Harnessing behavioral diversity to understand neural computations for cognition

Abstract

With the increasing acquisition of large-scale neural recordings comes the challenge of inferring the computations they perform and understanding how these give rise to behavior. Here, we review emerging conceptual and technological advances that begin to address this challenge, garnering insights from both biological and artificial neural networks. We argue that neural data should be recorded during rich behavioral tasks, to model cognitive processes and estimate latent behavioral variables. Careful quantification of animal movements can also provide a more complete picture of how movements shape neural dynamics and reflect changes in brain state, such as arousal or stress. Artificial neural networks (ANNs) could serve as artificial model organisms to connect neural dynamics and rich behavioral data. ANNs have already begun to reveal how a wide range of different behaviors can be implemented, generating hypotheses about how observed neural activity might drive behavior and explaining diversity in behavioral strategies.

Figure 1

Animals can exhibit a diverse range of behaviors and strategies even when solving the same task. Insight into this diversity might come from increasing task complexity, detailed quantification of animal behavior and examination of ANNs trained to solve the same problem. Image by Julia Kuhl.

Figure 2

A candidate workflow schematizing how new experimental approaches (black boxes) and analyses (grey boxes) can be combined to gain insight into neural computations that underlie behavior. A task-based approach is used to generate rich behavioral and neural data reflecting the animal’s task performance. Workflow with cognitive models, (right): Latent behavioral variables are estimated from different task features, using explicit models of cognitive computations. An observational approach measures ongoing movements and brain state. Variability in neural dynamics is explained by combining movements, brain state and latent behavioral variables. Workflow with ANNs (left): multiple ANNs are trained on the same task, using different cost functions and hyperparameters, such as network connectivity or those controlling the complexity of the dynamics. ANNs function as artificial model organisms with slight variations to discover possible network implementations that support a cognitive computation. The scientist has access to the underlying artificial neural dynamics and ANN parameters, from which computational variables can be reverse-engineered to interpret an animal’s neural data. These computational variables can be similar to those derived from cognitive models, but apply to a broader class of tasks where explicit cognitive models may not be available or tractable.