Anomaly Detection Systems


We Obtain Superior Performance Results with Right Approaches

Anomaly detection on multi-variate and multi-source time series data obtained from a real-life domain is not a trivial problem. Defining the anomaly plays a critical role towards developing correctly functional anomaly detection algorithms. If it is well defined, labeling the data set and using this labeled set to train algorithms yield much better results. We have developed various anomaly detection systems for real-time streaming multi-variate and multi-source data.

Our solutions to anomaly detection include:

  • Point anomaly detection: Deviation of a single sample or value from the distribution of the rest of the data.
  • Contextual anomaly detection: Deviation of the series in spatial and/or temporal axes.
  • Group anomaly detection: Cases when each series has a nominal behavior but causes an anomalous behavior when considered with other series.

Among applications and use-cases of our solutions are:

  • Preprocessing and post-processing on systems during data processing and analysis steps
  • Real-time analysis, detection, warning, visualization and reporting solutions for diagnostic and prognostic tools
  • Environmental awareness for the end-user
  • Predictive maintenance applications

Video Anomaly Detection

Our video anomaly detection systems can detect multiple anomalies with split second sensitivity in a domain-agnostic manner. Our solutions can be easily integrated into any existing system and used directly.  Our systems are designed to be generic, though it is possible to tune the system for specific needs easily.

Point, Contextual and Group Anomaly on Big Data

We have solutions that can detect contextual and temporal deviations from the nominal patterns via deep learning-based approaches.