Machine Learning Robustness and Security
September 2, 2022 • 1 min • 141 words •
Information
- Evaluating defenses against adversarial machine learning attacks that improve the robustness vs. accuracy trade-off.
- Creating and evaluating tools that use adversarial attacks for anti-surveillance and anti-face-recognition.
Relevant Projects
Hierarchical Robustness
-
DNNs are susceptible to adversarial examples. Fortunately, objects tend to have invariant salient features. Applying invariants to the classification task hierarchically makes robustness and accuracy feasible together.
Face-Off
-
Advances in deep learning have made face recognition technologies and surveillance pervasive. Face-Off is a privacy-preserving framework that introduces strategic perturbations to the user's face to prevent it from being correctly recognized.
Face Obfuscation Fairness
-
A recent, popular approach to address face recognition privacy concerns is to employ evasion attacks against the DNNs powering face recognition systems. This dependence of face obfuscation on DNNs, which are known to be unfair in the context of face recognition, surfaces the question of demographic fairness.
