Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Performance Prediction of Hub-Based Swarms

Puneet Jain, Chaitanya Dwivedi, Vigynesh Bhatt, Nick Smith, Michael A Goodrich

TL;DR

The paper tackles the challenge of predicting performance in hub-based swarms for the best-of-$N$ problem, where the collective state space grows rapidly with agent count. It introduces a graph-based representation of the ABM, encodes collective-state dynamics as a probabilistic graph, and learns low-dimensional embeddings via a GraphSAGE-based encoder. The authors demonstrate that tensor-derived collective states cluster by their likelihood of success and that 3D embeddings reveal trajectories and basins of attraction predictive of swarm performance. This approach enables scalable analysis, clustering, and potential regulation of swarm behavior across different agent and site configurations, with implications for planning and control in large-scale bio-inspired systems.

Abstract

A hub-based colony consists of multiple agents who share a common nest site called the hub. Agents perform tasks away from the hub like foraging for food or gathering information about future nest sites. Modeling hub-based colonies is challenging because the size of the collective state space grows rapidly as the number of agents grows. This paper presents a graph-based representation of the colony that can be combined with graph-based encoders to create low-dimensional representations of collective state that can scale to many agents for a best-of-N colony problem. We demonstrate how the information in the low-dimensional embedding can be used with two experiments. First, we show how the information in the tensor can be used to cluster collective states by the probability of choosing the best site for a very small problem. Second, we show how structured collective trajectories emerge when a graph encoder is used to learn the low-dimensional embedding, and these trajectories have information that can be used to predict swarm performance.

Performance Prediction of Hub-Based Swarms

TL;DR

The paper tackles the challenge of predicting performance in hub-based swarms for the best-of- problem, where the collective state space grows rapidly with agent count. It introduces a graph-based representation of the ABM, encodes collective-state dynamics as a probabilistic graph, and learns low-dimensional embeddings via a GraphSAGE-based encoder. The authors demonstrate that tensor-derived collective states cluster by their likelihood of success and that 3D embeddings reveal trajectories and basins of attraction predictive of swarm performance. This approach enables scalable analysis, clustering, and potential regulation of swarm behavior across different agent and site configurations, with implications for planning and control in large-scale bio-inspired systems.

Abstract

A hub-based colony consists of multiple agents who share a common nest site called the hub. Agents perform tasks away from the hub like foraging for food or gathering information about future nest sites. Modeling hub-based colonies is challenging because the size of the collective state space grows rapidly as the number of agents grows. This paper presents a graph-based representation of the colony that can be combined with graph-based encoders to create low-dimensional representations of collective state that can scale to many agents for a best-of-N colony problem. We demonstrate how the information in the low-dimensional embedding can be used with two experiments. First, we show how the information in the tensor can be used to cluster collective states by the probability of choosing the best site for a very small problem. Second, we show how structured collective trajectories emerge when a graph encoder is used to learn the low-dimensional embedding, and these trajectories have information that can be used to predict swarm performance.
Paper Structure (21 sections, 1 equation, 8 figures, 2 tables)

This paper contains 21 sections, 1 equation, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Graphs for the Best-of-N Problem
  4. Best-of-N Problem
  5. Agent Based Model (ABM)
  6. Representing Collective State as a Tensor
  7. Representing State Dynamics as A Graph
  8. Information in the Tensor
  9. Low-Dimensional Embeddings
  10. Input State Tensor for Embeddings
  11. Graph Convolutional Neural Network
  12. Loss Function
  13. Experiment Design
  14. Experiment Conditions
  15. ABM Results
  16. ...and 6 more sections

Figures (8)

  • Figure 1: Best-of-N problem with two sites (circles) and 50 agents (triangles). The hexagon represents the hub.
  • Figure 2: Agent-Based Model
  • Figure 3: Example trajectories generated from 1500 trials for a problem with ten agents and two sites. Each point represents a unique tensor, and each arrow is the edge between tensors. The visualization is made using the graphviz visualization method in networkx.
  • Figure 4: Clustering of 2D embedding using t-SNE hinton2002stochastic, and correlation of cluster with success probability. There were 10 agents, two sites ($q(s_1)=1$, $q(s_2)=0.5$) and a quorum thresold of 2 agents. Bernoulli parameters were set based on the mean times in a state: $O$ 8sec, $A$ 3sec, $R$$6q(s)$sec. The number of reassessing trips was $\propto 3q(s)$. The Bernoulli parameter for being recruited by a single recruiting agent was 40sec. Explore agents used $y=\delta(D/2)$ so agents deterministically stopped exploring when they reached half the world dimension. Sites were placed at $D/4$ from the hub.
  • Figure 5: Encoder architecture: i is the input dimension, h is the hidden dimension, o is the output dimension, and f = ReLU is the activation function.
  • ...and 3 more figures