Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neuroscience-Inspired Memory Replay for Continual Learning: A Comparative Study of Predictive Coding and Backpropagation-Based Strategies

Goutham Nalagatla, Shreyas Grandhe

TL;DR

This work tackles catastrophic forgetting in continual learning by comparing predictive coding-based generative replay with traditional backpropagation-based replay. It introduces a neuroscience-inspired framework that uses hierarchical predictive coding for generative replay and contrasts it against a VAE-based generator, evaluated on Split-MNIST, Split-CIFAR-10, and Split-CIFAR-100. The key finding is that predictive coding achieves substantially better task retention (average improvement around 15.3% in forgetting) while maintaining competitive transfer, suggesting that biologically plausible learning rules can yield robust continual learning. The study bridges neuroscience and AI, offering principled insights and a path toward more robust, biologically informed continual learning systems.

Abstract

Continual learning remains a fundamental challenge in artificial intelligence, with catastrophic forgetting posing a significant barrier to deploying neural networks in dynamic environments. Inspired by biological memory consolidation mechanisms, we propose a novel framework for generative replay that leverages predictive coding principles to mitigate forgetting. We present a comprehensive comparison between predictive coding-based and backpropagation-based generative replay strategies, evaluating their effectiveness on task retention and transfer efficiency across multiple benchmark datasets. Our experimental results demonstrate that predictive coding-based replay achieves superior retention performance (average 15.3% improvement) while maintaining competitive transfer efficiency, suggesting that biologically-inspired mechanisms can offer principled solutions to continual learning challenges. The proposed framework provides insights into the relationship between biological memory processes and artificial learning systems, opening new avenues for neuroscience-inspired AI research.

Neuroscience-Inspired Memory Replay for Continual Learning: A Comparative Study of Predictive Coding and Backpropagation-Based Strategies

TL;DR

This work tackles catastrophic forgetting in continual learning by comparing predictive coding-based generative replay with traditional backpropagation-based replay. It introduces a neuroscience-inspired framework that uses hierarchical predictive coding for generative replay and contrasts it against a VAE-based generator, evaluated on Split-MNIST, Split-CIFAR-10, and Split-CIFAR-100. The key finding is that predictive coding achieves substantially better task retention (average improvement around 15.3% in forgetting) while maintaining competitive transfer, suggesting that biologically plausible learning rules can yield robust continual learning. The study bridges neuroscience and AI, offering principled insights and a path toward more robust, biologically informed continual learning systems.

Abstract

Continual learning remains a fundamental challenge in artificial intelligence, with catastrophic forgetting posing a significant barrier to deploying neural networks in dynamic environments. Inspired by biological memory consolidation mechanisms, we propose a novel framework for generative replay that leverages predictive coding principles to mitigate forgetting. We present a comprehensive comparison between predictive coding-based and backpropagation-based generative replay strategies, evaluating their effectiveness on task retention and transfer efficiency across multiple benchmark datasets. Our experimental results demonstrate that predictive coding-based replay achieves superior retention performance (average 15.3% improvement) while maintaining competitive transfer efficiency, suggesting that biologically-inspired mechanisms can offer principled solutions to continual learning challenges. The proposed framework provides insights into the relationship between biological memory processes and artificial learning systems, opening new avenues for neuroscience-inspired AI research.

Paper Structure

This paper contains 27 sections, 3 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Continual Learning
  4. Generative Replay
  5. Predictive Coding
  6. Neuroscience-Inspired AI
  7. Methodology
  8. Problem Formulation
  9. Generative Replay Framework
  10. Predictive Coding-Based Generative Replay
  11. Backpropagation-Based Generative Replay
  12. Training Procedure
  13. Experiments
  14. Datasets
  15. Architecture
  16. ...and 12 more sections

Figures (2)

  • Figure 1: Architectural comparison between (left) predictive coding-based and (right) backpropagation-based generative replay. Predictive coding uses local learning rules with hierarchical predictions and error propagation, while backpropagation relies on global gradient computation.
  • Figure 2: Experimental results comparing predictive coding-based and backpropagation-based generative replay. (Left) Task retention performance showing average accuracy after learning each task. (Right) Forward transfer efficiency measuring improvement in learning new tasks. Error bars represent standard deviation across 5 independent runs.