> An interesting question arises, for example, whether the class of NP-complete problems could be attacked more efficiently in the network environment! It is w

> An interesting question arises, for example, whether the class of NP-complete problems could be attacked more efficiently in the network environment! It is well known that training even a two-layer feed-forward neural network is NP-hard, so I doubt that this is a reasonable direction. This was even known in 1993. [1] [1]: https://people.csail.mit.edu/rivest/pubs/BR93.pdf https://people.csail.mit.edu/rivest/pubs/BR93.pdf