Adaptive stimulus selection for multi-alternative psychometric functions with lapses.

Area of Science:

• Cognitive Neuroscience
• Computational Psychology
• Psychophysics

Background:

• Psychometric functions (PFs) are crucial for modeling sensory and cognitive processes by quantifying stimulus-behavior relationships.
• Current adaptive stimulus-selection methods are limited, especially for complex tasks with multiple response alternatives.

Purpose of the Study:

• To develop novel adaptive stimulus-selection methods for flexible psychometric function (PF) models.
• To improve the efficiency of estimating PFs in multi-alternative tasks.

Main Methods:

• Utilized a multinomial logistic regression mixture model to represent realistic psychophysical behavior, including lapses.
• Developed an information-theoretic criterion for stimulus selection.
• Implemented computationally efficient inference and selection using adaptive Markov-chain Monte Carlo (MCMC) sampling.

Main Results:

• Applied the new methods to macaque monkey motion-discrimination task data.
• Demonstrated substantial speed-up in PF estimation compared to random stimulus selection in simulations.
• Validated the model's ability to handle realistic psychophysical complexities like lapses and multiple alternatives.

Conclusions:

• The developed adaptive methods significantly reduce the number of trials needed to accurately model multi-alternative PFs.
• These techniques offer a more efficient approach to building models of sensory and cognitive processes.
• The methods are extendable to high-dimensional PFs, overcoming limitations of standard approaches.