The neural code of perceptual inference.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience

Background:

• Perception is fundamentally an inferential process, integrating sensory evidence with prior expectations.
• Current evaluations of the neural code for perception prioritize faithful sensory representation over optimal inference.
• This conventional approach leads to biases in neural code assessment, favoring discriminability and efficiency over generalizability and robustness.

Purpose of the Study:

• To propose a framework for evaluating the neural code of perception based on its capacity for perceptual inference.
• To highlight the misalignment between the computational goal of perception and conventional methods of neural code assessment.

Main Methods:

• This is a review article, synthesizing existing literature and theoretical concepts.
• It proposes a conceptual shift in evaluating neural codes for perception.

Main Results:

• Conventional evaluation metrics for neural codes are misaligned with the inferential nature of perception.
• A new evaluation framework focusing on perceptual inference can address current biases.

Conclusions:

• Re-evaluating the neural code for perception through the lens of inference is crucial.
• This shift will lead to a more accurate understanding of neural mechanisms underlying perception and improve computational models.