Introduction

In the previous chapter, we looked at various policy-based methods and their advantages. In this chapter, we are going to learn about gradient-free methods, namely genetic algorithms; develop these algorithms step by step; and use them to optimize neural networks and RL-based algorithms. This chapter discusses the limitations of gradient-based methods, such as getting stuck at local optima and slower convergence when dealing with noisy input. This chapter presents an alternative optimization solution to gradient methods through genetic algorithms, as they ensure global optimum convergence. You will examine and implement the structure of genetic algorithms and implement them through hyperparameter selection for neural networks and evolving network topologies, as well as using them in combination with RL for a cart-pole balancing activity. Hybrid neural networks that use genetic algorithms are used to solve complex problems, such as modeling plasma chemical reactors, designing...