3 ms·
If we want to "understand" what a network does, that really means we want to disentangle cause and effect and spit out simple algebraic models for it after dist
by TheOtherHobbes 10y ago
If we want to "understand" what a network does, that really means we want to disentangle cause and effect and spit out simple algebraic models for it after distilling them from a training set.
To the extent this is even possible - which is debatable, for all kinds of reasons - we're going to need a different set of tools. ML is not the right tool for that problem.
Something similar to ML may be, but ML itself definitely isn't.