How do animals learn the structure of their natural environment?

Cornell University, Ithaca NY

Investigators

Linked publications, trials & patents

Abstract

Project Summary/Abstract Most animals live in complex, changing environments. In order to search for food, shelter or to scape threats they need to learn how to navigate those environments in an adaptative manner. Since natural environments do not change in a complete random manner, it is possible to identify regularities, allowing animals to predict such changes. The ability to abstract latent structured relationships in the environment is known as structure learning. This ability is one of the fundamental aspects of intelligence, allowing to generalize and make inferences beyond one’s experience. However, its neural mechanisms are not known. The main goal of this proposal is to elucidate the neural circuit mechanisms of structure learning using rat foraging and social behavior as a model. As an animal interacts with its environment or other conspecifics, neurons in the hippocampus and associated cortical areas representing the same external variable fire together forming a functional assembly. The sequential activation of assemblies offers a mechanism to encode a relational map of the world that combines spatial, social and other types of information and can be flexibly reconfigured to track changes. This coding scheme is also predictive, since the activation of an initial assembly in the sequence can recruit subsequent ones, anticipating the occurrence of future events. For neuronal sequences to be able to support structure learning they also need to offer a means to perform inferences and generalize to new situations. This process involves identifying underlying principles form experience and applying them to novel situations. I will test the hypothesis that neuronal sequences are a mechanism that supports structure learning and inference through generalization. In their natural environments, rats live in large colonies and forage over extended areas, a complexity that is not captured by common laboratory assays. If different cells would be necessary to encode each contingency experienced by an animal in its natural environment, as the dominant parading in the field proposes, it would require more neurons that its brain has. A way to solve this problem, is to use structure learning to generalize common latent features and discard irrelevant information. The proposed work will investigate the neural circuit mechanisms that support the ability of animals to learn the latent structure of their natural environments by constructing internal predictive models and generalizing, and how they use such representations to guide flexible behavior. We will solve the two main obstacles that have prevented progress on these questions. One is the need for long-term stable recordings of neurons across brain areas together with specific manipulations of their interactions, without restricting animal behavior in large spaces or while interacting with other conspecifics. The second is to develop behavioral paradigms that capture the complexity of social and foraging behavior in natural environments and are amenable to neural recordings. We will deploy several technical innovations to overcome current limitations (AIM 0) and apply them to determine the neural circuit mechanisms that support socio-spatial structure learning in rats (AIM1). In AIM2 we will perform neural recordings in rats foraging in large outdoor enclosures to determine how the mechanisms identified in laboratory settings translate to more natural conditions. In AIM 3 we will investigate how rats learn complex social structures and how this affects their foraging behavior in naturalistic outdoor environments.