ResearchBenchmarksImage Classification on MNIST ε=0.1 (test)Image Classification on MNIST ε=0.1 (test)Follow99.25Standard AccuracyAD-CERT98.854898.957499.0699.1626Jun 30, 2026Evaluation resultsMethodMethodLinksStandard AccuracyCertified AccuracyAD-CERTε=0.1ε=0.12026.0699.2598.53MTL-IBPε=0.1ε=0.12026.0699.1898.37TAPS/STAPSε=0.1, Selection proto...ε=0.1, Selection protocol=Stronger of TAPS and STAPS based on certified accuracy2026.0699.1698.52SABRε=0.1ε=0.12026.0699.0898.4SORTNETε=0.1, Architecture=No...ε=0.1, Architecture=Non-ReLU-based2026.0699.0198.14IBPε=0.1ε=0.12026.0698.8798.26