Bayesian adaptive stimulus selection for dissociating models of psychophysical data

Abstract

15Comparing models facilitates testing different hypotheses regarding the computational basis of percep-16 tion and action. Effective model comparison requires stimuli for which models make different predictions. 17Typically, experiments use a predetermined set of stimuli or sample stimuli randomly. Both methods 18 have limitations; a predetermined set may not contain stimuli that dissociate the models whereas random 19 sampling may be inefficient. To overcome these limitations, we expanded the psi-algorithm (Kontsevich 20 & Tyler, 1999) from estimating the parameters of a psychometric curve to distinguishing models. To 21 test our algorithm, we applied it to two distinct problems. First, we investigated dissociating sensory 22 noise models. We simulated ideal observers with different noise models performing a 2-afc task. Stimuli 23 were selected randomly or using our algorithm. We found using our algorithm improved the accuracy 24 of model comparison. We also validated the algorithm in subjects by inferring which noise model un- in model comparison. We conclude that our technique is more efficient and more reliable than current 32 methods of stimuli selection for dissociating models. 33 Data Availability 34All data and code will be posted on our institutional repository system following acceptance. In the 35 meantime feel free to contact the authors if you would like any of the code.

Psychophysical Evaluation of Sensory Reweighting in Bilateral Vestibulopathy

“…Before reaching this stage, however, various aspects need to be optimized, from stimulus design to data recording, test duration, and data-analysis. One way to proceed is by incorporating modern adaptive psychometric procedures, which could improve efficiency in parameter estimation, both in terms of number of trials needed and the quality of the estimates ( 71 , 72 ).…”