Generating Adversarial Examples

Adversarial attacks and defense in Machine Learning applications.

Input(s):

• Image datasets

Main feature(s):

• Enhancing security and robustness of Neural Networks especially in the face of an adversary who wishes to fool the model

Output(s):

• A slightly perturbed image, still easily recognizable by human observers with the goal of producing a wrong output from the correct target class

A labeled dataset and a primary machine learning model for classification are needed. The security and Robustness of this Neural Network can be improved by generating an adversarial dataset.

AI-based system’s developer.

Licensing and permission required.