Denoisers and Input-Dependent Noise: Neural Network Certification, Part 9
Adapt randomized smoothing by denoising April’s noisy photographs and choosing a certifiably safe noise level for each input.
Recent posts
How Can Random Noise Produce a Certificate? Neural Network Certification, Part 8
Follow noisy copies of April’s photo from majority vote to a probabilistic robustness certificate, then examine training and evaluation.
Three Research Frontiers: Neural Network Certification, Part 7
Three surprising results separate what certifiable networks can represent, what training can find, and what a verifier can prove.
How Can Training Make Robustness Easier to Prove? Neural Network Certification, Part 6
Follow April’s worst-case loss from attacks to sound bounds, then separate sound certified training from unsound training surrogates.
What Should We Do If the Bound Is Inconclusive? Neural Network Certification, Part 5
Split April’s input region, bound each piece, and assemble the local results into a complete certificate.