A Study on Optimization Algorithm (OA) in Machine Learning and Hierarchical Information

摘要

Genetic Algorithm is a division of machine learning, where the computers are programmed to teach themselves to complete the given task over time. In our project, we simulate many rockets to fly towards the target specified. Genetic algorithm revolves around three main concepts. First generate a population of random rockets that fly in random directions. Each rocket is implemented as an array of Vectors, where each vector points to a specific direction at a given time. We then apply a fitness function that calculates the best performing rockets in each generation. With the fitness function, we now select the best rockets with which we form the next population. This involves two steps: First step is the crossover. Choose two parents i.e., two rockets and use their vector values to create a child rocket. This is done by retrieving the first half vectors from the first parent and second half vectors from the second parent and fuses them to build the child rocket, Second step is the mutation. This step is very crucial. If mutation is not applied, we will receive a new population that is only built around best performing ones from the previous population. We will then land in local maxima and may never reach the target. Mutation helps create individual rockets that go beyond the local maxima to reach the target. But over mutation will lead to too much diversity that is not beneficial to the system. Thus, define a mutation rate that is optimally balanced. In mutation, we choose a rocket with random probability, and alter its vector values randomly. This new population of rockets forms the next generation.