Reproduce “Alleviating Adversarial Attacks on Variational Autoencoders with MCMC” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Anna Kuzina, Max Welling, Jakub M. Tomczak
The authors add an inference-time Hamiltonian Monte Carlo (HMC) step to variational autoencoders, adjusting latent codes before reconstruction or classification without changing training. On MNIST, under an encoder attack with an L-infinity radius of 0.1, reconstruction MSSSIM rises from 0.70 to 0.88. This indicates improved reconstruction similarity in that setting, not robustness to every attack; HMC increases inference time.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.