|
|
|
| e-Pub |
Previous | 
Home
Section: New Results
Data Analytics
Massively Distributed Indexing of Time Series
Participants : Djamel-Edine Yagoubi, Reza Akbarinia, Boyan Kolev, Oleksandra Levchenko, Florent Masseglia, Patrick Valduriez, Dennis Shasha.
Indexing is crucial for many data mining tasks that rely on efficient and effective similarity query processing. Consequently, indexing large volumes of time series, along with high performance similarity query processing, have became topics of high interest. For many applications across diverse domains though, the amount of data to be processed might be intractable for a single machine, making existing centralized indexing solutions inefficient.
In [36], we consider the problem of finding highly correlated pairs of time series across multiple sliding windows. Doing this efficiently and in parallel could help in applications such as sensor fusion, financial trading, or communications network monitoring, to name a few. We have developed a parallel incremental random vector/sketching approach, called ParCorr, to this problem and compared it with the state-of-the-art nearest neighbor method iSAX. Whereas iSAX achieves 100% recall and precision for Euclidean distance, the sketching approach is, empirically, at least 10 times faster and achieves 95% recall and 100% precision on real and simulated data. For many applications this speedup is worth the minor reduction in recall. Our method scales up to 100 million time series and scales linearly in its expensive steps (but quadratic in the less expensive ones).
In [48], we propose a demonstration of our sketch-based solution to efficiently perform both the parallel indexing of large sets of time series and a similarity search on them. Because our method is approximate, we explore the tradeoff between time and precision. A video showing the dynamics of the demonstration can be found at http://parsketch.gforge.inria.fr/video/parSketchdemo_720p.mov.
Parallel Mining of Maximally Informative k-Itemsets in Data Streams
Participants : Mehdi Zitouni, Reza Akbarinia, Florent Masseglia.
The discovery of informative itemsets is a fundamental building block in data analytics and information retrieval. While the problem has been widely studied, only few solutions scale. This is particularly the case when the dataset is massive, or the length k of the informative itemset to be discovered is high.
In [63], we address the problem of mining maximally informative k-itemsets (miki) in data streams based on joint entropy. We propose PentroS, a highly scalable parallel miki mining algorithm. PentroS renders the mining process of large volumes of incoming data very efficient. It is designed to take into account the continuous aspect of data streams, particularly by reducing the computations of need for updating the miki results after arrival/departure of transactions to/from the sliding window. PentroS has been extensively evaluated using massive real-world data streams. Our experimental results confirm the effectiveness of our proposal which allows excellent throughput with high itemset length.
Spatio-Temporal Data Mining
Participants : Esther Pacitti, Florent Masseglia.
The problem of discovering spatiotemporal sequential patterns affects a broad range of applications. Many initiatives find sequences constrained by space and time. We address in [40] an appealing new challenge for this domain: find tight space-time sequences, i.e., find within the same process: i) frequent sequences constrained in space and time that may not be frequent in the entire dataset and ii) the time interval and space range where these sequences are frequent. The discovery of such patterns along with their constraints may lead to extract valuable knowledge that can remain hidden using traditional methods since their support is extremely low over the entire dataset. Our contribution is a new Spatio-Temporal Sequence Miner (STSM) algorithm to discover tight space-time sequences.