![Institutional Repository [SANDBOX]](images/dias.logo.png "Institutional Repository [SANDBOX]"){kind=logo}
Institutional Repository [SANDBOX]
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/371941FA-1DAC-45D2-8900-97FE91ADB9D9 | ||
| Year | 2023 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Michail Theologitis, "Algorithms for online federated machine learning", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2023 https://doi.org/10.26233/heallink.tuc.97515 | ||
| Appears in Collections |
In the contemporary landscape, the proliferation of data, driven by the internet, social media, and the Internet of Things (IoT), has catalyzed transformative shifts across various sectors, with Machine Learning (ML) standing at the forefront of this revolution. As the Big Data era fosters remarkable ML advancements in domains like image recognition and natural language processing, it also introduces substantial computational challenges, logistical hurdles, and privacy concerns. These complexities have pushed traditional ML approaches to the limit, often proving inadequate, prompting the rise of innovative paradigms such as Federated Learning (FL). We delve into the challenges posed by FL's traditional iterative training process, which is non-dynamic and prescribes predetermined operations for participating learners. We investigate a dynamic approach anchored in the principles of Functional Geometric Monitoring (FGM), a state-of-the-art technique for monitoring distributed data streams, with the aim to enhance the training process by significantly mitigating communication overhead. Through the lens of FGM and employing three approximation techniques, our work evaluates its efficacy in refining FL's conventional training procedures, supported by comprehensive experimental analyses.
Copyright © DIAS 2013
