Signal detection theory, uncertainty, and Poisson-like population codes

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Perception

Background:

• Established signal detection theory (SDT) models and recent claims on perceptual uncertainty encoding cause confusion.
• Distinguishing between optimal and probabilistic computation is key to understanding perceptual uncertainty.

Purpose of the Study:

• To precisely differentiate between optimal and probabilistic computation in perception.
• To clarify the neural basis for representing uncertainty in perception.

Main Methods:

• Theoretical analysis of optimal vs. probabilistic computation.
• Examination of single-neuron activity and population codes in relation to perceptual uncertainty.
• Comparison of Poisson-like and Gaussian neural variability.

Main Results:

• Optimal computation can occur without probabilistic computation, and vice versa.
• Behavioral evidence for neural uncertainty representation requires probabilistic computation models.
• Single-neuron activity supports optimal computation, but population codes are necessary for probabilistic computation.
• Poisson-like neural variability, unlike Gaussian variability, facilitates marginalizing nuisance parameters in population codes.

Conclusions:

• Probabilistic computation, essential for representing perceptual uncertainty, necessitates population neural codes.
• A theoretical framework linking SDT to Poisson-like population codes is established.