18075784. LEARNING HYPER-PARAMETER SCALING MODELS FOR UNSUPERVISED ANOMALY DETECTION simplified abstract (Oracle International Corporation)

LEARNING HYPER-PARAMETER SCALING MODELS FOR UNSUPERVISED ANOMALY DETECTION

Organization Name

Oracle International Corporation

Inventor(s)

Fatjon Zogaj of Zurich (CH)

Yasha Pushak of Vancouver (CA)

Hesam Fathi Moghadam of Sunnyvale CA (US)

Sungpack Hong of Palo Alto CA (US)

Hassan Chafi of San Mateo CA (US)

LEARNING HYPER-PARAMETER SCALING MODELS FOR UNSUPERVISED ANOMALY DETECTION - A simplified explanation of the abstract

This abstract first appeared for US patent application 18075784 titled 'LEARNING HYPER-PARAMETER SCALING MODELS FOR UNSUPERVISED ANOMALY DETECTION

Simplified Explanation

The abstract describes a patent application for a system that uses empirical validation scores to select the best training scenarios for an anomaly detector, optimizing hyperparameters using linear optimizers.

• The system sorts empirical validation scores of validated training scenarios of an anomaly detector.
• Each training scenario has a dataset with values for metafeatures and hyperparameters.
• A subset of the best training scenarios is selected based on predefined ranking percentages.
• Linear optimizers train to infer values for hyperparameters.
• Unvalidated training scenarios are generated with inferred hyperparameter values.
• The best linear optimizer is selected based on the highest combined inferred validation score.
• The best linear optimizer infers hyperparameter values for new datasets.

Potential Applications

This technology could be applied in various fields such as cybersecurity, fraud detection, and anomaly detection in industrial systems.

Problems Solved

This technology helps in automating the process of selecting and optimizing training scenarios for anomaly detectors, improving their performance and accuracy.

Benefits

The system saves time and resources by efficiently selecting the best training scenarios and optimizing hyperparameters, leading to better anomaly detection results.

Potential Commercial Applications

Potential commercial applications include security software companies, financial institutions, and manufacturing companies looking to enhance their anomaly detection systems.

Possible Prior Art

One possible prior art could be the use of machine learning algorithms to optimize hyperparameters in anomaly detection systems.

How does this system handle datasets with high dimensionality?

The system's performance with high-dimensional datasets is not explicitly mentioned in the abstract. It would be interesting to know if the system has any limitations or special considerations when dealing with such datasets.

Can this system adapt to changing data distributions over time?

It is not clear from the abstract how the system handles changes in data distributions over time. It would be important to understand if the system can adapt and continue to perform well in dynamic environments.

Original Abstract Submitted

A computer sorts empirical validation scores of validated training scenarios of an anomaly detector. Each training scenario has a dataset to train an instance of the anomaly detector that is configured with values for hyperparameters. Each dataset has values for metafeatures. For each predefined ranking percentage, a subset of best training scenarios is selected that consists of the ranking percentage of validated training scenarios having the highest empirical validation scores. Linear optimizers train to infer a value for a hyperparameter. Into many distinct unvalidated training scenarios, a scenario is generated that has metafeatures values and hyperparameters values that contains the value inferred for that hyperparameter by a linear optimizer. For each unvalidated training scenario, a validation score is inferred. A best linear optimizer is selected having a highest combined inferred validation score. For a new dataset, the best linear optimizer infers a value of that hyperparameter.