Open dlee138 opened 9 years ago
The layers differ from clusters in that the interconnectivity allows for relations between voxels to be layered as opposed to discretization that is done in clustering. Do you see any ways in which we can account for layering in connectomics data? Specifically, how would you teach the algorithm to differentiate between clusters and layers?
I thought the layers were referring to imaging processes in which one layer of the image may contain a neuron which synapses with a neuron in a different layer. Is this correct?
The way I understood it, layering in neural network classifiers were inspired by the different specialized processing layers in the brain. I would guess that the relationship between layers could be best summarized by the definition of a bipartite or multi-partite graph. You could design algorithms to detect these multi-partite relationships and they would be distinctly different from clustering algorithms. I suspect such an algorithm would be NP-complete.
The data mining paper describes a neural network as a layered graph, where the output of one node feeds into one or many other nodes in the next layer. Are "layers" the same thing as clusters then? How exactly does this layered structure make classification more challenging?