Protecting the Visual Fidelity of Machine Learning Datasets Using QR Codes

摘要

Machine learning is becoming increasingly popular in a variety of modern technology. However, research has demonstrated that, machine learning models are vulnerable to adversarial examples in their inputs. Potential attacks include poisoning datasets by perturbing input, samples to mislead a machine learning model into producing undesirable results. Such perturbations are often subtle and imperceptible from a human's perspective. This paper investigates two methods of verifying the visual fidelity of image based datasets by detecting perturbations made to the data using QR codes. In the first method, a verification string is stored for each image in a dataset. These verification strings can be used to determine whether an image in the dataset has been perturbed. In the second method, only a single verification string stored and is used to verify whether an entire dataset is intact.