# Adversarial Image Classification on EMNIST Letters Source: [Adversarial Image Classification on EMNIST Letters]() Collection index: [llms.txt]() Data updated: 2026-05-11T07:43:32.988Z Evaluation of Image Classification performance on the EMNIST Letters dataset, focusing on adversarial robustness metrics such as Accadv and AURAC, alongside clean accuracy. 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 | Image Classification | | Dataset (including any stated split) | EMNIST Letters | | Metric | Clean Accuracy | | 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 - [Clean Accuracy](): Higher is better; unit: PERCENT. - [Adversarial Accuracy (epsilon=0.30)](): Higher is better; unit: PERCENT. - [AURAC](): Higher is better; unit: PERCENT. - [Total Samples (K)](): Higher is better; unit: SIZE. ## Recorded results | Method | Clean Accuracy | Result date | Report paper | | --- | --- | --- | --- | | AdvTrain (off) | 81.4 | Not recorded | 6a01628edbc836b87d8230b7 | | AdvTrain | 81.4 | Not recorded | 6a01628edbc836b87d8230b7 | | Vanilla | 80.8 | Not recorded | 6a01628edbc836b87d8230b7 | | Vanilla | 80.7 | Not recorded | 6a01628edbc836b87d8230b7 | | Selective (off) | 80.7 | Not recorded | 6a01628edbc836b87d8230b7 | | Two-Stage Sel. | 80.7 | Not recorded | 6a01628edbc836b87d8230b7 | | Selective | 80.7 | Not recorded | 6a01628edbc836b87d8230b7 | | Selective | 80.7 | Not recorded | 6a01628edbc836b87d8230b7 | | Two-Stage Sel. | 80.7 | Not recorded | 6a01628edbc836b87d8230b7 | | AdvTrain (on) | 80.4 | Not recorded | 6a01628edbc836b87d8230b7 | | AdvTrain | 80.4 | Not recorded | 6a01628edbc836b87d8230b7 | | Selective (on) | 80.3 | Not recorded | 6a01628edbc836b87d8230b7 | | Sep-Aware | 80.3 | Not recorded | 6a01628edbc836b87d8230b7 | | Sep-Aware | 80.3 | Not recorded | 6a01628edbc836b87d8230b7 | ### AdvTrain (off) Clean Accuracy: 81.4. Result ID: 6a01888f47f34ce924e30fb3. [Reporting paper]() Recorded evaluation configuration: {"Attack":"Black-Box Transfer PGD-20","ρ":"0.9","Surrogate Model":"SimpleCNN"}. This record does not assert independent reproduction. ### AdvTrain Clean Accuracy: 81.4. Result ID: 6a01888f47f34ce924e30fcf. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Adversarial Training":"off"}. This record does not assert independent reproduction. ### Vanilla Clean Accuracy: 80.8. Result ID: 6a01888f47f34ce924e30fce. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Vanilla Clean Accuracy: 80.7. Result ID: 6a01888f47f34ce924e30fb2. [Reporting paper]() Recorded evaluation configuration: {"Attack":"Black-Box Transfer PGD-20","ρ":"0.9","Surrogate Model":"SimpleCNN"}. This record does not assert independent reproduction. ### Selective (off) Clean Accuracy: 80.7. Result ID: 6a01888f47f34ce924e30fb5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Attack":"Black-Box Transfer PGD-20","ρ":"0.9","Surrogate Model":"SimpleCNN"}. This record does not assert independent reproduction. ### Two-Stage Sel. Clean Accuracy: 80.7. Result ID: 6a01888f47f34ce924e30fb8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Attack":"Black-Box Transfer PGD-20","ρ":"0.9","Surrogate Model":"SimpleCNN"}. This record does not assert independent reproduction. ### Selective Clean Accuracy: 80.7. Result ID: 6a01888f47f34ce924e30fd1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Selective Mechanism":"off"}. This record does not assert independent reproduction. ### Selective Clean Accuracy: 80.7. Result ID: 6a01888f47f34ce924e30fd2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Selective Mechanism":"on"}. This record does not assert independent reproduction. ### Two-Stage Sel. Clean Accuracy: 80.7. Result ID: 6a01888f47f34ce924e30fd4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### AdvTrain (on) Clean Accuracy: 80.4. Result ID: 6a01888f47f34ce924e30fb4. [Reporting paper]() Recorded evaluation configuration: {"Attack":"Black-Box Transfer PGD-20","ρ":"0.9","Surrogate Model":"SimpleCNN"}. This record does not assert independent reproduction. ### AdvTrain Clean Accuracy: 80.4. Result ID: 6a01888f47f34ce924e30fd0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Adversarial Training":"on"}. This record does not assert independent reproduction. ### Selective (on) Clean Accuracy: 80.3. Result ID: 6a01888f47f34ce924e30fb6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Attack":"Black-Box Transfer PGD-20","ρ":"0.9","Surrogate Model":"SimpleCNN"}. This record does not assert independent reproduction. ### Sep-Aware Clean Accuracy: 80.3. Result ID: 6a01888f47f34ce924e30fb7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Attack":"Black-Box Transfer PGD-20","ρ":"0.9","Surrogate Model":"SimpleCNN"}. This record does not assert independent reproduction. ### Sep-Aware Clean Accuracy: 80.3. Result ID: 6a01888f47f34ce924e30fd3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.