Master of Arts (MA)
Mathematics and Statistics
We investigate two ideas in this thesis. First, we analyze the results of adaptingrecovery algorithms from linear inverse problems to defend neural networks against adversarial attacks. Second, we analyze the results of substituting sparsity priors with neural network priors in linear inverse problems. For the former, we are able to extend the framework introduced in  to defend neural networks against ℓ0, ℓ2,and ℓ∞ norm attacks, and for the latter, we find that our method yields an improvement over reconstruction results of .
Dhaliwal, Jasjeet, "Linear Inverse Problems and Neural Networks" (2021). Master's Theses. 5228.