Abstract: Distributed anomaly detection using combinable measurement value summaries is disclosed. A computing device obtains a global combined measurement value summary. The global combined measurement value summary includes a combined measurement value summary of a plurality of measurement values taken by a plurality of other computing devices measuring a first type of item. The computing device obtains a measurement value of an item of the first type. It is determined whether the new measurement value is an anomalous measurement value based at least partially on the global combined measurement value summary.

Abstract: Distributed anomaly detection using combinable measurement value summaries is disclosed. A computing device obtains a global combined measurement value summary. The global combined measurement value summary includes a combined measurement value summary of a plurality of measurement values taken by a plurality of other computing devices measuring a first type of item. The computing device obtains a measurement value of an item of the first type. It is determined whether the new measurement value is an anomalous measurement value based at least partially on the global combined measurement value summary.

Abstract: Performing usage-based software library decomposition is disclosed herein. In some examples, a processor device generates a first library graph representing a first software library including multiple functions. The first library graph comprises a plurality of nodes that each correspond to a function of the first software library. The processor device identifies a function within the first software library (“invoked function”) that is directly invoked by an application that depends on the first software library, then generates a call graph including nodes within the first library graph (“dependency nodes”) corresponding to either the invoked function or another function invoked by the invoked function during application execution. Using the call graph, the processor device generates a second software library including only functions of the first software library corresponding to dependency nodes of the call graph.

Abstract: Performing usage-based software library decomposition is disclosed herein. In some examples, a processor device generates a first library graph representing a first software library including multiple functions. The first library graph comprises a plurality of nodes that each correspond to a function of the first software library. The processor device identifies a function within the first software library (“invoked function”) that is directly invoked by an application that depends on the first software library, then generates a call graph including nodes within the first library graph (“dependency nodes”) corresponding to either the invoked function or another function invoked by the invoked function during application execution. Using the call graph, the processor device generates a second software library including only functions of the first software library corresponding to dependency nodes of the call graph.

Abstract: Distributed anomaly detection using combinable measurement value summaries is disclosed. A computing device obtains a global combined measurement value summary. The global combined measurement value summary includes a combined measurement value summary of a plurality of measurement values taken by a plurality of other computing devices measuring a first type of item. The computing device obtains a measurement value of an item of the first type. It is determined whether the new measurement value is an anomalous measurement value based at least partially on the global combined measurement value summary.

Abstract: Automatic detection of learning model drift is disclosed. A sidecar learning model receives operational input data submitted to a predictive learning model. The sidecar learning model was trained on a same training data used to train the predictive learning model. A deviation of the operational input data from the training data is determined. The sidecar learning model generates a drift signal that characterizes the deviation of the operational input data from the training data.