20240028944. ONLINE DRIFT DETECTION FOR FULLY UNSUPERVISED EVENT DETECTION IN EDGE ENVIRONMENTS simplified abstract (Dell Products L.P.)

From WikiPatents
Jump to navigation Jump to search

ONLINE DRIFT DETECTION FOR FULLY UNSUPERVISED EVENT DETECTION IN EDGE ENVIRONMENTS

Organization Name

Dell Products L.P.

Inventor(s)

Vinicius Michel Gottin of San Jose CA (US)

[[:Category:Herberth Birck Fr�hlich of Florianopolis (BR)|Herberth Birck Fr�hlich of Florianopolis (BR)]][[Category:Herberth Birck Fr�hlich of Florianopolis (BR)]]

Julia Drummond Noce of Rio de Janeiro (BR)

Ítalo Gomes Santana of Rio de Janeiro (BR)

ONLINE DRIFT DETECTION FOR FULLY UNSUPERVISED EVENT DETECTION IN EDGE ENVIRONMENTS - A simplified explanation of the abstract

This abstract first appeared for US patent application 20240028944 titled 'ONLINE DRIFT DETECTION FOR FULLY UNSUPERVISED EVENT DETECTION IN EDGE ENVIRONMENTS

Simplified Explanation

The abstract of this patent application describes a method for detecting drift in the performance of a model. The method involves receiving unlabeled data samples from the model and calculating a reconstruction error for these samples. A second reconstruction error is also calculated for a set of normative data. A margin is defined based on these reconstruction errors. The method then computes an initial proportion of normative data samples whose reconstruction errors fall within the margin. It also computes a new proportion of unlabeled data samples that fall within the margin. If the new proportion differs from the initial proportion by more than a predefined tolerance threshold, drift in the model's performance is signaled.

  • The method receives unlabeled data samples from a model.
  • It calculates a reconstruction error for the unlabeled data samples.
  • It calculates a second reconstruction error for a set of normative data.
  • A margin is defined based on the reconstruction errors.
  • An initial proportion of normative data samples within the margin is computed.
  • A new proportion of unlabeled data samples within the margin is computed.
  • Drift in the model's performance is signaled if the new proportion differs from the initial proportion by more than a predefined tolerance threshold.

Potential applications of this technology:

  • Monitoring the performance of machine learning models.
  • Detecting drift in data streams.
  • Quality control in manufacturing processes.

Problems solved by this technology:

  • Identifying when a model's performance is deteriorating.
  • Ensuring the accuracy and reliability of machine learning models.
  • Early detection of anomalies in data streams.

Benefits of this technology:

  • Improved performance monitoring and maintenance of machine learning models.
  • Timely detection of drift, allowing for prompt corrective actions.
  • Enhanced quality control and anomaly detection in various industries.


Original Abstract Submitted

one example method includes receiving a stream of unlabeled data samples from a model, obtaining a first reconstruction error for the unlabeled data samples, obtaining a second reconstruction error for a set of normative data, defining a margin based on the first reconstruction error and the second reconstruction error, computing an initial proportion of samples from the set of normative data whose reconstruction errors fall within a range of reconstruction errors defined by the margin, computing a new proportion of unlabeled data samples that fall within the range of reconstruction errors defined by the margin, and signaling drift in the performance of the model when said new proportion differs from said initial proportion by more than a predefined tolerance threshold.

Retrieved from "http://wikipatents.org/index.php?title=20240028944._ONLINE_DRIFT_DETECTION_FOR_FULLY_UNSUPERVISED_EVENT_DETECTION_IN_EDGE_ENVIRONMENTS_simplified_abstract_(Dell_Products_L.P.)&oldid=29838"