Pedro Domingos

Feynman

So what percentage of deep learning usages are more art and guessing than engineering? Is a deep learning use stable or unstable, bounded or not? Can this be determined a head of time or only discovered in testing or worse deployment?

That's it. I run the LSTM, but I cannot get how and why that thing can learn. I don't well understand how those tensors, cell and hidden states, may get the context of data and pass it.

Deep learning is a dead end, it's a dumb simple algo we managed to stretch to do cool tricks over the years but its writing is on the wall. @Numenta's "thousand brains" is the future with its sdr, spatial pooling, active columns, voting etc. And it's not a black box either