The cognitive paradigm of predictive processing is gaining immense theoretical footing, yet no attempts have been made to apply it in building artificial machines. We propose a novel approach to computer vision to investigate intrinsically motivated vision awareness created by top-down prediction errors between the algorithm’s probabilistic, generative model and sensory stimuli. Humans perceive a structured world from infinite data inputs in this way; prediction errors are propagated through layers, and once they are resolved by changing the model or the environment, a sensory experience is perceived. We successfully reconstruct this biologically plausible approach to vision awareness using a hierarchical neural system and suggest a theoretical framework away from traditional vision approaches.

Derek (Yue) Yu is a double HBSc candidate in cognitive science and computer science at the University of Toronto. In just his first year, he has gained extensive research experience as a research assistant for Dr. Nick Koudas, Dr. Jay Pratt, Dr. Daniel Wigdor. He is interested in computer applications of the predictive processing paradigm, creating novel cross-sections between cognitive and computer research.