorithm for learning in neural networks utilizes two mechanisms: first, stochastic gradient descent and second, initialization with small random weights, where the latter is essential to the effectiveness of the former. We show that in continual learning setups, Backprop performs well initially, but over time its performance degrades. Stochastic gradient descent alone is insufficient to learn continually; the initial randomness enables only initial learning but not continual learning. To the best of our knowledge, ours is the first result showing this degradation in Backprop's ability to learn. To address this degradation in Backprop's plasticity, we propose an algorithm that continually injects random features alongside gradient descent using a new generate-and-test process. We call this the \textit{Continual Backprop} algorithm. We show that, unlike Backprop, Continual Backprop is able to continually adapt in both supervised and reinforcement learning (RL) problems. Continual Backprop has the same computational complexity as Backprop and can be seen as a natural extension of Backprop for continual learning.