Rootconf proposals for round the year in 2018

A project to automatically detect anomalies in large-scale time-series data. It works by first building a reusable Deep-Learning model which is trained on seasonal data and used to create a one-day predict-ahead time series. Then a number of errors E are computed by comparing the expected value with the actual value at time t. The thresholds on E are automatically defined and most probable actionable anomalies are tagged. The drift is learned via predict ahead Deep-Learning, while the trend is learned using seasonality. Suitable for both system and business metrics and works best with correlated dashboards for both kind of metrics to establish a cause-effect relationship.

Outline

A project to automatically detect anomalies in large-scale time-series data. It works by first building a reusable Deep-Learning model which is trained on seasonal data and used to create a one-day predict-ahead time series. Then a number of errors E are computed by comparing the expected value with the actual value at time t. The thresholds on E are automatically defined and most probable actionable anomalies are tagged. The drift is learned via predict ahead Deep-Learning, while the trend is learned using seasonality. Suitable for both system and business metrics and works best with correlated dashboards for both kind of metrics to establish a cause-effect relationship.

Speaker bio

Senior Systems Engineer at Media.net (Directi), working on Deep Learning in Time-series data and Large scale distributed Deep Packet Inspection for Intrusion detection and performance troubleshooting systems