The Vision–Brain Hypothesis

Abstract

In this chapter, the vision–brain hypothesis is illustrated in three steps. First, we hypothesize that vision decides the robots’ attention and the attention could be regulated without brain-inspired objects detection and tracking. This could be highlighted by the difference in attention mechanisms between the manned and unmanned systems. Regulated attention in unmanned systems has a significant implication to the robots’ cognition accuracy and response speed. Therefore, the current learning systems must be optimized. Such optimization can be interpreted as an integration of deep learning with other hybrid adaptive algorithms. Second, we hypothesize that if a region of interest had been located by a “vision–brain,” then scene understanding and partition could be smoothly carried out, which help cognition systems to reduce the loss from mid-aligned samples addition by employing non-adaptive random projections instead of self-taught learning. Third, we hypothesize that cognition rates of the “vision–brain” could approach to 100%, which finally establishes the robustness and efficiency of the vision–brain. A broad learning system (BLS) was integrated with a decision layer (the vision–brain) to address the issue whether face recognition rates can reach 100%. And in the next chapter, we will show that face recognition rates can reach 100% in BLS with the vision–brain, as verified by a challenging AR database with real occlusion. BLS performance in face recognition on other bigger databases remains unknown and worthy of further attempts.