Innovative Two-Layer Neural Model Mimics Brain's Complex Visual Processing

Recent advancements in neuroscience have led to the development of a streamlined two-layer computational model that replicates the intricate way our brain processes visual information. Researchers from the Stringer and Pachitariu labs at Janelia have focused on understanding how individual neurons in the primary visual cortex— the first brain region to analyze visual data—encode details to distinguish objects such as leaves from rocks.

Building on the challenge of creating simple yet accurate models of neural activity, the team recorded neural responses from over 29,000 neurons in a mouse’s visual cortex. The mice viewed up to 65,000 natural texture images, including leaves and rocks. Using this extensive data, scientists tested various models to identify the minimal architecture capable of predicting neuron responses accurately.

They found a remarkably efficient model that could capture approximately 75% of the visual information processed by neurons. Unlike previous models that required multiple complex layers, this model achieved high performance with only two layers. By expanding the size of each layer—making them wider—and increasing the receptive field of each artificial neuron, the team demonstrated that a simplified model could replicate the complex feature extraction performed by the brain.

Furthermore, they introduced the concept of 'minimodels' for individual neurons, where smaller, interpretable models explain the specific visual feature selectivity of each neuron. These minimodels match the predictive power of larger, more complicated models and enable clearer understanding of visual computations at the single-neuron level.

Lead researcher Fengtong Du emphasized that this simplified approach not only advances our understanding of visual processing but also provides a more interpretable framework for studying neural responses. This research paves the way for better insights into how the brain implements complex visual functions using streamlined neural circuits.

The findings were published in "Nature Communications" and offer promising avenues for further exploration in neural modeling and understanding brain functions related to visual perception.

Source: Medical Xpress