神经网络 vs Genetic Algorithms

这是一个受 Google Tensorflow 的 Deep playground 启发而制作的应用。本程序大量代码都借鉴了开源项目 Deep playground 。 本程序致力于普及神经网络、人工智能的相关知识。通过本程序可以直观地感受到神经网络训练过程中的变化，包括连接的强弱、每个结点负责识别的特征和最终输出等图像等。 开源地址：https://github.com/ypwhs/NNPlayground_iPad

Genetic algorithms are one of the search and optimisation methods. The aim of optimisation is to increase efficiency in reaching a certain optimal value. Genetic algorithms are based on the mechanisms of natural selection and heredity. The basic genetic algorithm is built from three operations: reproduction, crossing, and mutation. Genetic algorithms operate on populations of coding sequences and use random selection rules to search for the global optimal value. However, these random rules are defined to give the appropriate direction of the search (through various selection mechanisms or scaling of fitness functions). This basic procedure is enhanced by certain genetic manipulations, such as those seen in nature. They include the mechanisms of dominance, diploidy, inversion, and other reconfiguration mechanisms that occur at the chromosome level. Users can design textual or graphical scaling patterns for the fitness function and then use them in simulations. The application also provides a preliminary analysis of statistical data concerning the distribution of the fitness function in the population. The program allows the user to compare non-random procedures (e.g. scaling of fitness functions) by using the same pseudorandom sequence (for procedures requiring randomization, such as selection, etc.) in successive simulations. In order to better visualize the genetic processes taking place during the simulation, some of them are presented to the user using animations. Additionally, some of the data obtained from simulations can be saved and shared.