neural field as inference
Active inference provides a theory about how agents employ a particular adaptive inference scheme. A neural field provides a description of cortical activity in particular induced by the 2d topology.
mix
What happens when we interpret waves 🌊 on the field as belief propagation? In other words, what kind of statistical distribution is induced by field topology?
shake
As with many questions, it can be useful to ask: if you had the answer, what would you do with it? What added value is there?
pour
Two possibilities come to mind:
the mix provides a model that can be inverted to fit existing brain data, in order to then interpret the data as evidence of the Bayesian brain
the mix is used itself for doing data analysis, as a sort of generic generative model for data where the field topology is a useful regularization
pay
Some implementation details need to be worked out:
- what kind of field topology?
- single fixed kernel or many kernels tuned to the data?
- adaptation mechanisms to keep the field in a critical régime?