| Paper: | PS-1B.41 |
| Session: | Poster Session 1B |
| Location: | Symphony/Overture |
| Session Time: | Thursday, September 6, 18:45 - 20:45 |
| Presentation Time: | Thursday, September 6, 18:45 - 20:45 |
| Presentation: |
Poster
|
| Publication: |
2018 Conference on Cognitive Computational Neuroscience, 5-8 September 2018, Philadelphia, Pennsylvania |
| Paper Title: |
A Perceptual Confirmation Bias from Approximate Online Inference |
| Manuscript: |
Click here to view manuscript |
| DOI: |
https://doi.org/10.32470/CCN.2018.1167-0 |
| Authors: |
Richard Lange, Ankani Chattoraj, Matthew Hochberg, University of Rochester, United States; Jeffrey Beck, Duke University, United States; Jacob Yates, Ralf Haefner, University of Rochester, United States |
| Abstract: |
The mechanisms underlying evidence accumulation in perceptual decision-making tasks have been the subject of much research. However, existing studies differ in their conclusions on whether the brain weighs evidence optimally over time, or whether it exhibits biases towards evidence presented early (primacy) or later (recency) in the trial. We resolve this discrepancy in the literature by proposing that previous tasks differ in how task-relevant information in the stimulus is partitioned into "sensory information" and "category information." We demonstrate that similar stimulus-dependent biases arise naturally in two common models of approximate inference: neural sampling-based inference, and parametric inference (Variational Bayes). Finally, we test our model by designing a psychophysics task that systematically trades off these two sources of uncertainty in the stimulus against each other while keeping all other aspects of the task the same. We find that subjects' evidence-weighting strategies change in the predicted direction and in a highly robust fashion, individually significant for every one of our 10 subjects. |
