Patent · US Active

Classification robust against multiple perturbation types

US11481681B2 · kind B2 · utility

0Cited by

1References

11Claims

0Family size

Assignees

Robert Bosch GmbH · DE
CARNEGIE MELLON UNIVERSITY · US

Inventors

Eric Wong · Lexington, US
Frank Schmidt · Leonberg, DE
Jeremy Zieg Kolter · Pittsburgh, US
Pratyush Maini · Bhangrola, IN

Key dates

Filing date	Apr 24, 2020
Grant date	Oct 25, 2022
Priority date	—
Expiry date	Mar 31, 2041

Classification

Technology area (CPC G)Physics
CPC primaryG06F18/217
WIPO fieldComputer technology
WIPO sectorElectrical engineering

Abstract

A system for training a classification model to be robust against perturbations of multiple perturbation types. A perturbation type defines a set of allowed perturbations. The classification model is trained by, in an outer iteration, selecting a set of training instances of a training dataset; selecting, among perturbations allowed by the multiple perturbation types, one or more perturbations for perturbing the selected training instances to maximize a loss function; and updating the set of parameters of the classification model to decrease the loss for the perturbed instances. A perturbation is determined by, in an inner iteration, determining updated perturbations allowed by respective perturbation types of the multiple perturbation types and selecting an updated perturbation that most increases the loss of the classification model.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.