Convergent Temperature Representations in Artificial and Biological Neural Networks

Abstract

Artificial Neural Networks (ANN) are increasingly successful in solving tasks long considered hallmarks of cognition in Biological Neural Networks (BNN), such as visual discrimination, playing Go and navigation. While the design of ANNs has been inspired by discoveries in BNNs, it is controversial whether both network types utilize the same fundamental principles and hence if ANNs can serve as models of animal cognition. However, if representations and algorithms are shared between BNNs and ANNs, then the analysis of processing in ANNs could lead to fundamental insights into their biological counterparts. Here, we generated and trained a deep convolutional neural network to solve a heat gradient navigation task using the behavioral repertoire of larval zebrafish. We found that these behavioral constraints led to striking similarities in temperature processing and representation in this ANN with biological circuits and neural dynamics underlying heat avoidance in larval zebrafish. This includes stimulus representation in ON and OFF types as well as ANN units showing adapting and sustained responses. Importantly, ANN performance critically relied on units representing temperature in a fish-like manner while other nodes were dispensable for network function. We next used the accessibility of the ANN to uncover new features of the zebrafish BNN. We 25 identified a novel neuronal response type in the zebrafish brain that was predicted by the ANN but escaped detection in previous brain wide calcium imaging experiments. Finally, our approach generalized since training the same ANN constrained by the C. elegans motor repertoire resulted in distinct neural representations that match closely features observed in the worm. Together, these results emphasize convergence of ANNs and BNNs on canonical representations and that man made ANNs form a powerful tool to understand their biological counterparts.