Hierarchical Bayesian inference

How does the visual system extract and represent statistical patterns from visual scenes? What are the mechanisms of perceptual inference? We have carried out a series of neurophysiological experiments demonstrating that early visual areas are strongly modulated by global contextual factors and higher order perceptual constructs. These experimental findings have lead us to re-consider the role of the early visual areas in visual computation and proposed the high-resolution buffer theory of V1. Recently, we proposed a hierarchical Bayesian framework for conceptualizing cortical inference, with the feedforward/feedback loops performing Bayesian belief propagation or particle-filtering. Our laboratory is actively pursuing a number integrated computational and neurophysiological projects to test this hypothesis.

• Project 1: Theory of hierarchical Bayesian inference and particle filtering
  • Project Leader: Tai Sing Lee
  
  
• Project 2: Influence of higher order contexts in early visual computation
  • Project Leader: Tai Sing Lee
  
  
• Project 3: 3D surface representation and inference
  • Project Leader: Brian Potetz
  
  
• Project 4: Analysis of motion in Bayesian hierarchy
  • Project Leader: Ryan Kelly
  
  
• Project 5: Object recognition in scene context
  • Project Leader: Tom Stepleton