12 Evolutionary machine learning and beyond

 

This chapter covers

  • Evolution and ML with gene expression programming
  • Revisiting reinforcement learning with Geppy
  • Instinctual learning
  • Generalized learning with genetic programming
  • The future of evolutionary machine learning
  • Generalization with instinctual deep and deep reinforcement learning

In the last chapter, we looked deeply at how evolutionary solutions like NEAT could be applied to solve RL. In this chapter, we continue with some of those same concepts but also take a step back and look at how evolutionary methods can be applied to expand our understanding of ML. Specifically, looking at what role evolutionary search plays can expand how we develop generalized ML.

Generalized ML (aka generalized AI)

Generalized machine learning, or generalized intelligence, is an area that focuses on building models that can solve more than one task. Typically, in ML, we develop models to classify or regress on a single source of data, training the model iteratively and validating performance with similar data. Generalized ML’s goal is to develop models that can predict over multiple disparate forms of data or environments. In data science, you may hear this problem referred to as cross-domain or multimodal, meaning we are building a model to tackle problems across domains.

12.1 Evolution and machine learning with gene expression programming

12.1.1 Learning exercises

12.2 Revisiting reinforcement learning with Geppy

12.2.1 Learning exercises

12.3 Introducing instinctual learning

12.3.1 The basics of instinctual learning

12.3.2 Developing generalized instincts

12.3.3 Evolving generalized solutions without instincts

12.3.4 Learning exercises

12.4 Generalized learning with genetic programming

12.4.1 Learning exercises