3 ms·
Tweaking 't' here seems to be similar to tweaking the learning rate in gradient descent. But in gradient descent you assume the problem is convex. Is there any
by rahimnathwani 10y ago
Tweaking 't' here seems to be similar to tweaking the learning rate in gradient descent. But in gradient descent you assume the problem is convex.
Is there any way to tell whether this logic problem (or any other) is convex? Is it possible that the correct solution's 5-away neighbours all have very high costs (like 9+)? What prevents you from doing worse that brute force?