Background
Various systems and applications perform monitoring of data streams in a distributed environment. Such applications include, for example, sensor networked, distributed web-sites, etc. Existing solutions are either limited in the set of monitoring functions (most solutions monitor only linear functions) or limited in the number of distributed nodes and monitored objects. In addition, the computational cost required in existing solutions is very expensive when monitoring non-linear functions.

Method
This technology employs novel geometric techniques, based on Minkowski sum operation for intelligently decomposing the monitoring of complex holistic conditions and functions into safe, local constraints that can be tracked independently at each node. First, values are collected from each monitored data stream. Second, based on the data distribution in each node and the monitoring function, a set of safe zones is defined such that the Minkowski sum of all safe zones is contained in S, where S ={z | f (z) ≤ t}. Finally, the safe zones are assigned to each stream. While the stream data is in its safe zone, the Minkowski sums guarantees that the global value of the monitored function has not crossed the threshold value.

Advantages
• Lower costs because the local constraint violation verification is very low in computational costs compared to existing algorithms
• Greater control because this method allows the user to set a tradeoff between the number of false alarms i.e. type I errors and the number of missed detections i.e. type II errors. This ability is unavailable with existing solutions
• This method supports different data distribution in each node, as opposed to previous algorithms which required identical data distribution among all nodes

Applications
• There are numerous tasks to which the suggested method can be applied, such as: Feature selection, Classification with kernel SVM, Association rule mining/correlation, Top-k, Monitoring global conditions, Dynamic data streams, etc.
 

Technical Keywords: Minkowski sum, distributed dataset, threshold function, multiple nodes, dynamic data stream

Market Keywords: Data mining, dynamic data streams,