Application of Artificial Neural Networksfor Characterizing Visual Information Encoding in Cortical Neurons and for Developing Stimulation Protocols for Prosthetic Vision

Abstract

Title: Application of Artificial Neural Networks for Characterizing Visual Informa- tion Encoding in Cortical Neurons and for Developing Stimulation Protocols for Prosthetic Vision Author: Luca Baroni katedra: Charles University, Faculty of Mathematics and Physics Supervisor: Mgr. Ján Antolík, Ph.D., KSVI Abstract: This cumulative thesis presents applications of artificial neural networks as response-predicting models of primary visual cortex (V1) to investigate neural encoding properties and to inform the design of cortical visual prosthesis stimulation protocols. It comprises four works, organized around these two themes. The first three works study V1 neural encoding properties using encoding models as digital twins of biological neurons. This approach enables systematic stimulus-space exploration difficult to achieve in in-vivo experiments. The first two works present a method that combines encoding models with implicit neural representations and contrastive learning to systematically characterize single-neuron invariance properties and enable population-level analyses. The third work examines how surround context modulates V1 responses, showing that contextual modulation is shaped by natural scenes statistics. The final work addresses prosthesis applications of encoding models. It introduces a...