EPITOME - Summarization and structuring of continuous data in concurrent processing pipelines

Synopses of data can provide ways for effective processing, structuring and analysis, either as analysis results per se, or as means to provide input in “distilled” form, for data mining and machine learning applications in high-rate data pipelines for digitized systems, without the need to log every single event.The EPITOME project targets methods to structure synopses of data for continuous, stateful, window-based analysis of continuous data, in presence of concurrency, which is a must, given the high-data rates and the parallelism available in all associated computing platforms.

Parallel processing requires maintaining shared information and can bring benefits when true concurrent data access is allowed, as opposed to the latter blocking each other or preventing each other from progressing.

Designing methods for concurrent data access implies tuning of trade-offs among the timeliness and accuracy of the results, consistency guarantees depending on allowed levels of concurrency and asynchrony, as well the cost in memory and processing capacity required.

The EPITOME project focuses on these issues, defining research for a PhD project, and targets methods for concurrent data synopses processing and structuring, to balance the implied trade-offs.This will facilitate cost-efficiency, energy-efficiency and usability of the data in digitized systems.