Microsoft Corporation
Machine-based recognition and dynamic selection of subpopulations for improved telemetry

Abstract:

Techniques for telemetry data collection including receiving first telemetry data items each including a first value for a first performance measurement, generating respective data objects providing values for a first set of attributes including multiple context attributes and a performance attribute determined based on the first value, performing a machine-based selection of outlier data objects from the data objects, obtaining a targeting rule specifying selection conditions for at least two of the context attributes and corresponding to a selected subset of the outlier data objects, providing the targeting rule to multiple telemetry sources, and receiving second telemetry data items each associated with a different telemetry source that satisfied the selection conditions, and each including a second value for a second performance measurement, with the first telemetry data items not including values for the second performance measurement.