Ioannis Lamprinidis, "Distributed machine learning framework on Akka", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2024
https://doi.org/10.26233/heallink.tuc.101190
The proliferation of data-driven applications has led to a growing demand for efficient and scalable machine learning algorithms. This thesis delves into the design and implementation of a distributed communication kernel in Akka for the Online Machine Learning and Data Mining system(OMLDM), a system that supports distributed online learning by utilizing the Parameter Server paradigm, for effortlessly deploying Online Machine Learning pipelines on streaming platforms. The objective was the implementation of an efficient, scalable, fault tolerant and robust kernel for the OMLDM, to analyze the performance overhead of Akka by comparing it to a local implementation of the OMLDM kernel, that utilizes Java Threads; To evaluate the performance speedup achieved by the kernel in a cluster environment. We demonstrate through experiments the communication overhead of Akka and the performance of the kernel in local and clustered environments.