Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evolutionary Dispersal of Ecological Species via Multi-Agent Deep Reinforcement Learning

Wonhyung Choi, Inkyung Ahn

TL;DR

This study uses multi-agent reinforcement learning with deep Q-networks to simulate single species and predator-prey interactions, incorporating SDD-type rewards, and adopts starvation-driven diffusion models as nonlinear diffusion to describe species dispersal based on local resource conditions.

Abstract

Understanding species dynamics in heterogeneous environments is essential for ecosystem studies. Traditional models assumed homogeneous habitats, but recent approaches include spatial and temporal variability, highlighting species migration. We adopt starvation-driven diffusion (SDD) models as nonlinear diffusion to describe species dispersal based on local resource conditions, showing advantages for species survival. However, accurate prediction remains challenging due to model simplifications. This study uses multi-agent reinforcement learning (MARL) with deep Q-networks (DQN) to simulate single species and predator-prey interactions, incorporating SDD-type rewards. Our simulations reveal evolutionary dispersal strategies, providing insights into species dispersal mechanisms and validating traditional mathematical models.

Evolutionary Dispersal of Ecological Species via Multi-Agent Deep Reinforcement Learning

TL;DR

This study uses multi-agent reinforcement learning with deep Q-networks to simulate single species and predator-prey interactions, incorporating SDD-type rewards, and adopts starvation-driven diffusion models as nonlinear diffusion to describe species dispersal based on local resource conditions.

Abstract

Understanding species dynamics in heterogeneous environments is essential for ecosystem studies. Traditional models assumed homogeneous habitats, but recent approaches include spatial and temporal variability, highlighting species migration. We adopt starvation-driven diffusion (SDD) models as nonlinear diffusion to describe species dispersal based on local resource conditions, showing advantages for species survival. However, accurate prediction remains challenging due to model simplifications. This study uses multi-agent reinforcement learning (MARL) with deep Q-networks (DQN) to simulate single species and predator-prey interactions, incorporating SDD-type rewards. Our simulations reveal evolutionary dispersal strategies, providing insights into species dispersal mechanisms and validating traditional mathematical models.

Paper Structure

This paper contains 12 sections, 3 theorems, 19 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Starvation-driven diffusion
  4. Framework
  5. Learning with a deep Q-Network
  6. Results
  7. Single species model
  8. Predator--prey model
  9. Conclusion
  10. Mathematical models with SDD
  11. Ideal free distribution
  12. Predator-prey model

Key Result

Theorem A.1

Let $\theta_\gamma$ be the solution of single. Suppose that $m(x)$ satisfies Then, $\theta_\gamma(x)$ converges to $m(x)$ as $\varepsilon\to0$.

Figures (6)

  • Figure 1: Discrete action space
  • Figure 2: Food resource distribution. The total amount of food resources is 1500 (500 in the left patch and 1000 in the right patch)
  • Figure 3: The number of populations in each resource patch (the red line is the number of species in the right-side patch, and the blue line is the number of species in the left-side patch).
  • Figure 4: Distribution of food resources in the grid (total amount $N_{food}=2000$).
  • Figure 5: Simulation results for the predator-prey ecosystem (a) without learning about both species (b) with only learning about the prey
  • ...and 1 more figures

Theorems & Definitions (3)

  • Theorem A.1
  • Theorem A.2
  • Theorem A.3