Extracting the free energy contributions for sensor selection

[lstagner]

I've been reading about active inference and one of the benefits people mention is that as an agent minimizes its free energy it can selectively filter out sensors that are not contributing new information. For instance, I could turn off sensors whose free energy contribution is less than some tolerance saving me computational cost. If there is a spike in the free energy I can then turn on sensors whose average constribution is approximetely equal the the spike in free energy.

Is there an example of this somewhere.

[bvdmitri]

No, but we have a research PhD project in BIASlab that explores how to do that properly, e.g. what is the threshold for those spikes or how to stop sending messages if there are no spikes. Currently RxInfer doesn't have any functioning example of that, but we are working on this topic