1
$\begingroup$

There are multiple examples in the machine learning literature of trying to learn the hierarchical structure of a reinforcement learning problem, however have there been any papers tying this learning to neurological activity in the brain?

The type of evidence I would like some indication of a neural substrate indicating some type of activity when learning a hierarchical structure is necessary to solve a problem vs. when a simple problem is attempted.

$\endgroup$
  • 1
    $\begingroup$ I am a bit confused on the question but probably it is on me as I have no knowledge in machine learning. Are you asking for the neural substrate of certain processes of human problem solving? Or for the substrate of reinforcement learning (i.e. the reward system)? Either way, I think linking the findings from machine learning to neural mechanisms would work only under the assumption that machine learning and human learning are analogous, and that's where it gets complicated. $\endgroup$ – Lazaros Mitskopoulos Nov 26 '14 at 17:56
  • $\begingroup$ @LazarosMitskopoulos I'm mostly looking for neural mechanisms that imply some sort of hierarchical deconstruction. $\endgroup$ – Seanny123 Feb 15 '18 at 17:11
  • $\begingroup$ I think some would argue that the sensory and associative areas of the brain are almost entirely a hierarchical problem solving structure...but rarely does an individual bother to learn the hierarchical structure, rather all problems that are solved in a hierarchical fashion learn the hierarchical structure implicitly. In that way it seems sort of an ill-posed question. $\endgroup$ – Bryan Krause Feb 16 '18 at 21:21
1
$\begingroup$

According to "Frontal Cortex and the Discovery of Abstract Action Rules" by David Badre, Andrew S. Kayser, and MarkD'Esposito:

... the rostro-caudal axis of frontal cortex support rule learning at higher levels of abstraction. Moreover, these results indicate that when humans confront new rule learning problems, this rostro-caudal division of labor supports the search for relationships between context and action at multiple levels of abstraction simultaneously.

They found this by contrasting fMRI signals collected while the subject performed two tasks, one requiring a hierarchical, abstract rule and one requiring a flat, basic mapping.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.