“Toward a Unified Theory of Neural Coding for Interpreting Brain-Wide Data”

Gaia Tavoni, PhD
Assistant Professor of Neuroscience
WashU Medicine

Large-scale, high-resolution neural recordings allow us to analyze neural activity simultaneously across sensory, decision-making and motor systems. Yet these systems are typically studied in isolation, leaving open the central question of how distributed brain networks jointly encode, integrate and predict information to guide behavior. At the same time, theoretical accounts of neural coding often assume distinct computational objectives — maximizing information transmission (efficient coding), minimizing prediction errors (predictive coding) or optimizing performance for specific tasks (task-driven coding). How these frameworks relate to one another and whether a single unifying principle can explain neural computation across brain areas and behavioral contexts remains unclear.

One line of research in my lab aims to establish a unified mathematical theory of neural coding that addresses these gaps. In this talk, I will present a framework we recently developed that generalizes efficient coding to multimodal networks — circuits that integrate independent streams of sensory and motor information. This theory: (a) provides a unified account of diverse multisensory and sensorimotor phenomena, including motor influences on auditory processing and cross-modal interactions among visual, auditory, somatosensory and olfactory systems; (b) shows that predictive computations arise naturally from efficient codes in low-noise regimes, linking two major coding principles within a single mathematical framework; (c) uncovers a circuit-level algorithm for predictive coding that differs fundamentally from previous proposals (for which empirical evidence has so far been limited), and situates predictive coding within a broader interpretative landscape; and (d) recovers unimodal coding as a limiting case, connecting classical results to a more general multimodal formulation. Finally, I will show how this theory maps onto biologically plausible learning dynamics in sparse neural networks.

Together, these results establish a foundation that we will extend to additional circuit motifs, noise regimes and behavioral contexts in pursuit of a general theory of neural coding that bridges normative objectives, algorithmic mechanisms and circuit-level implementations across the brain.