[jvanderbot]

AI + classical algorithms is my sweet daydream. Trained heuristics (even better domain specific ones), deployed for classical A*, ILP families, focal search, etc etc.

That is going to be really amazing.

[blt]

It is happening. There are papers on deep learning to improve the variable choice in branch-and-bound, etc

[jvanderbot]

Yeah my first exposure was from Marco Pavone's lab, on ML heuristics for MICP in the context of a cubical tumbling robot, IIRC. Really cool stuff.

[amelius]

Even solving large linear systems would already be amazing. But a SAT solver would be nice too :)

[jvanderbot]

I misunderstand your comment. We have those solvers. I'm suggesting AI would plug into existing solvers. This is a ripe area for research.