Incorporating Visual Cortical Lateral Connection Properties into CNN: Recurrent Activation and Excitatory-Inhibitory Separation

2025-09-18

AI Summary

Existing CNNs (e.g., ResNet), though inspired by the mammalian visual system, lack biologically grounded lateral connectivity within cortical layers, limiting both biological plausibility and representational capacity. To address this, we propose a biologically inspired recurrent convolutional network that augments standard feedforward CNNs with weight-shared local recurrent connections to explicitly model intra-feature-map lateral interactions. We design a custom loss function enabling separable learning of excitatory and inhibitory connections while preserving the original feedforward pathway and superimposing local feedback dynamics. Experiments demonstrate significant accuracy improvements on image classification tasks and spontaneous emergence of neurobiologically plausible activation patterns—including orientation selectivity and center-surround antagonism—as well as biologically consistent connection properties. This work constitutes the first systematic integration of cortex-inspired excitatory–inhibitory separation in lateral connectivity into CNN architecture, advancing artificial vision systems toward greater biological interpretability.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Bio-inspired Learning
• Computer Vision: Multi-modal Vision

Application Category

• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Search and Retrieval-Augmented AI: Assisted, interactive, and conversational search
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

The original Convolutional Neural Networks (CNNs) and their modern updates such as the ResNet are heavily inspired by the mammalian visual system. These models include afferent connections (retina and LGN to the visual cortex) and long-range projections (connections across different visual cortical areas). However, in the mammalian visual system, there are connections within each visual cortical area, known as lateral (or horizontal) connections. These would roughly correspond to connections within CNN feature maps, and this important architectural feature is missing in current CNN models. In this paper, we present how such lateral connections can be modeled within the standard CNN framework, and test its benefits and analyze its emergent properties in relation to the biological visual system. We will focus on two main architectural features of lateral connections: (1) recurrent activation and (2) separation of excitatory and inhibitory connections. We show that recurrent CNN using weight sharing is equivalent to lateral connections, and propose a custom loss function to separate excitatory and inhibitory weights. The addition of these two leads to increased classification accuracy, and importantly, the activation properties and connection properties of the resulting model show properties similar to those observed in the biological visual system. We expect our approach to help align CNN closer to its biological counterpart and better understand the principles of visual cortical computation.

Problem

• Modeling lateral cortical connections within CNN architecture
• Incorporating recurrent activation and excitatory-inhibitory separation
• Aligning CNN properties with biological visual system observations

Innovation

• Recurrent CNN with weight sharing
• Custom loss for excitatory-inhibitory separation
• Lateral connections within feature maps