Harnessing Vision Science to Overcome the Critical Limitations of Artificial Neural Networks

14.03.2024 bis 13.03.2027

Artificial neural networks (ANNs) drive the current interest in computer vision systems and artificial intelligence (AI) in general. However, ANNs used in computer vision still suffer from a number of shortcomings such as their susceptibility to adversarial attacks or their need for vast amounts of data to train them. Both limitations are not observed in biological vision systems. We believe that one fundamental difference between biological and artificial vision systems are the formers stronger, dynamic nonlinearities in the first stages of processing. In this grant we explore the intrinsically nonlinear receptive field (INRF) as a potential (partial) remedy for some of the shortcomings of current ANNs in vision science.