# Defense against adaptive adversarial attacks on JigsawToxicity

Source: [Defense against adaptive adversarial attacks on JigsawToxicity](<https://www.sota2.com/research/sota/defense-against-adaptive-adversarial-attacks-on-jigsawtoxicity>)

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

Data updated: 2026-03-24T13:55:05.514Z

Evaluation of model defense robustness against adaptive adversarial attacks on the JigsawToxicity dataset, measured using 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 | Defense against adaptive adversarial attacks |
| Dataset (including any stated split) | JigsawToxicity |
| Metric | Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 18 of 18 |
| 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

- [Accuracy](<https://www.sota2.com/research/sota/defense-against-adaptive-adversarial-attacks-on-jigsawtoxicity.md?metric=Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Pre | 73.58 | Not recorded | 68fa75521e209131cea67277 |
| Pre | 73.58 | Not recorded | 68fa75521e209131cea67277 |
| LiSSA | 73.3 | Not recorded | 68fa75521e209131cea67277 |
| Self-LiSSA | 71.41 | Not recorded | 68fa75521e209131cea67277 |
| Self-TracIn | 71.37 | Not recorded | 68fa75521e209131cea67277 |
| DataInf | 70.82 | Not recorded | 68fa75521e209131cea67277 |
| IP | 70.82 | Not recorded | 68fa75521e209131cea67277 |
| IP Ensemble | 70.67 | Not recorded | 68fa75521e209131cea67277 |
| IP Ensemble | 70.51 | Not recorded | 68fa75521e209131cea67277 |
| IP | 70.43 | Not recorded | 68fa75521e209131cea67277 |
| DataInf | 70.39 | Not recorded | 68fa75521e209131cea67277 |
| LiSSA | 70.13 | Not recorded | 68fa75521e209131cea67277 |
| EKFAC | 70.05 | Not recorded | 68fa75521e209131cea67277 |
| EKFAC | 70.04 | Not recorded | 68fa75521e209131cea67277 |
| Self-LiSSA | 69.18 | Not recorded | 68fa75521e209131cea67277 |
| Self-TracIn | 68.31 | Not recorded | 68fa75521e209131cea67277 |
| Post | 66.18 | Not recorded | 68fa75521e209131cea67277 |
| Post | 66.18 | Not recorded | 68fa75521e209131cea67277 |

### Pre

Accuracy: 73.58. Result ID: 69c2970b47f34ce924ab079e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

Recorded evaluation configuration: {"Defense Protocol":"Relabeling"}. This record does not assert independent reproduction.

### Pre

Accuracy: 73.58. Result ID: 69c2970b47f34ce924ab07a7.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

Recorded evaluation configuration: {"Defense Protocol":"Reweighting"}. This record does not assert independent reproduction.

### LiSSA

Accuracy: 73.3. Result ID: 69c2970b47f34ce924ab07a0.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa68f81e209131cea1f643.md>)

Recorded evaluation configuration: {"Defense Protocol":"Relabeling"}. This record does not assert independent reproduction.

### Self-LiSSA

Accuracy: 71.41. Result ID: 69c2970b47f34ce924ab07ad.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70ec1e209131cea4e0d0.md>)

Recorded evaluation configuration: {"Defense Protocol":"Reweighting"}. This record does not assert independent reproduction.

### Self-TracIn

Accuracy: 71.37. Result ID: 69c2970b47f34ce924ab07ac.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71141e209131cea4f251.md>)

Recorded evaluation configuration: {"Defense Protocol":"Reweighting"}. This record does not assert independent reproduction.

### DataInf

Accuracy: 70.82. Result ID: 69c2970b47f34ce924ab07ab.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72821e209131cea56e03.md>)

Recorded evaluation configuration: {"Defense Protocol":"Reweighting"}. This record does not assert independent reproduction.

### IP

Accuracy: 70.82. Result ID: 69c2970b47f34ce924ab07ae.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

Recorded evaluation configuration: {"Defense Protocol":"Reweighting"}. This record does not assert independent reproduction.

### IP Ensemble

Accuracy: 70.67. Result ID: 69c2970b47f34ce924ab07af.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

Recorded evaluation configuration: {"Defense Protocol":"Reweighting"}. This record does not assert independent reproduction.

### IP Ensemble

Accuracy: 70.51. Result ID: 69c2970b47f34ce924ab07a6.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

Recorded evaluation configuration: {"Defense Protocol":"Relabeling"}. This record does not assert independent reproduction.

### IP

Accuracy: 70.43. Result ID: 69c2970b47f34ce924ab07a5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

Recorded evaluation configuration: {"Defense Protocol":"Relabeling"}. This record does not assert independent reproduction.

### DataInf

Accuracy: 70.39. Result ID: 69c2970b47f34ce924ab07a2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72821e209131cea56e03.md>)

Recorded evaluation configuration: {"Defense Protocol":"Relabeling"}. This record does not assert independent reproduction.

### LiSSA

Accuracy: 70.13. Result ID: 69c2970b47f34ce924ab07a9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa68f81e209131cea1f643.md>)

Recorded evaluation configuration: {"Defense Protocol":"Reweighting"}. This record does not assert independent reproduction.

### EKFAC

Accuracy: 70.05. Result ID: 69c2970b47f34ce924ab07aa.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa703b1e209131cea49c24.md>)

Recorded evaluation configuration: {"Defense Protocol":"Reweighting"}. This record does not assert independent reproduction.

### EKFAC

Accuracy: 70.04. Result ID: 69c2970b47f34ce924ab07a1.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa703b1e209131cea49c24.md>)

Recorded evaluation configuration: {"Defense Protocol":"Relabeling"}. This record does not assert independent reproduction.

### Self-LiSSA

Accuracy: 69.18. Result ID: 69c2970b47f34ce924ab07a4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70ec1e209131cea4e0d0.md>)

Recorded evaluation configuration: {"Defense Protocol":"Relabeling"}. This record does not assert independent reproduction.

### Self-TracIn

Accuracy: 68.31. Result ID: 69c2970b47f34ce924ab07a3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71141e209131cea4f251.md>)

Recorded evaluation configuration: {"Defense Protocol":"Relabeling"}. This record does not assert independent reproduction.

### Post

Accuracy: 66.18. Result ID: 69c2970b47f34ce924ab079f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

Recorded evaluation configuration: {"Defense Protocol":"Relabeling"}. This record does not assert independent reproduction.

### Post

Accuracy: 66.18. Result ID: 69c2970b47f34ce924ab07a8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa75521e209131cea67277.md>)

Recorded evaluation configuration: {"Defense Protocol":"Reweighting"}. This record does not assert independent reproduction.

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