3 ms·
I see no reason the two approaches couldn't be combined - use machine learning to identify new patterns and weights to help evaluate positions, but compile them
by thom 9y ago
I see no reason the two approaches couldn't be combined - use machine learning to identify new patterns and weights to help evaluate positions, but compile them into efficient operations to use them in existing engines.
- dragontamer 9y agoThere's no easy way to turn a set of 10-layer Neural Network weights into faster optimized assembly language. You can optimize the Neural Network so that maybe it runs faster (using NN specific hardware like maybe TensorCores or whatever), but I don't think anyone has any way to "generalize" a Neural Network into faster code.