A genetic algorithm approach to shape coding in the ventral visual pathway


April 26, 2016 - 12:00pm
Northwest 243
About the Speaker
Ed Connor (Johns Hopkins)

We study neural coding of 2D and 3D shape in the ventral pathway of visual cortex. Because shape is a virtually infinite domain, neural coding is sparse and difficult to define. To compensate, we use genetic algorithms to focus stimulus sampling on the response ranges of specific neurons. This yields datasets that can constrain quantitative models relating stimulus geometry to neural responses. In previous studies, we have used genetic algorithms to show how objects are encoded as configurations of 3D surface fragments and medial axis elements. In recent work, we have studied large scale shape coding of scenes and rooms in the ventral pathway. In addition, we are analyzing how shape tuning functions change with learning to support discrimination of familiar shapes.