# Image Classification on CIFAR-10 original (test) (TPR-95 Robustness Metrics)

Source: [Image Classification on CIFAR-10 original (test) (TPR-95 Robustness Metrics)](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-original-test-tpr-95-robustness-metrics>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-26T06:05:59.404Z

Evaluation of Image Classification performance on the original test split of the CIFAR-10 dataset, reporting specialized TPR-95 accuracy metrics under various conditions (Multi-target, GAMA PGD, GAMA FW).

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) | CIFAR-10 original (test) |
| Metric | TPR-95 Accuracy (Multi-target) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 9 of 9 |
| 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

- [TPR-95 Accuracy (Multi-target)](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-original-test-tpr-95-robustness-metrics.md?metric=TPR-95%20Accuracy%20(Multi-target)>): Higher is better; unit: PERCENT.

- [TPR-95 Accuracy (GAMA PGD)](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-original-test-tpr-95-robustness-metrics.md?metric=TPR-95%20Accuracy%20(GAMA%20PGD)>): Higher is better; unit: PERCENT.

- [TPR-95 Accuracy (GAMA FW)](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-original-test-tpr-95-robustness-metrics.md?metric=TPR-95%20Accuracy%20(GAMA%20FW)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | TPR-95 Accuracy (Multi-target) | Result date | Report paper |
| --- | --- | --- | --- |
| PGD-AT + RR | 56.18 | Not recorded | 68fa6c291e209131cea32724 |
| PGD-AT + EBD | 55.4 | Not recorded | 68fa6c291e209131cea32724 |
| PGD-AT + SNet | 55.02 | Not recorded | 68fa6c291e209131cea32724 |
| TRADES + RR | 54.83 | Not recorded | 68fa6c291e209131cea32724 |
| TRADES + Con. | 53.69 | Not recorded | 68fa6c291e209131cea32724 |
| CARL + Margin | 46.17 | Not recorded | 68fa6c291e209131cea32724 |
| CCAT + RR | 36.48 | Not recorded | 68fa6c291e209131cea32724 |
| CCAT + Con. | 34.21 | Not recorded | 68fa6c291e209131cea32724 |
| ATRO + Margin | 32.53 | Not recorded | 68fa6c291e209131cea32724 |

### PGD-AT + RR

TPR-95 Accuracy (Multi-target): 56.18. Result ID: 698ee417729b416a4aa59b5b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Implementation](<https://github.com/P2333/Rectified-Rejection>)

Recorded evaluation configuration: {"AT Framework":"PGD-AT","Rejector":"Rectified Rejection (RR)","Backbone":"ResNet-18","Threat Model":"l\_inf, eps=8/255"}. This record does not assert independent reproduction.

### PGD-AT + EBD

TPR-95 Accuracy (Multi-target): 55.4. Result ID: 698ee417729b416a4aa59b56.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69cc1e209131cea24242.md>)

Recorded evaluation configuration: {"AT Framework":"PGD-AT","Rejector":"EBD","Backbone":"ResNet-18","Threat Model":"l\_inf, eps=8/255"}. This record does not assert independent reproduction.

### PGD-AT + SNet

TPR-95 Accuracy (Multi-target): 55.02. Result ID: 698ee417729b416a4aa59b55.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65f21e209131cea0d4c0.md>)

Recorded evaluation configuration: {"AT Framework":"PGD-AT","Rejector":"SNet","Backbone":"ResNet-18","Threat Model":"l\_inf, eps=8/255"}. This record does not assert independent reproduction.

### TRADES + RR

TPR-95 Accuracy (Multi-target): 54.83. Result ID: 698ee417729b416a4aa59b5d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Implementation](<https://github.com/P2333/Rectified-Rejection>)

Recorded evaluation configuration: {"AT Framework":"TRADES","Rejector":"Rectified Rejection (RR)","Backbone":"ResNet-18","Threat Model":"l\_inf, eps=8/255"}. This record does not assert independent reproduction.

### TRADES + Con.

TPR-95 Accuracy (Multi-target): 53.69. Result ID: 698ee417729b416a4aa59b5a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65f71e209131cea0d897.md>)

Recorded evaluation configuration: {"AT Framework":"TRADES","Rejector":"Confidence","Backbone":"ResNet-18","Threat Model":"l\_inf, eps=8/255"}. This record does not assert independent reproduction.

### CARL + Margin

TPR-95 Accuracy (Multi-target): 46.17. Result ID: 698ee417729b416a4aa59b57.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66b71e209131cea120f5.md>)

Recorded evaluation configuration: {"AT Framework":"CARL","Rejector":"Margin","Backbone":"ResNet-18","Threat Model":"l\_inf, eps=8/255"}. This record does not assert independent reproduction.

### CCAT + RR

TPR-95 Accuracy (Multi-target): 36.48. Result ID: 698ee417729b416a4aa59b5c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Implementation](<https://github.com/P2333/Rectified-Rejection>)

Recorded evaluation configuration: {"AT Framework":"CCAT","Rejector":"Rectified Rejection (RR)","Backbone":"ResNet-18","Threat Model":"l\_inf, eps=8/255"}. This record does not assert independent reproduction.

### CCAT + Con.

TPR-95 Accuracy (Multi-target): 34.21. Result ID: 698ee417729b416a4aa59b59.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67d51e209131cea18cd6.md>)

Recorded evaluation configuration: {"AT Framework":"CCAT","Rejector":"Confidence","Backbone":"ResNet-18","Threat Model":"l\_inf, eps=8/255"}. This record does not assert independent reproduction.

### ATRO + Margin

TPR-95 Accuracy (Multi-target): 32.53. Result ID: 698ee417729b416a4aa59b58.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c291e209131cea32724.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa68a31e209131cea1d570.md>)

Recorded evaluation configuration: {"AT Framework":"ATRO","Rejector":"Margin","Backbone":"ResNet-18","Threat Model":"l\_inf, eps=8/255"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
