# The Potential of Cognitive‐Inspired Neural Network Modeling Framework for Computer Vision

## TL;DR

This paper introduces a new framework that bridges cognitive science and AI by improving vision neural networks with cognitive-inspired modeling.

## Contribution

A cognitive modeling framework (CMF) and memory modeling method (UMA) are proposed to better align vision deep neural networks with human visual cognition.

## Key findings

- VCogM and VCNU achieved state-of-the-art performance on tasks like natural scene recognition and agricultural image classification.
- The model's learning process is independent of data distribution and scale, showing the effectiveness of cognitive-inspired principles.

## Abstract

Vision deep neural networks (VDNNs) only simulate the attention‐based significance selection function in human visual perception, rather than the full spectrum of visual cognition, reflecting the divide between cognitive science (CS) and artificial intelligence (AI). To address this problem, this work proposes a cognitive modeling framework (CMF) comprising three stages: functional abstraction, operator structuring, and program agent. Then, this work defines the prior information of basic image features as the long‐term memory content in VDNNs. Meanwhile, this work introduces a memory modeling method for VDNNs based on the fast Fourier transform (FFT) and statistical methods—the unbiased mapping algorithm (UMA). Finally, this work develops visual cognitive neural units (VCNUs) and a baseline model (VCogM) based on CMF and UMA, and conduct performance testing experiments on different datasets such as natural scene recognition and agricultural image classification. The results show that VCogM and VCNU achieved state‐of‐the‐art (SOTA) performance across various recognition tasks. The model's learning process is independent of data distribution and scale, fully demonstrating the rationality of cognitive‐inspired modeling principles. The research findings provide new insights into the deep integration of CS and AI.

In article number 202507730, Guorun Li, Lei Liu, Yuefeng Du, and co‐workers present a cognitive modeling framework (CMF) to bridge the ‘representation gap’ and ‘conceptual gap’ between cognitive theory and vision deep neural networks (VDNNs). The research findings provide new insights and solid theoretical support for VDNN modeling inspired by cognitive theory.