Try removing BN from the critic :)

> It is not for lack of trying that all the top papers in visual question answering end up doing this as a classification task. Results are really poor when it is used as RNN generation

I'd be curious to know if you have a reference for this. Given that the answers are one word, a word-level RNN language model output should basically be the same thing as a straight 1000-way softmax.

This is also not quite right. Models whose parameters are too big to fit on one GPU can be trained by splitting them across multiple GPUs, as was done here, for example: http://papers.nips.cc/paper/5346-sequence-to-sequence-learni...

Cool algorithm! One problem is that its probability density calculation assumes a random ship placement. If the opponent were aware of how the algorithm works (let's say they deduced it from its behavior, which seems plausible) then they could devise a counter-strategy of placing ships on the squares which have the fewest other possible ship placements (e.g. in the beginning of the game, this is along the edges of the board).

According to their "About" page, a one-way ride costs $22. Even if you buy 50 rides at once, the price is still $17.6 per one-way ride. That's comparable to what a typical UberX ride around Palo Alto costs, in my experience.

Patenting it doesn't necessarily mean they won't let other people use it. It could, for example, be for protection against Apple patenting it themselves and not letting anyone use it.

It's not a problem if those connections use self-signed certificates, right? If that's the case, then setting up SSL from CloudFlare to your servers should be pretty easy.

I don't know whether it's possible with ZIP, but there are definitely compression schemes for which this can be done in time proportional to the size of the compressed file.