Modelling zebrafish collective behaviours with multilayer perceptrons optimised by evolutionary algorithms

Abstract

Collective movements are pervasive behaviours among social organisms and have led to the development of many models. However, modelling animal trajectories and social interactions in simple bounded environments remains a challenge. Moreover, advances in the understanding of the sensory-motor loop and the information processing by animals are leading to revisions of the traditional assumptions made in decision-making algorithms. In this context, we develop a methodology based on artificial neural networks (ANN) to describe the collective motion of small zebrafish groups in a bounded environment. Although ANN models are commonly used in artificial systems they are still under-explored to model animal collective behaviours. Here, we present a methodology to calibrate Multilayer Perceptrons by learning from real fish experimental data. The ANNs are trained using either supervised learning or various forms of evolutionary reinforcement learning methods (using the CMA-ES and NSGA-III algorithms). We reveal that ANN models trained using evolutionary methods are capable of generating realistic collective motions for groups of 5 zebrafish including the tank wall effects, a feature that is lacking in previous models. Finally, we also discuss the benefits of optimised ANNs as candidates for driving robotic lure with biologically realistic behaviour, a method that is becoming increasingly popular to gather data and validate assumptions on collective behaviours.

Automatic Calibration of Artificial Neural Networks for Zebrafish Collective Behaviours Using a Quality Diversity Algorithm

“…We apply the same experimental method, fish handling and set-up as in [6], [29], [7], [8]. During experiments, fish are placed in an immersed square white plexiglass arena of 1000 × 1000 × 100 mm.…”

“…Artificial neural networks (ANN) are universal function approximators able to model phenomena with a priori information. They were used in previous studies [7], [8], [20] to model fish collective behaviour and generate biomimetic trajectories of fish in groups. However this problem is challenging, and it is still possible to improve upon the biomimetism of resulting trajectories.…”

“…However this problem is challenging, and it is still possible to improve upon the biomimetism of resulting trajectories. Our methodology builds on Cazenille et al [8] and calibrates Multilayer Perceptron (MLP) [2] artificial neural networks to drive simulated fish-like agents in groups of 5 individuals. All simulations involve 5 simulated agents driven by the optimised MLP (see workflow on Fig 1).…”

“…They can be employed in a wide variety of modelling and control tasks [26]. As in [7], [8], our approach uses MLP with one hidden layer of 10 neurons with a hyperbolic tangent activation function. We use this simple and limited ANN as a baseline for bench-marking the various optimisation algorithms.…”

“…We do not look for understanding specific collective behaviour mechanisms. In recent works, we explored the use of artificial neural network models (multilayer perceptrons) to generate realistic collective motion and trajectories of a group of five zebrafish arXiv:1907.09209v1 [cs.NE] 22 Jul 2019 in a bounded environment [7], [8]. We compared supervised learning and reinforcement learning techniques to optimise the behaviour of artificial Zebrafish, so that they would match the trajectories obtained from real-world experimental data.…”