Posted on Categories:CS代写, Machine Learning, 机器学习, 计算机代写

# 计算机代写|机器学习代写Machine Learning代考|COMP4702 Population-based optimization

avatest™

## avatest™帮您通过考试

avatest™的各个学科专家已帮了学生顺利通过达上千场考试。我们保证您快速准时完成各时长和类型的考试，包括in class、take home、online、proctor。写手整理各样的资源来或按照您学校的资料教您，创造模拟试题，提供所有的问题例子，以保证您在真实考试中取得的通过率是85%以上。如果您有即将到来的每周、季考、期中或期末考试，我们都能帮助您！

•最快12小时交付

•200+ 英语母语导师

•70分以下全额退款

## 计算机代写|机器学习代写Machine Learning代考|Population-based optimization

Stochastic local search (SLS) maintains a single “best guess” at each step, $\boldsymbol{x}_t$. If we run this for $T$ steps, and restart $K$ times, the total cost is $T K$. A natural alternative is to maintain a set or population of $K$ good candidates, $\mathcal{S}_t$, which we try to improve at each step. This is called an an evolutionary algorithm (EA). If we run this for $T$ steps, it also takes $T K$ time; however, it can often get better results than multi-restart SLS, since the search procedure explores more of the space in parallel, and information from different members of the population can be shared. Many versions of EA are possible, depending on how we update the population at each step, as we discuss in Section $6.3 .1$.

An alternative to maintaining an explicit set of promising candidates is to maintain a probability distribution over promising candidates. We call this distribution-based optimization (DBO). ${ }^2$ We can think of EAs as using a nonparametric representation of this distribution in terms of a “bag of points”. We can sometimes get better performance by using a parametric distribution, with suitable inductive bias. We discuss some examples in Section $6.3 .3$.

## 计算机代写|机器学习代写Machine Learning代考|Evolutionary algorithms

Since EA algorithms draw inspiration from the biological process of evolution, they also borrow a lot of its terminology. The fitness of a member of the population is the value of the objective function (possibly normalized across population members). The members of the population at step $t+1$ are called the offspring. These can be created by randomly choosing a parent from $\mathcal{S}_t$ and applying a random mutation to it. This is like asexual reproduction. Alternatively we can create an offspring by choosing two parents from $\mathcal{S}_t$, and then combining them in some way to make a child, as in sexual reproduction; combining the parents is called recombination. (It is often followed by mutation.)

The procedure by which parents are chosen is called the selection function. In truncation selection, each parent is chosen from the fittest $K$ members of the population (known as the elite set). In tournament selection, each parent is the fittest out of $K$ randomly chosen members. In fitness proportionate selection, also called roulette wheel selection, each parent is chosen with probability proportional to its fitness relative to the others. We can also “kill off” the oldest members of the population, and then select parents based on their fitness; this is called regularized evolution [Rea $+19]$ ).

In addition to the selection rule for patents, we need to specify the recombination and mutation rules. There are many possible choices for these heuristics. We briefly mention a few of them below.

In a genetic algorithm (GA) [Gol89; Hol92], we use mutation and a particular recombination method based on crossover. To implement crossover, we assume each individual is represented as a vector of integers or binary numbers, by analogy to chromosomes. We pick a split point along the chromosome for each of the two chosen parents, and then swap the strings, as illustrated in Figure 6.5.

In genetic programming [Koz92], we use use a tree-structured representation of individuals, instead of a bit string. This representation ensures that all crossovers result in valid children, as illustrated in Figure 6.7. Genetic programming can be useful for finding good programs as well as other structured objects, such as neural networks. In evolutionary programming, the structure of the tree is fixed and only the numerical parameters are evolved.

In surrogate assisted EA, a surrogate function $\hat{f}(s)$ is used instead of the true objective function $f(s)$ in order to speed up the evaluation of members of the population (see [Jin11] for a survey). This is similar to the use of response surface models in Bayesian optimization (??), except it does not deal with the explore-exploit tradeoff.

In a memetic algorithm [MC03], we combine mutation and recombination with standard local search.
Evolutionary algorithms have been applied to a large number of applications, including training neural networks (this combination is known as neuroevolution [Sta $+19]$ ). An efficient JAX-based library for (neuro)-evolution can be found at https://github.com/google/evojax.

## MATLAB代写

MATLAB 是一种用于技术计算的高性能语言。它将计算、可视化和编程集成在一个易于使用的环境中，其中问题和解决方案以熟悉的数学符号表示。典型用途包括：数学和计算算法开发建模、仿真和原型制作数据分析、探索和可视化科学和工程图形应用程序开发，包括图形用户界面构建MATLAB 是一个交互式系统，其基本数据元素是一个不需要维度的数组。这使您可以解决许多技术计算问题，尤其是那些具有矩阵和向量公式的问题，而只需用 C 或 Fortran 等标量非交互式语言编写程序所需的时间的一小部分。MATLAB 名称代表矩阵实验室。MATLAB 最初的编写目的是提供对由 LINPACK 和 EISPACK 项目开发的矩阵软件的轻松访问，这两个项目共同代表了矩阵计算软件的最新技术。MATLAB 经过多年的发展，得到了许多用户的投入。在大学环境中，它是数学、工程和科学入门和高级课程的标准教学工具。在工业领域，MATLAB 是高效研究、开发和分析的首选工具。MATLAB 具有一系列称为工具箱的特定于应用程序的解决方案。对于大多数 MATLAB 用户来说非常重要，工具箱允许您学习应用专业技术。工具箱是 MATLAB 函数（M 文件）的综合集合，可扩展 MATLAB 环境以解决特定类别的问题。可用工具箱的领域包括信号处理、控制系统、神经网络、模糊逻辑、小波、仿真等。