题目：A fresh, old look on vision

报告人：Michael Herzog(EPFL, Switzerland)

报告人简介：

Michael Herzog is a professor for psychophysics at the Brain Mind Institute (BMI) at the EPFL in Lausanne (Switzerland). He studied Mathematics, Biology, and Philosophy at the Universities of Erlangen, Tübingen (both Germany), and the Massachusetts Institute of Technology (USA). He got a ''Diplom'' in mathematics . and received a Master in philosophy with an investigation about current approaches to intentionality and representation (Prof. Keuth, Tübingen). Under the supervision of Prof. Fahle at the Section of Visual Science (Tübingen) and Prof. Poggio at MIT, He finished a thesis on ''mathematical models and psychophysical experiments of perceptual learning'' earning a Ph.D. in biology.In 1998-1999, he did a post-doc in the lab of Prof. Koch at Caltech (USA) investigating the characteristics of temporal processing and feature integration. From 1999-2004, he have been a senior researcher at the section of Human Neurobiology at the University of Bremen (Head: Prof. Fahle) and was a leader of a research project at the Center of Excellence 517 ''Neurocognition'' of the DFG (German Research Council).

时间：2015年12月08日 星期二 13:00-15:00

地点： 北京大学王克桢楼1113室

摘要：

In classic models of vision, vision proceeds in afeedforward andhierarchical fashion, from low-level analysis (edges and lines) tofigural processing (shapes and objects). Low-level processing fully determineshigh-level processing. For example, the responses of a neuron sensitive to a square are fully determined by the neural activity of neurons coding for the vertical and horizontal edges making up the square. First, using crowding as a paradigm, I will show that shapeprocessing determines low-level visual processing as much as the other way around. For example, we presenteda vernier stimulus and asked observers to indicate its offset direction.Performance strongly deteriorated when the vernier was surrounded by asquare, in line with most models of vision. Surprisingly, performanceimproved when more squares were added. All classic models of vision predictjust the opposite,i.e., a further deterioration of performance. Second, using visual masking, EEG, and transcranial magnetic stimulation (TMS), I will show that features are unconsciously integrated over half a second before consciousness is reached. During unconscious processing, shapeprocessinginteracts with low-level processing in a recurrent fashion. Third, I will quickly sketch a computational model, which is very little related to the classic models of vision, employing recurrent connections. Computer simulations will show how neural dynamics explain the results of the crowding experiments mentioned above (cooperation with Greg Francis and the Human Brain Project).Even though the models is state of the art with millions of neural connection, it reverberates the tradition of Gestalt theory.