In ever-changing environments, where animals encounter diverse stimuli, flexibility is crucial for survival. This flexibility encompasses the ability to adapt by manipulating inputs and modifying behavior. In this thesis, we demonstrate that 'flexibility' is a pervasive property found at multiple levels of brain organization. It is inherent to individual neurons and also manifests in neural networks, contributing to adaptive behavior.

We begin by investigating the substrate of computations, by exploring how neurons represent their inputs. Our findings reveal that auditory neurons possess composite receptive fields, a capability that enables them to process complex stimuli effectively. We then examine how neurons combine their inputs, revealing their capacity for flexible computations. In these two studies, we probe the neuron with natural stimuli, which are ethologically relevant to the animal in its environment.

The ability to process and operate over information flexibly should confer a survival advantage to the animal. Expanding our scope from single neurons to behaviour, and from instantaneous computations to learning from experiences over longer timescales, we delve into the neural bases of flexible behaviour within a reward-learning framework. We show that the control of neuromodulators over synaptic plasticity enhances reversal learning for agents navigating reward changes.

These investigations, which span single-neuron computations to complex behaviours, offer insight into the neural mechanisms for adaptive responses and representations for navigating a dynamic environment.