Amazon Technologies, Inc.

Mohammed Talal Yassar Azam

Ketan Vijayvargiya

Mohammad Adnan

Ahmed Gamal Hamed

Fnu Syed Furqhan Ulla

Ankur Mehrotra

Aditya Bahuguna

Laurent Callot

Jill Blue Lin

Devesh Ratho

Patrick Bloebaum

Lenon Alexander Minorics

David Kernert

Sadanand Murthy Sachidananda

Ankita Verma

Ali Caner Turkmen

Mohammed Saad Ather

Dominik Janzing

Peter Goetz

Syed Ahsan Ishtiaque

Shekhar Agrawal

Shashank Srivastava

Techniques for detecting regressions with respect to the accuracy of an anomaly detection compute service in detecting anomalies in users' time series data. The techniques include providing an instrumented time series instrumented with a set of one or more anomalies to the anomaly detection service. The anomaly detection service detects a set of one or more anomalies in the instrumented time series. The precision and recall of the detected anomalies with respect to the instrumented anomalies is computed. From the computed precision and recall, an anomaly detection accuracy is computed as an F-score or F-measure. It is then determined whether a regression in anomaly detection accuracy has occurred by comparing the computed accuracy score to a threshold. If a regression has occurred, an alert can be generated or a recent change to the anomaly detection service can be rolled back.

Accuracy regression detection for time series anomaly detection compute services

Techniques for anomaly and causality detection are described. An example includes receiving time series data; performing anomaly detection on the received time series data to detect at least one anomaly using an anomaly detection model; detecting a causal relationship between measures, wherein a set of measures are related when a first of the set of measures has a causal influence on a second of the set of measures, wherein a single time series is a metric and a measure is a numerical or categorical quantity a metric describes; and outputting a result of the anomaly and causality relationship detections.

Anomaly and causality detection

Systems and methods are described for to detecting anomalies in various forms of data, including non-time series data. In one example, a data ingestion interval for customer data may be determined, where the data ingestion interval specifies a frequency at which data is analyzed to detect analogies in portions of the data corresponding to time windows. A portion of the customer data may then be obtained and aggregated from a data source according to the data ingestion interval. The portion of data may be converted into time series data by appending a time stamp corresponding to the time window of the portion of the data. The anomaly detection service may then process the time series data, using a time series data anomaly model, to detect one or more anomalies in the time series data.

Converting non time series data to time series data

A service to provide anomaly detection receives a request to back-test the service. The request includes information for accessing a dataset of historical data. The service executes workflows to ingest the data, train a plurality of machine learning models to perform anomaly detection, and detect anomalies in the dataset. A representation of the detect anomalies is generated and presented to a user. The service receives an indication to activate the service to provide ongoing anomaly detection services.

Anomaly back-testing

Systems and methods are described for using a recurring event based scheduler to continuously monitor data to detect anomalies within the data. In some aspects, an anomaly detection schedule may be determined for monitoring time series data to detect anomalies based on a data ingestion interval. A plurality of anomaly detection events may be sequentially generated and stored in an event queue at times specified by the anomaly detection schedule. The anomaly detection events may then be processed sequentially from the event queue to trigger execution of a plurality of anomaly detection workflow tasks at the times specified by the anomaly detection schedule. In some cases, execution of individual anomaly detection workflow tasks causes individual portions of time series data to be obtained from a customer data source and processed by an anomaly detection model to detect anomalies in the time series data.

Continuous execution engine algorithm

Growing community of inventors

Snoqualmie, WA, United States of America

Mohammed Talal Yassar Azam