TechRxiv
manuscript_techrxiv.pdf (1.37 MB)

Streaming Algorithms for Subspace Analysis: an Edge- and Hardware-oriented review

Download (1.37 MB)
preprint
posted on 28.05.2021, 18:04 by Alex MarchioniAlex Marchioni, Luciano Prono, Mauro MangiaMauro Mangia, Fabio PareschiFabio Pareschi, Riccardo Rovatti, Gianluca SettiGianluca Setti
Subspace analysis is a basic tool for coping with high-dimensional data and is becoming a fundamental step in early processing of many signals elaboration tasks. Nowadays trend of collecting huge quantities of usually very redundant data by means of decentralized systems suggests these techniques be deployed as close as possible to the data sources. Regrettably, despite its conceptual simplicity, subspace analysis is ultimately equivalent to eigenspace computation and brings along non-negligible computational and memory requirements. To make this fit into typical systems operating at the edge, specialized streaming algorithms have been recently devised that we here classify and review giving them a coherent description, highlighting features and analogies, and easing comparisons. Implementation of these methods is also tested on a common edge digital hardware platform to estimate not only abstract functional and complexity features, but also more practical running times and memory footprints on which compliance with real-world applications hinges.

Funding

Smart-Data@PoliTO center

Italian Ministry for University and Research (MUR) - “Dipartimenti di Eccellenza (2018-2022)"

History

Email Address of Submitting Author

alex.marchioni@unibo.it

ORCID of Submitting Author

0000-0002-3441-3634

Submitting Author's Institution

University of Bologna

Submitting Author's Country

Italy