Abstract

Most patients with functional impairments following brain damage show some degree of recovery over time—a pattern observed across language, motor, and visual domains. Yet, recovery remains highly variable and depends on lesion characteristics. What principles govern this variability? In this talk, I will present a computational approach to understanding how neural circuits reorganise to compensate for damage. Using generative models that mirror neural architectures, we can simulate specific lesion patterns and predict their functional outcomes. I will focus on speech processing, where our work demonstrates how precision-weighting between primary and alternative neural circuits shapes recovery trajectories, enabling paradoxical recovery through circuit rebalancing. I will then show how these insights align with our neuroimaging experiment, revealing that neural circuits flexibly engage different pathways even for basic cognitive operations like speech processing. Building on these findings, I will discuss our recent efforts to develop domain-general models of perceptual function. For this, we have formalised how information can be mapped across sensory modalities using optimal transport theory. This approach provides a quantitative framework to study cross-modal plasticity and the emergence of compensatory mechanisms through neural circuit interactions. Ultimately, this lays the groundwork for understanding brain-wide damage and functional resilience, and generating testable predictions that can guide future neuroimaging studies and therapeutic approaches.