Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Energy Costs and Neural Complexity Evolution in Changing Environments

Sian Heesom-Green, Jonathan Shock, Geoff Nitschke

TL;DR

The paper tests whether changing environments and energy costs shape neural complexity by evolving ANN controllers for RL agents across static to highly seasonal environments. Using NEAT with lifetime learning and PPO, it measures ANN size ($N_S$) and structural complexity ($N_C$) under two energy regimes (NEC and EC). Results show that environmental change increases neural complexity only when energy costs are imposed, with highly seasonal environments pushing toward smaller networks, supporting the Expensive Brain Hypothesis; structural complexity appears largely as a byproduct of network size. The findings have implications for understanding energy-efficient brain evolution and guiding energy-constrained robotic controller design in dynamic environments.

Abstract

The Cognitive Buffer Hypothesis (CBH) posits that larger brains evolved to enhance survival in changing conditions. However, larger brains also carry higher energy demands, imposing additional metabolic burdens. Alongside brain size, brain organization plays a key role in cognitive ability and, with suitable architectures, may help mitigate energy challenges. This study evolves Artificial Neural Networks (ANNs) used by Reinforcement Learning (RL) agents to investigate how environmental variability and energy costs influence the evolution of neural complexity, defined in terms of ANN size and structure. Results indicate that under energy constraints, increasing seasonality led to smaller ANNs. This challenges CBH and supports the Expensive Brain Hypothesis (EBH), as highly seasonal environments reduced net energy intake and thereby constrained brain size. ANN structural complexity primarily emerged as a byproduct of size, where energy costs promoted the evolution of more efficient networks. These results highlight the role of energy constraints in shaping neural complexity, offering in silico support for biological theory and energy-efficient robotic design.

Energy Costs and Neural Complexity Evolution in Changing Environments

TL;DR

The paper tests whether changing environments and energy costs shape neural complexity by evolving ANN controllers for RL agents across static to highly seasonal environments. Using NEAT with lifetime learning and PPO, it measures ANN size () and structural complexity () under two energy regimes (NEC and EC). Results show that environmental change increases neural complexity only when energy costs are imposed, with highly seasonal environments pushing toward smaller networks, supporting the Expensive Brain Hypothesis; structural complexity appears largely as a byproduct of network size. The findings have implications for understanding energy-efficient brain evolution and guiding energy-constrained robotic controller design in dynamic environments.

Abstract

The Cognitive Buffer Hypothesis (CBH) posits that larger brains evolved to enhance survival in changing conditions. However, larger brains also carry higher energy demands, imposing additional metabolic burdens. Alongside brain size, brain organization plays a key role in cognitive ability and, with suitable architectures, may help mitigate energy challenges. This study evolves Artificial Neural Networks (ANNs) used by Reinforcement Learning (RL) agents to investigate how environmental variability and energy costs influence the evolution of neural complexity, defined in terms of ANN size and structure. Results indicate that under energy constraints, increasing seasonality led to smaller ANNs. This challenges CBH and supports the Expensive Brain Hypothesis (EBH), as highly seasonal environments reduced net energy intake and thereby constrained brain size. ANN structural complexity primarily emerged as a byproduct of size, where energy costs promoted the evolution of more efficient networks. These results highlight the role of energy constraints in shaping neural complexity, offering in silico support for biological theory and energy-efficient robotic design.

Paper Structure

This paper contains 26 sections, 2 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Research Objective:
  3. Methods
  4. Task Environment
  5. Agent Rewards:
  6. Food Colors and Seasonal Changes:
  7. Lifetime Learning via Reinforcement Learning
  8. Agent Neuro-evolution
  9. Task Performance Metric: Net Energy Intake
  10. Rationale for Energy Budget Design
  11. Neural Complexity Metrics
  12. ANN size ($\mathbf{N_S}$):
  13. ANN structural complexity ($\mathbf{N_C}$):
  14. Energy Costs
  15. Experiments
  16. ...and 11 more sections

Figures (3)

  • Figure 1: 20×20 grid-world. Cell color indicates food type (edible or poisonous). The agent is the white circle. The legend shows seasonal food-color mappings.
  • Figure 2: Left: Box plots for neural complexity metrics and task performance. The x-axis represents the environments, labeled by the number of seasons, from the static 1-season environment to the most dynamic 4-season environment. The y-axis displays neural complexity metrics / task performance for the fittest evolved ANNs from the final generation (fitness evaluated after lifetime learning using the same evaluation seeds as during evolution), averaged over 20 runs. Right: Neural complexity metrics and task performance over evolutionary time of the current fittest genome, averaged over 20 runs. (NEC = No Energy Costs; EC = Energy Costs)
  • Figure 3: Network size ($N_S$) versus structural complexity ($N_C$) over generations. Grey points and crosses are random-walk evolution without and with an ANN size penalty, respectively. Colored markers show task-based evolution (1-4 season environments). Overlap across conditions suggests $N_C$ primarily reflects evolved $N_S$ given mutation rather than task driven selection.