reinforcement learning

References

Summary

K-Armed Bandit

One of the simplest problems

A slot machine with K levers

Temporal Difference Learning

You assign internal rewards at all of the hidden layer steps (steps before the final step)

The further the internal step is from the final step, the more its reward is decreased. So for instance, the final reward might be 100, the second last step gets 90, the third last gets 85, etc.

Bellman Equation

Backgammon

The TD-Gammon program from 1995 can become master-level after playing about 1.5M games with itself