Massachusetts Institute of Technology researchers determined that the brain's demand and language systems encode specific code properties that correlate with machine-learned representations.
The researchers performed functional magnetic resonance imaging (fMRI) scans of programmers' brains as they processed computer program elements and saw a difference in encoding properties related to dynamic analysis, in which information is encoded much better in the brain's multiple demand network compared to the language processing center.
A second experimental series incorporating neural networks confirmed the brain signals observed when participants examined pieces of code closely resembled the networks' own activation patterns.
Mariya Toneva at Germany's Max Planck Institute for Software System said such findings "raise the possibility of using computational models of code to better understand what happens in our brains as we read programs."
From MIT News
View Full Article
Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA
No entries found