According to Feather, “if you train models in noise, they give better brain predictions than if you don’t” because “a lot of real-world hearing involves hearing in noise, and that’s plausibly something the auditory system is adapted to.” This is intuitively reasonable because “they give better brain predictions than if you don’t.”
The new concentrate additionally upholds the possibility that the human hear-able cortex has some level of progressive association, wherein handling is separated into stages that help particular computational capabilities. As in the 2018 review, the scientists found that portrayals created in before phases of the model most intently look like those found in the essential hear-able cortex, while portrayals produced in later model stages all the more intently look like those produced in cerebrum locales past the essential cortex.
Also, the scientists observed that models that had been prepared on various assignments were better at duplicating various parts of tryout. For instance, models prepared on a discourse related task all the more firmly looked like discourse specific regions.