Lp-Convolution: Brain-inspired AI improves accuracy/ efficiency of machine vision image recognition while reducing computational burden

https://www.eurekalert.org/news-releases/1081245

"multivariate p-generalized normal distribution reshapes Convolutional Neural Network filters dynamically (stretching horizontally or vertically based on task, like how human brain selectively focuses on relevant details), unlike traditional method that uses fixed square filters... increasing filter size to 7×7 kernels or larger usually does not improve performance despite adding more parameters, overcome using flexible, connectivity patterns focusing on most relevant parts of image... also robust against corrupted data"

Related:

All-topographic neural networks more closely mimic the human visual system
https://techxplore.com/news/2025-06-topographic-neural-networks-mimic-human.html

Researchers develop two-layer neural model that matches complex visual processing in the brain
https://medicalxpress.com/news/2025-07-layer-neural-complex-visual-brain.html