Chapter 6. Designing a neural network for Go data

 

This chapter covers

  • Building a deep-learning application that can predict the next Go move from data
  • Introducing the Keras deep-learning framework
  • Understanding convolutional neural networks
  • Building neural networks to analyze spatial Go data

In the preceding chapter, you saw the fundamental principles of neural networks in action and implemented feed-forward networks from scratch. In this chapter, you’ll turn your attention back to the game of Go and tackle the problem of how to use deep-learning techniques to predict the next move for any given board situation of a Go game. In particular, you’ll generate Go game data with tree-search techniques developed in chapter 4 that you can then use to train a neural network. Figure 6.1 gives an overview of the application you’re going to build in this chapter.

Figure 6.1. How to predict the next move in a game of Go by using deep learning

As figure 6.1 illustrates, to connect your working knowledge of neural networks from the preceding chapter, you have to address a few critical steps first:

6.1. Encoding a Go game position for neural networks

6.2. Generating tree-search games as network training data

6.3. Using the Keras deep-learning library

6.4. Analyzing space with convolutional networks

6.5. Predicting Go move probabilities

6.6. Building deeper networks with dropout and rectified linear units

6.7. Putting it all together for a stronger Go move-prediction network

6.8. Summary

sitemap