In May 2017, Google revealed AutoML, an automated machine learning system that could create an artificial intelligence solution without the assistance of a human engineer. IBM Cloud and Amazon Web Services (AWS) offer machine learning solutions that do not require AI developers. GitHub and other cloud platforms already provide thousands of machine learning programs, reducing the need of having an AI expert at hand. These cloud platforms will slowly but surely reduce the need for artificial intelligence developers. Google Cloud's AI provides intuitive machine learning services. Microsoft Azure offers user-friendly machine learning interfaces.
At the same time, Massive Open Online Courses (MOOC) are flourishing everywhere. Anybody anywhere can pick up a machine learning solution on GitHub, follow a MOOC without even going to college, and beat any engineer to the job.
Today, artificial intelligence is mostly mathematics translated into source code which makes it difficult to learn for traditional developers. That is the main reason why Google, IBM, Amazon, Microsoft, and others have ready-made cloud solutions that will require fewer engineers in the future.
As you will see, starting with this chapter, you can occupy a central role in this new world as an adaptive thinker. There is no time to waste. In this chapter, we are going to dive quickly and directly into reinforcement learning, one of the pillars of Google Alphabet's DeepMind asset (the other being neural networks). Reinforcement learning often uses the Markov Decision Process (MDP). MDP contains a memoryless and unlabeled action-reward equation with a learning parameter. This equation, the Bellman equation (often coined as the Q function), was used to beat world-class Atari gamers.
The goal here is not to simply take the easy route. We're striving to break complexity into understandable parts and confront them with reality. You are going to find out right from the start how to apply an adaptive thinker's process that will lead you from an idea to a solution in reinforcement learning, and right into the center of gravity of Google's DeepMind projects.
The following topics will be covered in this chapter:
- A three-dimensional method to implement AI, ML, and DL
- Reinforcement learning
- MDP
- Unsupervised learning
- Stochastic learning
- Memoryless learning
- The Bellman equation
- Convergence
- A Python example of reinforcement learning with the Q action-value function
- Applying reinforcement learning to a delivery example
