Apparatus and method for determining vulnerability of deep learning model
AssigneeIndustry Academy Cooperation Foundation of Sejong University
InventorsJoo Beom YUN, Ji Hyeon YU, Hyun Jun MUN
Life of this patent
Estimating the term…
Abstract
An apparatus for determining a vulnerability of a deep learning model according to an embodiment includes a converter configured to generate an input image for the deep learning model by transforming an original image selected from an image dataset, a measurer configured to measure neuron coverage of the deep learning model by inputting the input image into the deep learning model, and an inspector configured to detect, based on a prediction result of the deep learning model for a class of the input image and a class of the original image, an error in the prediction result.
Claims12
An apparatus for determining vulnerability of a deep learning model, the apparatus comprising:
- a converter configured to generate an input image for the deep learning model by transforming an original image selected from an image dataset;
- a measurer configured to measure neuron coverage of the deep learning model by inputting the input image into the deep learning model;
- and
- an inspector configured to detect, based on a prediction result of the deep learning model for a class of the input image and a class of the original image, an error in the prediction result.
A method for determining vulnerability of a deep learning model, the method comprising:
- generating an input image for the deep learning model by transforming an original image selected from an image dataset;
- measuring neuron coverage of the deep learning model by inputting the input image into the deep learning model;
- and
- detecting based on a prediction result of the deep learning model for a class of the input image and a class of the original image, an error in the prediction result.
Dependencies are inferred from the claim text.