The sandbox does it different

"Theory vs. Data" in statistics too

There is an economic debate about 'deep learning'.  

In the sandbox, 'deep learning' means to mark to market inside information with the minimum number of transactions.  This is pure cash, by definition. It works because we know the distribution of information and distribution of goods have isomorphic, reduced graphs.  We know this because we can show a Huffman decoding tree also minimizes congestion on the decoding tree. And under the assumption of maximizing entropy we get isomorphic equivalence.

We need not know much more, we need to get our  automatic liquidity trading pits working.