As an ML/AI guy, I'm curious about the problems encountered by HFT systems. What I would love to see is some data, along with the desired outcome. Maybe also include the amount of time available to make the said decision. One could go further and institute a "Netflix prize" with such data.

Right now, I'm clueless about what kind of data these systems deal with, and what are they supposed to do.

Any HFT/ULLT people willing to share data?