# Machine Unlearning on ToxiGen (test)

Source: [Machine Unlearning on ToxiGen (test)](<https://www.sota2.com/research/sota/machine-unlearning-on-toxigen-test>)

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

Data updated: 2026-03-30T07:59:48.713Z

Evaluation of Machine Unlearning effectiveness on the test split of the ToxiGen dataset, reporting accuracy metrics for the forget set (Df) and adjacent/remote retain sets (Dr\_adj, Dr\_rem).

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 | Machine Unlearning |
| Dataset (including any stated split) | ToxiGen (test) |
| Metric | Accuracy ($D\_f$) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 13 of 13 |
| 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 ($D\_f$)](<https://www.sota2.com/research/sota/machine-unlearning-on-toxigen-test.md?metric=Accuracy%20(%24D_f%24)>): Higher is better; unit: PERCENT.

- [Accuracy ($D\_r^{adj}$)](<https://www.sota2.com/research/sota/machine-unlearning-on-toxigen-test.md?metric=Accuracy%20(%24D_r%5E%7Badj%7D%24)>): Higher is better; unit: PERCENT.

- [Accuracy ($D\_r^{rem}$)](<https://www.sota2.com/research/sota/machine-unlearning-on-toxigen-test.md?metric=Accuracy%20(%24D_r%5E%7Brem%7D%24)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy ($D\_f$) | Result date | Report paper |
| --- | --- | --- | --- |
| SSD | 86.9 | Not recorded | 69ca03a80636e5bf4afc949f |
| Original | 78.06 | Not recorded | 69ca03a80636e5bf4afc949f |
| SCRUB | 57.67 | Not recorded | 69ca03a80636e5bf4afc949f |
| Munba | 49.27 | Not recorded | 69ca03a80636e5bf4afc949f |
| FT | 47.73 | Not recorded | 69ca03a80636e5bf4afc949f |
| ℓ1-sparse | 46.31 | Not recorded | 69ca03a80636e5bf4afc949f |
| GA | 43.78 | Not recorded | 69ca03a80636e5bf4afc949f |
| DELETE | 39.53 | Not recorded | 69ca03a80636e5bf4afc949f |
| GDR | 19.83 | Not recorded | 69ca03a80636e5bf4afc949f |
| Our method | 14.29 | Not recorded | 69ca03a80636e5bf4afc949f |
| Two-Stage Machine Unlearning | 14.29 | Not recorded | 69ca03a80636e5bf4afc949f |
| SalUn | 12.42 | Not recorded | 69ca03a80636e5bf4afc949f |
| Retrain | 12.13 | Not recorded | 69ca03a80636e5bf4afc949f |

### SSD

Accuracy ($D\_f$): 86.9. Result ID: 69ca2d5847f34ce924b2a017.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa717d1e209131cea51995.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Original

Accuracy ($D\_f$): 78.06. Result ID: 69ca2d5847f34ce924b2a012.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### SCRUB

Accuracy ($D\_f$): 57.67. Result ID: 69ca2d5847f34ce924b2a018.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71011e209131cea4e7b9.md>)

[Implementation](<https://github.com/Meghdad92/SCRUB>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Munba

Accuracy ($D\_f$): 49.27. Result ID: 69ca2d5847f34ce924b2a016.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b241e209131cea86cda.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### FT

Accuracy ($D\_f$): 47.73. Result ID: 69ca2d5847f34ce924b2a013.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70381e209131cea49a4c.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### ℓ1-sparse

Accuracy ($D\_f$): 46.31. Result ID: 69ca2d5847f34ce924b2a015.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72001e209131cea540dc.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### GA

Accuracy ($D\_f$): 43.78. Result ID: 69ca2d5847f34ce924b2a014.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6bf01e209131cea31038.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### DELETE

Accuracy ($D\_f$): 39.53. Result ID: 69ca2d5847f34ce924b2a01a.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77ae1e209131cea74ad5.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### GDR

Accuracy ($D\_f$): 19.83. Result ID: 69ca2d5847f34ce924b2a01b.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Our method

Accuracy ($D\_f$): 14.29. Result ID: 69ca2d5847f34ce924b2a01c.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Two-Stage Machine Unlearning

Accuracy ($D\_f$): 14.29. Result ID: 69ca2d5847f34ce924b2a038.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### SalUn

Accuracy ($D\_f$): 12.42. Result ID: 69ca2d5847f34ce924b2a019.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72cd1e209131cea5888f.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Retrain

Accuracy ($D\_f$): 12.13. Result ID: 69ca2d5847f34ce924b2a037.

[Reporting paper](<https://www.sota2.com/research/paper/69ca03a80636e5bf4afc949f.md>)

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.
