József Fiser, Department of Cognitive Science, Central European University, Budapest, Hungary
Monday, March 17 2014, 11h Salle de réunion du LPP, H432
The role of time in making perceptual decisions
In models of perceptual decision making within the classical signal processing framework (e.g. integration-to-bound), time is solely used to accumulate evidence. In the recently proposed probabilistic, sampling-based frameworks, time is necessary to collect samples from subjective posterior distributions for the decision regardless whether sensory evidence is still entering the system. Which of these two roles does time assume and how do those roles relate to each other during perceptual decisions in everyday perceptual tasks? In my talk, first I will give a brief overview of the evidence integration and probabilistic sampling frameworks, then I will present results of an analytically derivation to show the theoretical progression of the error and subjective uncertainty in time for these two models of decision making. I will demonstrate that the correlation between subjects’ error and their subjective uncertainty shows a very differently evolving pattern under sampling and evidence integration. Under sampling, after a brief initial period, the correlation always increases monotonically to a non-zero asymptote with this increase continuing long after the error itself has reached its asymptote. In contrast, integration-to-bound with additive, non-negligible behavioral noise always shows a decreasing correlation. Next, I will present two sets of experiments utilizing these antagonistic predictions about the correlations. In the first decision making study, where subjects had to perform time-limited orientation matching and report their uncertainty about their decisions, the results confirmed both predictions of the sampling-based model: the correlation converged to a non-zero asymptote long after no additional evidence was provided, and correlations increased with time. The second experiment used the classical decision making task concerning the direction of random dot motion displays under various coherence. In each individual result, we found a marked decrease in error-uncertainty correlation in the first part of the trial, indicating evidence integration, and a significant increase in the second part, indicating probabilistic sampling. Moreover, the transition between these segments shifted in accordance with the change in signal coherence. These findings support a novel interpretation of the role of time during decision making processes: under typical conditions, time is mostly used for assessing probabilistically what we really know and not for gathering more information. In addition, during any decision making, probabilistic sampling works in parallel with evidence integration, the former taking the lead early but probabilistic sampling determining the later part of the process.