Biologically Inspired Autonomous Mental Development Model with Visual Selective Attention Mechanism

Abstract

We propose an autonomous mental development model that can voluntarily decide where and what it wants to see based on a bottom-up and top-down visual selective attention model in conjunction with human interaction. The proposed bottom-up saliency map model was developed by mimicking the functions of the visual pathway from the retina to the visual cortex through LGN. A low level topdown attention model implemented by a modified hierarchical Fuzzy ART network can incrementally inhibit uninteresting areas and reinforce interesting areas through human interaction. And a high level top-down attention model with human interaction for object non-specific representation and detection is proposed, which consists of a Gaussian mixture model and a maximum likelihood method for object representation and detection, respectively. The proposed model can generate a plausible attention map and give control signals to the effectors in robots to increase the machine intelligence through human interaction, autonomously.