| C. Karras, A. Karras, G. Drakopoulos, C. Makris, Ph. Mylonas, S. Sioutas |
| Weighted Reservoir Sampling On Evolving Streams: A Sampling Algorithmic Framework For Stream Event Identification |
| 12th Conference on Artificial Intelligence (SETN 2022), September 7-9, 2022, Corfu, Greece |
|
ABSTRACT
|
| Data streams are becoming increasingly important across a wide array of fields and are generally expected to be the preferred form of big data as aggregators and smart stream analytics in general can efficiently yield stream descriptions in various levels. Among them, event detection analytics are paramount since they typically allow the identification of distinct cases of interest like the so called black swans. Reservoir sampling refers to probabilistic class of techniques for keeping representative values of a stream given limited memory capacity. In the proposed framework event detection takes place once reservoir sampling is complete by clustering its output. The rationale behind this is that repeated representative values correspond to normal stream states, whereas any outliers indicate rare yet noteworthy events. With that information a probabilistic stream state graph can be constructed in order to examine the transition dynamics between states and to evaluate the role black swans play in the overall stream stability. A major part of the descriptive power of said graph lies on its inherent geometric interpretation in addition to the algebraic one. Results from two benchmark datasets, one coming from real world and a random one, are encouraging. The proposed framework is planned to be executed in Raspberry Pi as part of an IoT stack since it is sufficiently lightweight.
|
| 07 September, 2022 |
| C. Karras, A. Karras, G. Drakopoulos, C. Makris, Ph. Mylonas, S. Sioutas, "Weighted Reservoir Sampling On Evolving Streams: A Sampling Algorithmic Framework For Stream Event Identification", 12th Conference on Artificial Intelligence (SETN 2022), September 7-9, 2022, Corfu, Greece |
[  PDF] [
BibTex] [
Print] [
 Back] |
