3 ms·
The classic failure of this approach is a form of the hill climbing problem (fastest way from Cambridge to Palo Alto is to start driving east to the airport).
by gumby 3y ago
The classic failure of this approach is a form of the hill climbing problem (fastest way from Cambridge to Palo Alto is to start driving east to the airport).
- bschne 3y agoIt simply says that if you want to evaluate all paths from Cambridge to Palo Alto that go through the airport, you only need to consider those that use the quickest way of getting from Cambridge to the airport, because all others can't be faster than that. This is different from getting stuck in a local minimum, it's merely a dynamic programming approach to reducing the computational burden of finding the global one.
- benreesman 3y agoWith a little hand-waving, same insight as in Q-learning I think right?
- bschne 3y agoI'm unfortunately not familiar with that, but maybe someone else will see the connection