Knowledge-augmented face perception: Prospects for the Bayesian brain-framework to align AI and human vision

Area of Science:

• Cognitive Science
• Computer Vision
• Artificial Intelligence

Background:

• Human visual perception is efficient, flexible, and context-sensitive, explained by the Bayesian brain's probabilistic inference.
• Artificial Neural Networks (ANNs) have advanced computer vision but lack adaptive use of context and prior knowledge.

Purpose of the Study:

• To propose methods for aligning human and computer vision, focusing on facial expression recognition.
• To explore how the Bayesian brain framework can enhance ANNs for improved human-machine interaction.

Main Methods:

• Reviewing human face perception studies on knowledge-augmentation and context-sensitivity.
• Examining current approaches in computer vision that leverage contextual information.
• Discussing the potential for an "epistemic loop" between neuroscience and AI.

Main Results:

• Identified a gap in ANNs' ability to utilize contextual cues and prior knowledge adaptively.
• Highlighted human perception's reliance on top-down influences and prior experience.
• Proposed ANNs inspired by the Bayesian brain framework for more flexible AI.

Conclusions:

• Aligning ANNs with the Bayesian brain framework can improve flexibility and utility in human-machine interaction.
• ANNs can serve as powerful tools to advance empirical research into human knowledge-augmented perception.