Reinforcement Learning: A Tutorial.
Mance E. Harmon, Stephanie S. Harmon · 1997
sumed that the reader has some knowledge of learning algorithms that rely on gradient descent (such as the backpropagation of errors algorithm). 1 Introduction There are many unsolved problems that computers could solve if the appropriate software existed. Flight control systems for aircraft, automated manufacturing systems, and sophisticated avionics systems all present difficult, nonlinear control problems. Many of these problems are currently unsolvable, not because current computers are too slow or have too little memory, but simply because it is too difficult to determine what the program should do. If a computer could learn to solve the problems through trial and error, that would be of great practical value. Reinforcement Learning is an approach to machine intelligence that combines two disciplines to successfully solve problems that neither discipline can address individually. Dynamic Programming is a field of mathematics that has traditionally