公开(公告)号：US11531917B1

公开(公告)日：2022-12-20

申请号：US16147147

申请日：2018-09-28

Applicant: Amazon Technologies, Inc.

Inventor： Jan Gasthaus ,  Konstantinos Benidis ,  Yuyang Wang ,  David Salinas ,  Valentin Flunkert

IPC: G06N3/08 ,  G06N7/00

Abstract: Techniques are described for a time series probabilistic forecasting framework that combines recurrent neural networks (RNNs) with a flexible, nonparametric representation of the output distribution. The representation is based on the nonparametric quantile function (instead of, for example, a parametric density function) and is trained by minimizing a continuous ranked probability score (CRPS) derived from the quantile function. Unlike methods based on parametric probability density functions and maximum likelihood estimation, the techniques described herein can flexibly adapt to different output distributions without manual intervention. Furthermore, the nonparametric nature of the quantile function provides a significant boost in the approach's robustness, making it more readily applicable to a wide variety of time series datasets.

公开(公告)号：US11537874B1

公开(公告)日：2022-12-27

申请号：US16101118

申请日：2018-08-10

Applicant: Amazon Technologies, Inc.

Inventor： Yuyang Wang ,  Alexander Johannes Smola ,  Dean P. Foster ,  Tim Januschowski

IPC: G06N3/08 ,  G06N7/00 ,  G06K9/62 ,  G06Q10/04 ,  G06N20/00

Abstract: Techniques for forecasting using deep factor models with random effects are described. A forecasting framework combines the strengths of both classical and neural forecasting methods in a global-local framework for forecasting multiple time series. A global model captures the common latent patterns shared by all time series, while a local model explains the variations at the individual level.

公开(公告)号：US12189506B1

公开(公告)日：2025-01-07

申请号：US18082521

申请日：2022-12-15

Applicant: Amazon Technologies, Inc.

Inventor： Xiao Wang ,  Jasmeet Chhabra ,  Yuyang Wang ,  Willem Conradie Visser

IPC: G06F11/30 ,  G06F11/07

Abstract: Systems and methods are described relating to aggregating log anomalies. In some examples, a plurality of log anomaly instances may be obtained, from a log anomaly detector, where individual instances are associated with a first log anomaly type and a first anomalous log event. Log anomaly instances associated with the first log anomaly type and the first anomalous log event may be combined into a first log anomaly class. The first log anomaly class may be combined with a second log anomaly class, including log anomaly instances associated with the first anomalous log event and a second log anomaly type, into a log anomaly group, which may correlate the occurrences of the first and second anomaly types to the same first anomalous log event over a period of time. An indication of the log anomaly group may then be output.