Checking Brain Expertise Using Rough Set Theory Andrzej W. Przybyszewski 1 Dept Psychology McGill University, Montreal, Canada 2 Dept of Neurology, University of Massachusetts Medical Center, Worcester, MA US przy@ego.psych.mcgill.ca Abstract. Most information about the external world comes from our visual brain. However, it is not clear how this information is processed. We will analyze brain responses using machine learning methods based on rough set theory. We will test the expertise of the visual area V4, which is responsible for shape classifications. Characteristic of each stimulus are treated as a set of learning attributes. We assume that bottom-up infor- mation is related to hypotheses, while top-down information is related to predictions. Therefore, neuronal responses are divided into three cat- egories. Category 0 occurs if cell response is below 20 spikes/s (sp/s), indicating that the hypothesis is not valid. Category 1 occurs if cell ac- tivity is higher than 20 spikes, implying the hypothesis is valid. Category 2 occurs if cell response is above 40 sp/s; in this case we conclude that the hypothesis and prediction are valid. By using experimental data we make a decision table for each cell, and generate equivalence classes. We express the brains basic concepts by means of the learners basic cate- gories. By approximating stimulus categories with concepts of different cells we determine core properties of cells, and differences between them. On this basis we have created profiles of their receptive field properties. Keywords: V4, machine learning, bottom-up, top-down processes, neu- ronal activity. 1 Introduction Most of our knowledge about function of the brain is based on electrophysiologi- cal recordings from single neurons. In the lower visual areas like the retina, LGN or V1 (primary visual cortex) it is relatively easy to find an optimal stimulus for each neuron. The receptive fields in these areas are small and simple. On the other end, in the area designated as IT (inferotemporal cortex), receptive fields are very large and optimal stimuli are generally unknown, though they could be as complex as faces. In consequence, different laboratories propose different often contradictory hypotheses on the basis of their different testing stimuli. An- other part of the confusion is related to non-uniform properties of neurons in area V4 of the brain. Therefore we do not know if different experimental results and hypotheses are related to different methods and classifications or to different classes of cells. M. Kryszkiewicz et al. (Eds.): RSEISP 2007, LNAI 4585, pp. 746–755, 2007. c  Springer-Verlag Berlin Heidelberg 2007