The upcoming avenue of IoT, with its massive generated data, makes it really hard to train centralized systems with machine learning in real time. This problem can be addressed with learning-based edge computing systems where the learning is performed in a distributed way on the nodes. In particular, this work focuses on developing multi-agent systems for implementing learning-based edge computing systems. The diversity of methodologies in agent-oriented software engineering reflects the complexity of developing multi-agent systems. The division of the development processes into method fragments facilitates the application of agent-oriented methodologies and their study. In this line of research, this work proposes a database for implementing a repository of method fragments considering the development of learning-based edge computing systems and the information recommended by the FIPA technical committee. This repository makes method fragments available from different methodologies, and computerizes certain metrics and queries over the existing method fragments. This work compares the performance of several combinations of dimensionality reduction methods and machine learning techniques (i.e., support vector regression, k-nearest neighbors, and multi-layer perceptron neural networks) in a simulator of a learning-based edge computing system for estimating profits and customers.

中文翻译：

---

基于学习的边缘计算系统的面向代理开发的方法片段存储库


即将到来的物联网途径及其生成的海量数据，使得通过机器学习实时训练集中式系统变得非常困难。这个问题可以通过基于学习的边缘计算系统来解决，其中学习在节点上以分布式方式执行。特别是，这项工作侧重于开发多代理系统来实现基于学习的边缘计算系统。面向代理的软件工程方法的多样性反映了开发多代理系统的复杂性。将开发过程划分为方法片段有利于面向主体的方法的应用及其研究。在这一研究领域中，考虑到基于学习的边缘计算系统的发展以及 FIPA 技术委员会推荐的信息，这项工作提出了一个用于实现方法片段存储库的数据库。该存储库使不同方法中的方法片段可用，并对现有方法片段的某些指标和查询进行计算机化。这项工作比较了基于学习的边缘计算系统模拟器中降维方法和机器学习技术（即支持向量回归、k-近邻和多层感知器神经网络）的几种组合的性能，以估计利润和客户。