US 11,636,377 B1Granted

Artificial intelligence system incorporating automatic model updates based on change point detection using time series decomposing and clustering

AssigneeAmazon Technologies, Inc.

InventorsQiuping Xu, Joshua Allen Edgerton

Life of this patent

Estimating the term…

Loading patent…

Abstract

Computer systems and associated methods are disclosed to detect a future change point in time series data used as input to a machine learning model. A forecast for the time series data is generated. In some embodiments, a fitting model is generated from the time series data, and residuals of the fitting model are obtained for respective portions of the data both before and after a potential change point in the future. The change point is determined based on a ratio of residual metrics for the two portions. In some embodiments, data features are extracted from individual segments in the time series data, and the segments are clustered based on their data features. A change point is determined based on a dissimilarity in cluster assignments for segments before and after the point. In some embodiments, when a change point is predicted, an update of the machine learning model is triggered.

Claims20

Dependent claims

A system, comprising:

one or more computers configured to implement a first machine learning model that consumes a time series data to generate output;
one or more other computers configured to implement a change point detection service, configured to:
monitor the time series data using a second machine learning model to generate, via a forecasting technique implemented by the second machine learning model, future data for the time series data consumed by the first machine learning model;
group the time series data and the future data into a plurality of segments, including one or more future segments that contain at least some of the future data;
extract data features of individual ones of the segments;
group the segments into a plurality of clusters via a clustering technique applied to the data features;
determine a dissimilarity metric that quantifies a dissimilarity between a first cluster assignment distribution of segments before the point and a second cluster assignment distribution of segments that include or occur after the point;
and
determine that a point in a future segment represents a future change point of the time series data based at least in part on the dissimilarity metric exceeding a threshold;
and
wherein the first machine learning model is updated by a model updater in response to the determination of the future change point of the time series data to adapt the first machine learning model to a changing condition of the time series data at the future change point.

8 more elements

A method, comprising:

monitoring a time series data consumed by a first machine learning model, wherein the monitoring includes generating future data for the time series data via a forecasting technique implemented by a second machine learning model;
grouping the time series data and the future data into a plurality of segments, including one or more future segments that contain at least some of the future data;
extracting data features of individual ones of the segments;
grouping the segments into a plurality of clusters via a clustering technique applied to the data features;
determining a dissimilarity metric that quantifies a dissimilarity between a first cluster assignment distribution of segments before the point and a second cluster assignment distribution of segments that include or occur after the point;
determining that a point in a future segment represents a future change point of the time series data based at least in part on the dissimilarity metric exceeding a threshold;
and
in response to the determination that the point represents a future change point, updating the first machine learning model to adapt the first machine learning model to a changing condition of the time series data at the future change point.

5 more elements

A non-transitory computer-accessible storage medium storing program instructions that when executed on one or more processors cause the one or more processors to:

monitor a time series data consumed by a first machine learning model to generate future data for a time series data via a forecasting technique implemented by a second machine learning model;
group the time series data and the future data into a plurality of segments, including one or more future segments that contain at least some of the future data;
extract data features of individual ones of the segments;
group the segments into a plurality of clusters via a clustering technique applied to the data features;
determine a dissimilarity metric that quantifies a dissimilarity between a first cluster assignment distribution of segments before the point and a second cluster assignment distribution of segments that include or occur after the point;
determine that a point in a future segment represents a future change point of the time series data based at least in part on the dissimilarity metric exceeding a threshold;
and
in response to the determination of the future change point of the time series data, update the first machine learning model to adapt the first machine learning model to a changing condition of the time series data at the future change point.

5 more elements

Dependencies are inferred from the claim text.