2 ms·
Funny but I think that this competition is a blatant case for not using deep learning for everything. Apparently the software can have access to the depthbuffer
by Iv 10y ago
Funny but I think that this competition is a blatant case for not using deep learning for everything. Apparently the software can have access to the depthbuffer. That makes the "defend the center" exercise extremely trivial and should make aiming and shooting much better than the demo we see.
I think that a scripted AI will be much better than a blind deep-learnt one. A level of deep-learning may add a good layer of intelligence, but I think that building it on some optimal bricks would be a good start.
- ghgfhgfhn 10y agoPerhaps, but the idea is to work our way to general machines that can solve a range of problems. Learning to play chess with deep learning even though it is currently not going to beat a purely algorithmic approach is still worthwhile.
- modeless 10y agoDepth buffer access was disallowed for the competition. It would be sad if the result of the competition was a hand coded AI that would work on DOOM but fail on anything else. A pure deep learning approach should be easily generalizable to other games.