Harnessing cortical geometry, wiring, and function as inductive biases for recurrent neural networks

2026-06-12

AI Summary

This study proposes a biologically inspired recurrent neural network (RNN) architecture designed to enhance both learning efficiency and neurobiological plausibility. Leveraging the MICrONS dataset—comprising spatial coordinates, anatomical connectivity, and functional responses of nearly 12,000 excitatory neurons in the mouse visual cortex—the authors integrate real cortical geometry, wiring principles, and functional organization as holistic inductive biases into the RNN. Specifically, they employ functionally driven weight initialization derived from co-registered calcium imaging and electron microscopy reconstructions, alongside spatially aware communication constraints and non-negative weight restrictions. The resulting model significantly outperforms baseline architectures across three cognitive decision-making tasks while exhibiting brain-like network properties, including low entropy, modularity, and small-world topology, thereby demonstrating dual advantages in performance and biological fidelity.

Abstract

How the wiring and functional organization of cortex shape recurrent computation remains a central question in both neuroscience and machine learning. Here, we leverage data released through the Machine Intelligence from Cortical Networks (MICrONS) program--a functional connectomics resource spanning multiple areas of mouse visual cortex, in which dense calcium imaging is co-registered with high-resolution electron microscopy reconstruction from the same animal--to build biologically grounded recurrent neural networks. Using neuronal spatial coordinates, anatomical connectivity, and function-derived relationships from nearly 12,000 coregistered excitatory neurons, we initialize recurrent weights and impose communication-aware spatial constraints during learning. Across three cognitive decision-making tasks, networks constrained by cortical structure and function consistently outperform baseline and partially constrained models. Functional weight initialization provides the largest gain, while real spatial embedding yields robust additional improvements across conditions. These biologically grounded networks also develop low-entropy, modular, and small-world organization, and retain strong performance even when recurrence is restricted to positive weights. Together, our results show that the machinery of cortex--its geometry, wiring, and functional structure--can be harnessed as a powerful inductive basis for building recurrent networks that learn more effectively while converging toward key organizational principles of biological computation.

Problem

• cortical geometry
• wiring
• functional organization
• recurrent neural networks
• inductive biases

Innovation

• cortical geometry
• functional connectomics
• inductive bias
• recurrent neural networks
• biologically grounded AI