Deep Neural Networks for Modeling Visual Perceptual Learning

• DOI: 10.1523/jneurosci.1620-17.2018
• Publication Date: 2018-05-23T15:35:43Z
• Authors: Li K. Wenliang; Aaron R. Seitz
• Keywords: Perceptual Learning; Stimulus (psychology); Computational model

Understanding visual perceptual learning (VPL) has become increasingly more challenging as new phenomena are discovered with novel stimuli and training paradigms. Although existing models aid our knowledge of critical aspects VPL, the connections shown by these between behavioral plasticity across different brain areas typically superficial. Most explain VPL readout from simple representations to decision not easily adaptable findings. Here, we show that a well -known instance deep neural network (DNN), whereas designed specifically for provides computational model enough complexity be studied at many levels analyses. After Gabor orientation discrimination task, DNN reproduced key results, including increasing specificity higher task precision, also suggested precise discriminations could transfer asymmetrically coarse when stimulus conditions varied. Consistent findings, distribution moved toward lower layers precision increased this was modulated tasks types. Furthermore, in units demonstrated close resemblance extant electrophysiological recordings monkey areas. Altogether, fulfilled predictions theories regarding findings tuning changes neurons primate comparisons were mostly qualitative, method studying can serve test bed theories, assists generating physiological investigations. <b>SIGNIFICANCE STATEMENT</b> Visual been found cause multiple stages hierarchy. We (DNN) on an produced patterns similar those human experiments. Unlike models, pre-trained natural images reach high performance object recognition, but VPL; however, it When used care, unbiased deep-hierarchical provide ways behavior physiology.