公开(公告)号：US20240386047A1

公开(公告)日：2024-11-21

申请号：US18198975

申请日：2023-05-18

Applicant: Oracle International Corporation

Inventor： Ankit Aggarwal ,  Chirag Ahuja ,  Vikas Pandey ,  Sharmily Sidhartha ,  Hariharan Balasubramanian ,  Jie Xing

IPC: G06F16/38 ,  G06F16/31

Abstract: Techniques are described herein for cold-start forecasting datasets using backcasting and composite embedding. An example method can include a system receiving a set of time series and metadata text comprising a first subset of metadata text and a second subset of metadata text. The system can generate a plurality of embeddings, each embedding comprising a numerical representation of a metadata text of the set of metadata text. The system can generate a plurality of vectors, each vector comprising a time series of the set of time series each time series associated with a metadata text of the first subset of metadata text. The system can generate a plurality of composite embeddings based at least in part on combining each embedding with a respective vector of the plurality of vectors. The system can determine a forecasted value associated with the second subset of metadata text based on the composite embeddings.

公开(公告)号：US20240362210A1

公开(公告)日：2024-10-31

申请号：US18139492

申请日：2023-04-26

Applicant: Oracle International Corporation

Inventor： Ankit Aggarwal ,  Jie Xing ,  Chirag Ahuja ,  Vikas Pandey ,  Hariharan Balasubramanian

IPC: G06F16/242 ,  G06F16/2455

CPC classification number: G06F16/244 ,  G06F16/24553

Abstract: Techniques are described herein for forecasting datasets using blend of temporal aggregation and grouped aggregation. An example method can include a device accessing a first and second time series, comprising a first data point associated with a first time step and a first value and a second data point associated with a second time step and a second value. The method can further include the device determining a grouped aggregated data point using the first and second time series by aligning the first and second data point. The method can further include the device determining the grouped aggregated data point by summing the first and second value. The method can further include determining a grouped aggregated time series. The method can further include the device determining a first set of input values for a machine learning model. The method can further include the device determining a first forecasted future value.

公开(公告)号：US20250094861A1

公开(公告)日：2025-03-20

申请号：US18470220

申请日：2023-09-19

Applicant: Oracle International Corporation

Inventor： Ankit Kumar Aggarwal ,  Vikas Pandey ,  Chirag Ahuja ,  Jie Xing ,  Hariharan Balasubramanian

IPC: G06N20/00

Abstract: Techniques for time-bound hyperparameter tuning are disclosed. The techniques enable the determination of optimized hyperparameters for a machine learning (ML) model given a specified time bound using a three-stage approach. A series of trials are executed, during each of which the ML model is trained using a distinct set of hyperparameters. In the first stage, a small number of trials are executed to initialize the algorithm. In the second and third stages, a certain number of trials are executed in each stage. The number of trials to run in each stage are determined using one or more computer-implemented techniques. The computer-implemented techniques can also be used to narrow the hyperparameter search space and the feature space. Following the third stage, a set of optimized hyperparameters is adopted based a predefined optimization criterion like minimization of an error function.

公开(公告)号：US20230274195A1

公开(公告)日：2023-08-31

申请号：US17854482

申请日：2022-06-30

Applicant: Oracle International Corporation

Inventor： Ankit Kumar Aggarwal ,  Vikas Pandey ,  Praneet Pabolu ,  Samik Raychaudhuri

IPC: G06N20/20

CPC classification number: G06N20/20

Abstract: The present embodiments relate to using feature engineering to generate time-varying features via metadata. A first exemplary embodiment provides a method for performing feature engineering to generate time-varying features. The method can include receiving a first value and a second value of the time-series data. The method can further include receiving metadata that describes a relationship between the first value and the second value. The method can further include detecting the relationship between the first value and the second value based on the metadata. The method can further include generating, a time-varying feature from a combination of the first value and the second value based on the relationship detected from the metadata. The method can further include generating, by implementing the machine learning forecasting model, a forecasted value for the time-series data based on the time-varying feature.