Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/451A5CA8-A6BE-449C-9B8D-EBC95D16466C | ||
| Year | 2021 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Eftychia Seisaki, "Distributed multivariate regression via functional geometric monitoring", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2021 https://doi.org/10.26233/heallink.tuc.91161 | ||
| Appears in Collections |
Multivariate linear regression is an important and massively used technique formodeling and predicting data behavior in many fields. In scenarios where the data evolves over time, it is essential to monitor the model in order to identify possible changes. This becomes more challenging, when the data is distributed at a number of different nodes and the regression model must be recomputed to avoid inaccuracy. In such dynamic settings, data centralization and periodic model recomputation can be wasteful. Therefore, the goal is to develop a technique which conserves a precise approximation of the model over the union of all nodes’ data in a communicationefficient fashion. We propose a monitoring algorithm for multivariate regression models of distributed data streams, based on the basic notions of Functional Geometric Monitoring (FGM), which guarantees a bounded model error and demands communication only when the estimated model has fairly departed from the current global. Our experimental results clearly demonstrate a reduction in communication cost while maintaining the desired model accuracy, compared to similar existing models.
Copyright © DIAS 2013
