# Certified Robustness on ImageNet Source: [Certified Robustness on ImageNet]() Collection index: [llms.txt]() Data updated: 2026-06-29T06:50:24.982Z Evaluation of model certified robustness on the standard ImageNet dataset, reporting certified accuracy across various L1 perturbation radii (0.5 to 3.0). A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Certified Robustness | | Dataset (including any stated split) | ImageNet | | Metric | Certified Accuracy (L1 R=0.5) | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 14 of 14 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Certified Accuracy (L1 R=0.5)](): Higher is better; unit: PERCENT. - [Certified Accuracy (L1 R=1.0)](): Higher is better; unit: PERCENT. - [Certified Accuracy (L1 R=1.5)](): Higher is better; unit: PERCENT. - [Certified Accuracy (L1 R=2.0)](): Higher is better; unit: PERCENT. - [Certified Accuracy (L1 R=2.5)](): Higher is better; unit: PERCENT. - [Certified Accuracy (L1 R=3.0)](): Higher is better; unit: PERCENT. - [Certified Accuracy](): Higher is better; unit: PERCENT. - [Robustness Delta (%)](): Lower is better; unit: PERCENT. - [Exit Iterations](): Lower is better; unit: SCALAR. ## Recorded results | Method | Certified Accuracy (L1 R=0.5) | Result date | Report paper | | --- | --- | --- | --- | | Uniform, LBS | 71.6 | Not recorded | 69f1907655a3c32af89fb9b2 | | Laplace, LBS | 70.5 | Not recorded | 69f1907655a3c32af89fb9b2 | | Uniform (w/), Yang et al. (2020) | 53.6 | Not recorded | 69f1907655a3c32af89fb9b2 | | Laplace (w/), Teng et al. (2020) | 47.1 | Not recorded | 69f1907655a3c32af89fb9b2 | | Laplace (w/), Lecuyer et al. (2019) | 42.8 | Not recorded | 69f1907655a3c32af89fb9b2 | | Laplace (w/o), Teng et al. (2020) | 5.8 | Not recorded | 69f1907655a3c32af89fb9b2 | | Uniform (w/o), Yang et al. (2020) | 5.7 | Not recorded | 69f1907655a3c32af89fb9b2 | | Laplace (w/o), Lecuyer et al. (2019) | 5.4 | Not recorded | 69f1907655a3c32af89fb9b2 | ### Uniform, LBS Certified Accuracy (L1 R=0.5): 71.6. Result ID: 69f1b0d047f34ce924d7bfd0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Noise Distribution":"Uniform"}. This record does not assert independent reproduction. ### Laplace, LBS Certified Accuracy (L1 R=0.5): 70.5. Result ID: 69f1b0d047f34ce924d7bfcf. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Noise Distribution":"Laplace"}. This record does not assert independent reproduction. ### Uniform (w/), Yang et al. (2020) Certified Accuracy (L1 R=0.5): 53.6. Result ID: 69f1b0d047f34ce924d7bfce. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Noise Distribution":"Uniform","Noise-Augmented Training":"true"}. This record does not assert independent reproduction. ### Laplace (w/), Teng et al. (2020) Certified Accuracy (L1 R=0.5): 47.1. Result ID: 69f1b0d047f34ce924d7bfcd. [Reporting paper]() Recorded evaluation configuration: {"Noise Distribution":"Laplace","Noise-Augmented Training":"true"}. This record does not assert independent reproduction. ### Laplace (w/), Lecuyer et al. (2019) Certified Accuracy (L1 R=0.5): 42.8. Result ID: 69f1b0d047f34ce924d7bfcc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Noise Distribution":"Laplace","Noise-Augmented Training":"true"}. This record does not assert independent reproduction. ### Laplace (w/o), Teng et al. (2020) Certified Accuracy (L1 R=0.5): 5.8. Result ID: 69f1b0d047f34ce924d7bfca. [Reporting paper]() Recorded evaluation configuration: {"Noise Distribution":"Laplace","Noise-Augmented Training":"false"}. This record does not assert independent reproduction. ### Uniform (w/o), Yang et al. (2020) Certified Accuracy (L1 R=0.5): 5.7. Result ID: 69f1b0d047f34ce924d7bfcb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Noise Distribution":"Uniform","Noise-Augmented Training":"false"}. This record does not assert independent reproduction. ### Laplace (w/o), Lecuyer et al. (2019) Certified Accuracy (L1 R=0.5): 5.4. Result ID: 69f1b0d047f34ce924d7bfc9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Noise Distribution":"Laplace","Noise-Augmented Training":"false"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.