Object Recognition Using a Bayesian Network Imitating Human Neocortex

After Mountcastle proposed the theory that all parts of mammalian neocortex are uniform, more and more evidences were found to prove that no matter what function the field of neocortex provide, the organizing of cortex cells are same and have an hierarchical structure. It can be inferred every unit of neocortex process information using an identical algorithm. This means if people can find and imitate this algorithm, all intelligence work will be resolved in one way, the only difference is using different sensors. Base on this, Hawkins proposed a topdown model of neocortical operation, in which model, both continuous time and prediction play important roles in human's invariant recognition. Then George and Hawkins developed an inspired framework named "hierarchical temporal memory", using a hierarchical Bayesian network to perform prediction and temporal sequence in training the network, which partially implements Hawkins' model, and can provide a rather good performance in invariant recognition. We current study is extending their work by implementing the framework in visual category recognition field.