A Predictive Approach to Nonparametric Inference for Adaptive Sequential Sampling of Psychophysical Experiments.

Area of Science:

• Psychophysics
• Bayesian statistics
• Machine learning

Background:

• Psychophysical experiments often involve ordinal stimuli and limited prior structural knowledge, hindering traditional parametric modeling.
• Adaptive sequential sampling is crucial for efficient data acquisition in such scenarios.

Purpose of the Study:

• To present a predictive account of adaptive sequential sampling for stimulus-response relations in psychophysical experiments.
• To develop a flexible modeling framework applicable when parametric assumptions are not feasible.

Main Methods:

• Introduced partial exchangeability to develop a hierarchical Bayesian model.
• Utilized a mixture of Pólya urn processes for sequential sampling.
• Developed efficient algorithms for computing posterior predictions and information-theoretic utility measures.

Main Results:

• The proposed hierarchical Bayesian model enables adaptive sequential sampling.
• The framework efficiently computes exact solutions for posterior predictions and utility measures, overcoming computational challenges.
• Demonstrated the framework's advantages on a hypothetical sampling problem.

Conclusions:

• The developed method provides an effective approach for optimizing experimental sampling in psychophysics.
• This framework is particularly useful for situations with ordinal stimuli and weak structural information.
• The efficient computation of utility measures enhances the practicality of adaptive sampling strategies.