Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evolved Developmental Artificial Neural Networks for Multitasking with Advanced Activity Dependence

Yintong Zhang, Jason A. Yoder

TL;DR

The paper addresses multitask learning in artificial neural networks by combining developmentally guided growth with extended activity-dependent plasticity (AD). Building on Cartesian Genetic Programming-based developmental models, it adds AD mechanisms that regulate soma parameters—bias, health, and position—and evaluates their impact on solving a reinforcement learning task (CartPole) and a classification task (Bank Authentication). The study finds that AD on health and position improves performance over the base model, and that combining AD across all soma parameters yields the strongest gains, suggesting synergistic benefits for multitask development. These findings highlight the potential of evolvable AD in developmental neuroevolution and point to future work on dendrite AD, evolved AD configurations, and efficiency enhancements for scalable exploration.

Abstract

Recently, Cartesian Genetic Programming has been used to evolve developmental programs to guide the formation of artificial neural networks (ANNs). This approach has demonstrated success in enabling ANNs to perform multiple tasks while avoiding catastrophic forgetting. One unique aspect of this approach is the use of separate developmental programs evolved to regulate the development of separate soma and dendrite units. An opportunity afforded by this approach is the ability to incorporate Activity Dependence (AD) into the model such that environmental feedback can help to regulate the behavior of each type of unit. Previous work has shown a limited version of AD (influencing neural bias) to provide marginal improvements over non-AD ANNs. In this work, we present promising results from new extensions to AD. Specifically, we demonstrate a more significant improvement via AD on new neural parameters including health and position, as well as a combination of all of these along with bias. We report on the implications of this work and suggest several promising directions for future work.

Evolved Developmental Artificial Neural Networks for Multitasking with Advanced Activity Dependence

TL;DR

The paper addresses multitask learning in artificial neural networks by combining developmentally guided growth with extended activity-dependent plasticity (AD). Building on Cartesian Genetic Programming-based developmental models, it adds AD mechanisms that regulate soma parameters—bias, health, and position—and evaluates their impact on solving a reinforcement learning task (CartPole) and a classification task (Bank Authentication). The study finds that AD on health and position improves performance over the base model, and that combining AD across all soma parameters yields the strongest gains, suggesting synergistic benefits for multitask development. These findings highlight the potential of evolvable AD in developmental neuroevolution and point to future work on dendrite AD, evolved AD configurations, and efficiency enhancements for scalable exploration.

Abstract

Recently, Cartesian Genetic Programming has been used to evolve developmental programs to guide the formation of artificial neural networks (ANNs). This approach has demonstrated success in enabling ANNs to perform multiple tasks while avoiding catastrophic forgetting. One unique aspect of this approach is the use of separate developmental programs evolved to regulate the development of separate soma and dendrite units. An opportunity afforded by this approach is the ability to incorporate Activity Dependence (AD) into the model such that environmental feedback can help to regulate the behavior of each type of unit. Previous work has shown a limited version of AD (influencing neural bias) to provide marginal improvements over non-AD ANNs. In this work, we present promising results from new extensions to AD. Specifically, we demonstrate a more significant improvement via AD on new neural parameters including health and position, as well as a combination of all of these along with bias. We report on the implications of this work and suggest several promising directions for future work.
Paper Structure (7 sections, 3 figures)

This paper contains 7 sections, 3 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Neural Network Model
  4. Methodology
  5. Results
  6. Discussion
  7. Conclusion and future work

Figures (3)

  • Figure 1: CGP ANN overview in which soma and dendrite refer, respectively, to nodes and connections in an artificial neural network.
  • Figure 2: Inputs and Outputs for the Soma and Dendrite Program. The inputs come from parameters of nodes and connections in an ANN, and the outputs are fed back to the network as updated values for the same parameters.
  • Figure 3: Comparing Performance for different Activity Dependence (AD) models. The base model includes no AD, and the combined model includes AD of all soma's parameters (bias, health, and position). Each experimental group was run 50 times and the lines show the mean of the best individuals from each run. The shaded area represents the standard error for each of the fitness lines.