Supervised and unsupervised machine learning have had many successes in various areas, but their applications have mainly involved narrowly-defined tasks. To move a step higher in a learning scheme, machines must not only learn but also remember what they have learned previously. While companies are now using deep learning for various machine learning applications, reinforcement learning has found only limited use so far. But RL can be used in conjunction with neural networks, which have achieved recent breakthroughs in problems such as computer vision, to tackle complex problems.