Re:Tricks That Can Outsmart Deepfake Videos for Now
by lmm on Saturday October 20, 2018 @04:54PM
(#18265410)
Is there any reason to believe generating 2D video with the properties that a recording of a 3D world has is a difficult mathematical problem though?
Re:Tricks That Can Outsmart Deepfake Videos for Now
by amluto on Saturday October 20, 2018 @01:25PM
(#18264309)
But you can solve TSP pretty well with a TSP solver. I’d be impressed if a neural network did that well. But maybe it’s possible.
Re:Tricks That Can Outsmart Deepfake Videos for Now
by currymj on Saturday October 20, 2018 @03:53PM
(#18265080)
See Dai et al, "Learning Combinatorial Optimization Algorithms Over Graphs", experiment section A.1.3 "Traveling Salesman Problem".
https://www.cc.gatech.edu/~lsong/papers/arxiv_rl_combopt.pdf
It does at least respectably well against the Concorde TSP solver although (unsurprisingly) doesn't beat it. Note that this also isn't a neural network that just maps a graph to a TSP solution; instead, the neural network function is a heuristic that guides the greedy construction of a solution one vertex at a time.
You can't solve the travelling salesman problem in polynomial time just by throwing machine learning at it...
On a mathematical level you can prove that machine learning will do a bad job for certain things, and this is what the parent talked about.