| 研究実績の概要 |
We have two main accomplishments with respect to the Kakenhi grant in fiscal year 2014 for which we have presented at international conferences (SfN and CoSyNe) and are preparing manuscripts for submission.
1. Switching between prior and evidence. We have examined short-term priors in a motion estimation task and found behavior that summary statistics suggested followed optimal bayesian inference was in fact a heuristic approximation. Subjects are biased towards their prior when sensory evidence is weak and back towards sensory evidence when the prior is weak just as optimality theory predicts. However, closer examination of the distribution of choices suggests that subjects were switching between their prior and evidence on different trials and not integrating the two pieces of information as predicted by standard theory. We believe that this is an important heuristic that humans can use to approximate optimality and are actively investigating its neural basis.
2. Adaptability of choice history biases. We have been examining choice history biases in a simple detection task. We have found that subjects are biased by their previous choices (typically by switching after a failure) even though this is non-optimal for the task. This suggests that this is a general prior strategy that humans have that is not overridden by current task demands. So, we therefore asked whether we could override initial biases by imposing stronger task demands to switch biases and found that we could adapt choice history biases, but this was hard to do when trying to go against existing biases.
|
