Practical AI on the Google Cloud Platform

Chapter 9. Putting AI Assistants to Work

Throughout this book, we have used a variety of deep learning architectures and systems to do our bidding. Many of the applications we have developed have obvious use cases for commercialization, while some do not. In the previous chapter, we introduced the concept of smarter AI assistants that could be powered by deep reinforcement learning. For the examples in that chapter, we looked at using AI agents to play games and solve puzzles. The purpose of that was to demonstrate how AI was evolving into something beyond rules-based, or supervised, learning. As we learned, reinforcement learning allows us to build constantly learning and evolving agents. In this chapter, we extend that concept to a full agent assistant that can recommend the food you should or shouldn’t eat.

In this chapter, we build the Eat/No Eat agent assistant. The purpose of this smart assistant is to monitor your food intake and suggest which dishes you should or shouldn’t consume. The agent will be able to do this just by looking at a picture of the food you plan to eat. Building an agent that can consume images of food and decide whether or not the user eats is no easy task. It will require us to revisit many of the things we’ve learned in this book.

We will start this chapter off by looking at what food datasets may be good candidates for a basis of training the agent. Reinforcement learning is powerful, but it also requires extensive training. In this case, we are ...