The effect of temporal pattern of injury on disability in learning networks

TL;DR

This paper investigates how the rate and pattern of injury over time affect neural network disability, revealing that slow-growing damage results in less impairment than acute damage, aligning with clinical observations.

Contribution

It introduces a comparison between slow-growing and acute damage in neural networks using simple models, highlighting the impact of injury rate on network disability.

Findings

Slow-growing damage causes less network disability than acute damage.

The effect is consistent across three-layer and homeostasis models.

Results align with clinical reports on injury and recovery.

Abstract

How networks endure damage is a central issue in neural network research. This includes temporal as well as spatial pattern of damage. Here, based on some very simple models we study the difference between a slow-growing and acute damage and the relation between the size and rate of injury. Our result shows that in both a three-layer and a homeostasis model a slow-growing damage has a decreasing effect on network disability as compared with a fast growing one. This finding is in accord with clinical reports where the state of patients before and after the operation for slow-growing injuries is much better that those patients with acute injuries.