Puneet Jain’s research while affiliated with Brigham Young University and other places

There are powerful tools for modelling swarms that have strong spatial structures like flocks of birds, schools of fish and formations of drones, but relatively little work on developing formalisms for other swarm structures like hub-based colonies doing foraging, maintaining a nest or selecting a new nest site. We present a method for finding low-dimensional representations of swarm state for simulated homogeneous hub-based colonies solving the best-of-N problem. The embeddings are obtained from latent representations of convolution-based graph neural network architectures and have the property that swarm states which have similar performance have very similar embeddings. Such embeddings are used to classify swarm state into binned estimates of success probability and time to completion. We demonstrate how embeddings can be obtained in a sequence of experiments that progressively require less information, which suggests that the methods can be extended to larger swarms in more complicated environments. This article is part of the theme issue ‘The road forward with swarm systems’.

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.

Agent-based simulations and differential equation models have been used to analyze distributed solutions to the best-of-N problem. This paper shows that the best-of-N problem can be also solved using a graph-based formalism that abstractly represents (a) agents and solutions as vertices, (b) individual agent states as graph edges, and (c) agent state dynamics as edge creation (attachment) or deletion (detachment) between agent and solution. The paper identifies multiple candidate attachment and detachment processes from the literature, and then presents a comparative study of how well various processes perform on the best-of-N problem. Results not only identify promising attachment and detachment processes but also identify model parameters that provide probable convergence to the best solution. Finally, processes are identified that may be suitable for the best-M-of-N problem.