Basic Information
The Course Will Focus On Theoretical Studies and Practical Aspects Of Adversarial Attacks On Deep-learning Models and On Cyber-security Systems. in The Theoretical Studies, We Will Focus On The Various Types of Adversarial Attacks, and Cover Papers Aiming to Explain Their Success in Harming The Proper Conduct of Models and Systems. We Will Discuss Approaches to Attacks Such As White-box and Black-box Attacks, Universal Attacks, Sparse Attacks, Patch-based Attacks, And Attacks On Classification Models As Opposed to Regression Models. We Will Discuss Various Adversarial Defenses As Well Which Are Meant To Mitigate The Effect of Adversarial Attacks. We Will Discuss Approaches to Defenses Such As Adversarial Training, Certified Defenses, Empirical Randomized Smoothing, and Adversarial Repair. In Addition, We Will Discuss Other Utilizations of Adversarial Attacks, Such As Un-adversarial Perturbations and Methods of Revealing The Structure and Parameters of Attacked Models. in The Tutorials We Will Discuss How to Make Use of Known Attacks and Defenses, How to Develop New Attacks and Defenses, and How to Evaluate The Success of Attacks Against Defenses. Learning Outcomes# at The End of The Course, The Students Will Be Able# 1. Understand The Theories Explaining The Success of Adversarial Attacks in Harming The Proper Conduct of Models and Systems._ 2. Understand The Methodological Approaches and Theoretical Principles That Existing Attacks and Defenses Are Based On. 3. Develop New Adversarial Attacks and Evaluate Their Success. 4. Develop New Adversarial Defenses and Evaluate Their Success Against Attacks. (5) Make Use of The Phenomenon of Adversarial Attacks For Various Purposes, and Specifically For Improving The Performance of Models On Clean Input. 6
Faculty: Computer Science
|Graduate Studies
Pre-required courses
46211 - Deep Learning or 97200 - תיאוריה ומעשה or 236781 - Deep Learning On Computation