Title: Decoding the brain's learning algorithms: From offline population analysis to real-time closed-loop validation

Abstract: Intelligent behavior requires more than learning from experience, it requires learning how to learn. Imagine moving from a city with reliable weather forecasts to one where they are frequently wrong. Checking the forecast each morning no longer works; you must learn that the source itself has become unreliable and adjust your strategy. The brain faces this challenge constantly, detecting when a once-effective strategy no longer matches the ever-changing environment. This ability to flexibly switch between learning strategies, known as meta-learning, is fundamental to intelligent behavior. Yet, we lack the tools to understand how the brain computes and implements it. This project will develop methods to uncover the brain's meta-learning algorithms by combining neuroscience with machine learning and real-time neural decoding. Here, I will record from large neural populations in the anterior cingulate cortex, a region critical for adaptive decision-making, while rats perform a task that requires them to adjust their learning strategies. Using dynamical systems theory, I will identify which learning algorithm the brain is using and precisely when there are switches between strategies. I will then build a platform capable of reading out abstract cognitive states, such as uncertainty, directly from neural activity in real-time. Currently, researchers can show that a brain region is necessary for behavior, but not which computation was lost when it was disrupted as the computational state active at the moment of disruption is not typically known. The decoding platform developed here will lay the technological foundation to remove this bottleneck, enabling future experiments in which neural activity can be perturbed at the precise moment a specific computational state is detected, establishing computational specificity. Together, the project aims to shift the field from correlational questions about what neurons encode, toward falsifiable tests of computations the brain carries out to produce adaptive-behaviors.