Both rational and irrational: Understanding how the brain makes rational and irrational probabilistic inference
| Autor: | Yun-Yen Yang, 楊勻硯 |
|---|---|
| Rok vydání: | 2016 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 104 Uncertainty is a central feature in many decisions we face. Making decision under uncertainty often requires humans to have accurate knowledge about probability distribution over possible outcomes. However, information about probabilities can come from different sources. Hence, making source integration is a key computational problem. In this study, we investigated how humans combine prior experience and current information so as to make the appropriate choice. In particular, we asked how the reliability of these two sources of information affects the weights that subjects assign to them. In a probabilistic inference task, using functional magnetic resonance imaging (fMRI), on each trial, subjects were presented with two visual stimuli – one associated with past experience on rewards (prior knowledge) obtained in a previous session and the other is a new and independent piece of sensory evidence for reward probability (likelihood information). In order to maximize rewards, subjects should integrate both sources of information by taking into account the reliability of each source so as to compute the optimal estimates of probability. At the behavioral level, we found a unique pattern of suboptimal performance: When prior information was highly reliable, subjects exhibited patterns similar to base-rate neglect – they underweight the prior and overweight the likelihood information. In contrast, when prior information was highly variable, subjects achieved near-optimal integration. By using distribution-based model, we were able to understand subjects’ evaluation on prior and likelihood distribution. We found that suboptimal integration did not result from the underweighting the prior distribution. Rather, it came from overweighting the likelihood function. Using fMRI, we found that brain activity in the central orbitofrontal cortex (cOFC), the anterior insula and visual cortex were positively correlated with the reliability of the prior distribution, while the medial prefrontal cortex (mPFC) and medial orbitofrontal cortex (mOFC) represented the reliability of the likelihood function. Furthermore, independent region-of-interest analyses suggested that cOFC not only represented the group average of weight (wL) subjects assigned to likelihood relative to prior information, but also the individual differences in wL. These results indicated that cOFC reflected the reliability-dependent asymmetry in information weighting. Finally, by using PPI analysis, we found that prior coding in the visual cortex showed a decrease in functional connectivity with cOFC when subjects assigned less weight to prior information than they should. This indicated the possibility that suboptimal inference is driven by insufficient connectivity between the visual system and cOFC. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|