Learning where to look and what to attention (Back)


Learning where to look and what to attend is the subject of one of our research directions.
This line of research is designed to address one of major problems in design of cognitive robots; that is the design of feature space. There are a large set of extractable features in multimodal sensory system but attending to all of them for decision making is impossible due to bounded computational resources and response time. In addition, careful selection of features can highly improve quality of cognitive decisions. On the other hand, optimum set of features varies from one decision to another. Therefore, feature selection can be framed in attention control framework. In this research we look for some mathematical frameworks and learning methods to learn what to attend along with learning decision. Moreover, we study the effect of attention control on the structure of recognition and decision making. We use findings in neuroscience and cognitive science to interpret our hypothesis and methods. 


< Back