Late Bayesian inference in mental transformations

Evan D Remington,Mehrdad Jazayeri,Tiffany V Parks

doi:10.1038/s41467-018-06726-9

Evan D Remington, Mehrdad Jazayeri + Show 1 more

Open Access

https://doi.org/10.1038/s41467-018-06726-9

Copy DOI

Abstract

Many skills rely on performing noisy mental computations on noisy sensory measurements. Bayesian models suggest that humans compensate for measurement noise and reduce behavioral variability by biasing perception toward prior expectations. Whether a similar strategy is employed to compensate for noise in downstream mental and sensorimotor computations is not known. We tested humans in a battery of tasks and found that tasks which involved more complex mental transformations resulted in increased bias, suggesting that humans are able to mitigate the effect of noise in both sensorimotor and mental transformations. These results indicate that humans delay inference in order to account for both measurement noise and noise in downstream computations.

Highlights

Because the transformation in this task did not involve a gain factor, the noMTN prediction for BIAS in the remapped context was equal to that measured in the identity context. Because in this task we modeled both measurement and production noise as non-scalar[1,2,30], the no-mental transformation noise (MTN) prediction for √VAR was equivalent to that computed in the identity context
The observer-actor model fits for the remapped context were associated with higher values of wpre, while we found no systematic effect on wpost
The key advance in our work is the finding that the brain has the capacity to establish a late inference strategy so that behavior can account for the sources of noise associated with sensorimotor and mental transformations of sensory inputs

Summary

Methods

Human subjects aged 18–65 years participated in this study after giving informed consent. All experiments were approved by the Committee on the Use of Humans as Experimental Subjects at the Massachusetts Institute of Technology. Semi-overlapping groups of 11 or 12 subjects completed experiments 1 and 3, and experiments 2 and 4, respectively. Eleven subjects completed an additional experiment shown in Supplementary Figure 4. Differences in number of subjects across tasks were due to subject attrition. All subjects had normal or corrected-tonormal vision

Results

Discussion

Conclusion