# Machine Unlearning on TOFU (Attack Robustness and Forgetting)

Source: [Machine Unlearning on TOFU (Attack Robustness and Forgetting)](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-attack-robustness-and-forgetting>)

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

Data updated: 2026-03-05T06:57:18.256Z

Evaluation of Machine Unlearning effectiveness on the base TOFU dataset, focusing on forgetting scores (FS), retention (RS), utility (US), and performance against harmful and benign data extraction attacks.

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) | TOFU |
| Metric | Forgetting Score (FS) |
| Unit | PERCENT |
| Direction | Lower is better |
| Results loaded | 7 of 7 |
| 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

- [Forgetting Score (FS)](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-attack-robustness-and-forgetting.md?metric=Forgetting%20Score%20(FS)>): Lower is better; unit: PERCENT.

- [Retention Score (RS)](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-attack-robustness-and-forgetting.md?metric=Retention%20Score%20(RS)>): Higher is better; unit: PERCENT.

- [Utility Score (US)](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-attack-robustness-and-forgetting.md?metric=Utility%20Score%20(US)>): Higher is better; unit: PERCENT.

- [Harmful Attack Success Rate (p=0.1)](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-attack-robustness-and-forgetting.md?metric=Harmful%20Attack%20Success%20Rate%20(p%3D0.1)>): Lower is better; unit: PERCENT.

- [Harmful Attack Success Rate (p=0.3)](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-attack-robustness-and-forgetting.md?metric=Harmful%20Attack%20Success%20Rate%20(p%3D0.3)>): Lower is better; unit: PERCENT.

- [Benign Attack Success Rate](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-attack-robustness-and-forgetting.md?metric=Benign%20Attack%20Success%20Rate>): Lower is better; unit: PERCENT.

## Recorded results

| Method | Forgetting Score (FS) | Result date | Report paper |
| --- | --- | --- | --- |
| RMU | 14.04 | Not recorded | 68fa7baa1e209131cea89c3d |
| SSIUU | 17.75 | Not recorded | 68fa7baa1e209131cea89c3d |
| KLUE | 18.71 | Not recorded | 68fa7baa1e209131cea89c3d |
| GD | 19.32 | Not recorded | 68fa7baa1e209131cea89c3d |
| GA | 19.49 | Not recorded | 68fa7baa1e209131cea89c3d |
| DPO | 19.57 | Not recorded | 68fa7baa1e209131cea89c3d |
| Default | 90.42 | Not recorded | 68fa7baa1e209131cea89c3d |

### RMU

Forgetting Score (FS): 14.04. Result ID: 69a927df6dbb7075b951f492.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7baa1e209131cea89c3d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa732f1e209131cea5af82.md>)

Recorded evaluation configuration: {"Model":"Llama-3.2"}. This record does not assert independent reproduction.

### SSIUU

Forgetting Score (FS): 17.75. Result ID: 69a927df6dbb7075b951f494.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7baa1e209131cea89c3d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7baa1e209131cea89c3d.md>)

Recorded evaluation configuration: {"Model":"Llama-3.2"}. This record does not assert independent reproduction.

### KLUE

Forgetting Score (FS): 18.71. Result ID: 69a927df6dbb7075b951f493.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7baa1e209131cea89c3d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa772b1e209131cea71878.md>)

Recorded evaluation configuration: {"Model":"Llama-3.2"}. This record does not assert independent reproduction.

### GD

Forgetting Score (FS): 19.32. Result ID: 69a927df6dbb7075b951f490.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7baa1e209131cea89c3d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71c71e209131cea52ea2.md>)

Recorded evaluation configuration: {"Model":"Llama-3.2"}. This record does not assert independent reproduction.

### GA

Forgetting Score (FS): 19.49. Result ID: 69a927df6dbb7075b951f48f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7baa1e209131cea89c3d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e421e209131cea3ec45.md>)

[Implementation](<https://github.com/joeljang/knowledge-unlearning>)

Recorded evaluation configuration: {"Model":"Llama-3.2"}. This record does not assert independent reproduction.

### DPO

Forgetting Score (FS): 19.57. Result ID: 69a927df6dbb7075b951f491.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7baa1e209131cea89c3d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74511e209131cea6105d.md>)

Recorded evaluation configuration: {"Model":"Llama-3.2"}. This record does not assert independent reproduction.

### Default

Forgetting Score (FS): 90.42. Result ID: 69a927df6dbb7075b951f48e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7baa1e209131cea89c3d.md>)

Recorded evaluation configuration: {"Model":"Llama-3.2"}. This record does not assert independent reproduction.

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