Bayesian integration in setting classification criterion
Hee‐Seung Lee, Sang‐Hun Lee · IBRO Reports · 2019
Classification is a systematic arrangement of stimuli in classes according to predefined criteria.So, classification is composed of three independent computational steps: (1) perceiving stimuli, (2) setting criteria and (3) comparing the stimuli with the criteria.However, literatures of classification have mostly investigated classification focusing on how variability in stimulus perception modulates decision, while how the classification criterion is formed in mind and affect decision are widely overlooked.Here we tested two hypotheses for explaining the underlying setting process for classification criteria.The first hypothesis is the adaptive account.It explains that an agent forms criteria by heuristically weighting memory of past stimuli in a way that recent stimulus is more weighted than the past.The rationale for the weighting is that ever-changing environment forces us to consider more recent experience as more informative to current situation.The second hypothesis is the Bayesian account, in which an agent integrates past stimuli by weighting them according to their representational reliability, not to the lapse of the past stimuli.By varying uncertainty of stimuli trial-to-trial in perceptual classification task, we could identify that human subjects set criteria in accordance with the Bayesian account.Our results imply that human agents process not only current stimuli but also past stimuli in memory for classification by the normative probabilistic computation.