Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

A functional geometric approach to distributed support vector machine (SVM) classification

Kampioti Sofia

Full record

---

Summary

We live in the information age, and with every passing year, our environment becomes more and more heavily defined by data, leading to a major need for better decision-making models. The breakthroughs in data analytics have already seen through machine learning. Support vector machines (SVM) are a popular, adaptive, multipurpose machine learning algorithm with the ability to capture complex relationships between data points without having to perform difficult transformations. We study the problem of prohibitive communication costs that a centralized architecture implies if most of the data is generated or received on different remote machines. The past few years notable efforts have been made to achieve parallelism on the training procedure of machine learning models. We propose the use of Functional Geometric Monitoring (FGM) communication protocol which is used to monitor high-volume, rapid distributed streams to decrease the communication cost on a distributed SVM architecture. Our main goal is both to achieve centralized-like prediction loss and to minimize communication costs. In our proposal, the sklearn library, for centralized machine learning, is used in a distributed manner, with the use of Dask library, resulting in a notable speedup for the training procedure.

Available Files

Services