Neural network based reinforced learning

Reinforcement learning equations for modifying neural network backpropagation weights were derived. Subsequent convergence analysis showed guaranteed convergence. Experiments conducted in simulation and on a physical system showed the algorithm considered here learned to control quickly, quicker than other learning algorithms doing the same task. It could also adapt to changes in the physical parameters. There were also clear indications that the algorithm could generalize, and accommodate changes in the control environment, without the need for further training. This is due to the distributed knowledge representation ability supported by neural networks