Neural circuits for learning under perceptual uncertainty

How does the brain learn to make efficient decisions in an uncertain world? A hallmark of biological learning is strategy discovery: over long periods of time, animals can transition through a range of behaviors on their way from naïve to expert performance. Decades of research have highlighted the role of midbrain dopamine (DA) neurons as well as major DA-receiving brain regions, striatum and frontal cortex, in learning. However, previous studies did not probe these neural circuits from the very first trial of the task, while animals test and select among possible strategies for gaining reward under perceptual uncertainty. Moreover, past studies often examined these neural circuits at a small scale and in isolation, and could not systematically investigate how neural signals across these brain regions underlie learning. A large gap thus remains between our understanding of neural computations and the long-term process of strategy discovery under uncertainty. This proposal will combine novel behavioral tasks and advanced neural circuit tools to fill this gap. We will longitudinally measure and manipulate neural signals from naïve to expert performance in mice learning a visual decision task that admits multiple behavioral strategies, and will use computational tools to formalize the relation between neural activity and learning dynamics. The work will address these questions:

1. How do DA neural signals develop during learning under uncertainty, and how do they relate to behavioral strategies?
2. What does DA release across striatum and frontal cortex encode during learning, and does DA in different brain regions play distinct causal roles in learning?
3. How do neural signals across striatum and frontal cortex develop and relate to behavioral strategies during learning?

This work will rectify the paucity of data on the neural bases of learning, generating a step change in our understanding of neural circuits regulating learning under perceptual uncertainty.