# Image Classification on CIFAR10 epsilon = 4/255 (test)

Source: [Image Classification on CIFAR10 epsilon = 4/255 (test)](<https://www.sota2.com/research/sota/image-classification-on-cifar10-epsilon-4-255-test>)

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

Data updated: 2026-02-26T05:39:22.381Z

Evaluation of Image Classification robustness on the CIFAR-10 test set under an L-infinity perturbation constraint of epsilon=4/255, reporting clean error, certified error, and autoattack error rates.

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) | CIFAR10 epsilon = 4/255 (test) |
| Metric | Clean Error Rate |
| Unit | PERCENT |
| Direction | Lower is better |
| Results loaded | 5 of 5 |
| 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 Error Rate](<https://www.sota2.com/research/sota/image-classification-on-cifar10-epsilon-4-255-test.md?metric=Clean%20Error%20Rate>): Lower is better; unit: PERCENT.

- [Autoattack Error Rate](<https://www.sota2.com/research/sota/image-classification-on-cifar10-epsilon-4-255-test.md?metric=Autoattack%20Error%20Rate>): Lower is better; unit: PERCENT.

- [Certified Error Rate](<https://www.sota2.com/research/sota/image-classification-on-cifar10-epsilon-4-255-test.md?metric=Certified%20Error%20Rate>): Lower is better; unit: PERCENT.

## Recorded results

| Method | Clean Error Rate | Result date | Report paper |
| --- | --- | --- | --- |
| A5/RC | 35.26 | Not recorded | 68fa6f6a1e209131cea456c6 |
| A5/RC | 36.96 | Not recorded | 68fa6f6a1e209131cea456c6 |
| A5/RC | 41.24 | Not recorded | 68fa6f6a1e209131cea456c6 |
| A5/RC | 46.76 | Not recorded | 68fa6f6a1e209131cea456c6 |
| A5/RC | 50.24 | Not recorded | 68fa6f6a1e209131cea456c6 |

### A5/RC

Clean Error Rate: 35.26. Result ID: 6990bbf33a6c31484902da63.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f6a1e209131cea456c6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f6a1e209131cea456c6.md>)

[Implementation](<https://github.com/NVlabs/A5>)

Recorded evaluation configuration: {"epsilon\_CA":"8/255","epsilon\_D":"32/255","epsilon\_A\_tr":"4/255"}. This record does not assert independent reproduction.

### A5/RC

Clean Error Rate: 36.96. Result ID: 6990bbf33a6c31484902da62.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f6a1e209131cea456c6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f6a1e209131cea456c6.md>)

[Implementation](<https://github.com/NVlabs/A5>)

Recorded evaluation configuration: {"epsilon\_CA":"8/255","epsilon\_D":"16/255","epsilon\_A\_tr":"4/255"}. This record does not assert independent reproduction.

### A5/RC

Clean Error Rate: 41.24. Result ID: 6990bbf33a6c31484902da61.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f6a1e209131cea456c6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f6a1e209131cea456c6.md>)

[Implementation](<https://github.com/NVlabs/A5>)

Recorded evaluation configuration: {"epsilon\_CA":"8/255","epsilon\_D":"8/255","epsilon\_A\_tr":"4/255"}. This record does not assert independent reproduction.

### A5/RC

Clean Error Rate: 46.76. Result ID: 6990bbf33a6c31484902da60.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f6a1e209131cea456c6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f6a1e209131cea456c6.md>)

[Implementation](<https://github.com/NVlabs/A5>)

Recorded evaluation configuration: {"epsilon\_CA":"8/255","epsilon\_D":"4/255","epsilon\_A\_tr":"4/255"}. This record does not assert independent reproduction.

### A5/RC

Clean Error Rate: 50.24. Result ID: 6990bbf33a6c31484902da5f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f6a1e209131cea456c6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f6a1e209131cea456c6.md>)

[Implementation](<https://github.com/NVlabs/A5>)

Recorded evaluation configuration: {"epsilon\_CA":"8/255","epsilon\_D":"0/255","epsilon\_A\_tr":"4/255"}. This record does not assert independent reproduction.

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