Title: Cortical mechanism of continual learning.

Human and other animals can learn and execute many different tasks throughout their lifespan, a process known as continual learning. This natural ability, however, challenges most artificial neural networks (ANN), which face the problem of catastrophic forgetting: learning of new tasks can interfere with performance on previously learned tasks due to conflicts in task representations. How can a network, biological or artificial, learn multiple tasks while preventing catastrophic forgetting? ANN research suggests that unique information about new tasks can be encoded through expansion of representation, while shared information between old and new tasks can be integrated into shared representation. These solutions provide testable hypotheses for how the brain avoids catastrophic forgetting. We thus propose to investigate how neural system solve the problem of continual learning. We trained mice to perform two distinct auditory dsicrimination tasks. We will use two-photon calcium imaging to track activity of L2/3 pyramidal cells in the auditory cortex (AC) and the posterior parietal cortex (PPC) at single-cell resolution throughout learning and aim to characterize the evolution of neural representation along two independent dimensions: representational expansion and integration. Furthermore, we ask if behaviorally manipulating the learning configuration of multi-task learning alters both behavioral performance and the underlying neural representation of multiple tasks. Together, this study will shed light on how continual learning is achieved in the brain through dynamic evolution of neural representations, how manipulation of the learning process can alter these neural representations, and inspire new solutions to continual learning for artificial neural network research.