# Evasion Attack Detection on MNIST

Source: [Evasion Attack Detection on MNIST](<https://www.sota2.com/research/sota/evasion-attack-detection-on-mnist>)

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

Data updated: 2026-05-19T07:22:55.252Z

Evaluation of model robustness against evasion attacks on the MNIST dataset, reporting detection accuracy and QUBO advantage metrics.

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 | Evasion Attack Detection |
| Dataset (including any stated split) | MNIST |
| Metric | Detection Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 12 of 12 |
| 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

- [Detection Accuracy](<https://www.sota2.com/research/sota/evasion-attack-detection-on-mnist.md?metric=Detection%20Accuracy>): Higher is better; unit: PERCENT.

- [QUBO Advantage](<https://www.sota2.com/research/sota/evasion-attack-detection-on-mnist.md?metric=QUBO%20Advantage>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Detection Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| QUBO-based client selection | 95.11 | Not recorded | 6a0bee88dbc836b87d82c388 |
| QUBO-based client selection | 86.67 | Not recorded | 6a0bee88dbc836b87d82c388 |
| Classical MultiKrum | 81.33 | Not recorded | 6a0bee88dbc836b87d82c388 |
| QUBO-based client selection | 80.89 | Not recorded | 6a0bee88dbc836b87d82c388 |
| QUBO-based client selection | 78.76 | Not recorded | 6a0bee88dbc836b87d82c388 |
| Classical MultiKrum | 72.44 | Not recorded | 6a0bee88dbc836b87d82c388 |
| QUBO-based client selection | 71.11 | Not recorded | 6a0bee88dbc836b87d82c388 |
| Classical MultiKrum | 69.95 | Not recorded | 6a0bee88dbc836b87d82c388 |
| Classical MultiKrum | 69.33 | Not recorded | 6a0bee88dbc836b87d82c388 |
| Classical MultiKrum | 66.67 | Not recorded | 6a0bee88dbc836b87d82c388 |
| Classical MultiKrum | 60 | Not recorded | 6a0bee88dbc836b87d82c388 |
| QUBO-based client selection | 60 | Not recorded | 6a0bee88dbc836b87d82c388 |

### QUBO-based client selection

Detection Accuracy: 95.11. Result ID: 6a0c0fb1fbc373592c01b78d.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"ALIE"}. This record does not assert independent reproduction.

### QUBO-based client selection

Detection Accuracy: 86.67. Result ID: 6a0c0fb1fbc373592c01b78f.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"Lie"}. This record does not assert independent reproduction.

### Classical MultiKrum

Detection Accuracy: 81.33. Result ID: 6a0c0fb1fbc373592c01b78c.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"ALIE"}. This record does not assert independent reproduction.

### QUBO-based client selection

Detection Accuracy: 80.89. Result ID: 6a0c0fb1fbc373592c01b791.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"Sparse Lie"}. This record does not assert independent reproduction.

### QUBO-based client selection

Detection Accuracy: 78.76. Result ID: 6a0c0fb1fbc373592c01b797.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"Average (5 attacks)"}. This record does not assert independent reproduction.

### Classical MultiKrum

Detection Accuracy: 72.44. Result ID: 6a0c0fb1fbc373592c01b78e.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"Lie"}. This record does not assert independent reproduction.

### QUBO-based client selection

Detection Accuracy: 71.11. Result ID: 6a0c0fb1fbc373592c01b793.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"Sign Flip"}. This record does not assert independent reproduction.

### Classical MultiKrum

Detection Accuracy: 69.95. Result ID: 6a0c0fb1fbc373592c01b796.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"Average (5 attacks)"}. This record does not assert independent reproduction.

### Classical MultiKrum

Detection Accuracy: 69.33. Result ID: 6a0c0fb1fbc373592c01b794.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"Shuffle"}. This record does not assert independent reproduction.

### Classical MultiKrum

Detection Accuracy: 66.67. Result ID: 6a0c0fb1fbc373592c01b790.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"Sparse Lie"}. This record does not assert independent reproduction.

### Classical MultiKrum

Detection Accuracy: 60. Result ID: 6a0c0fb1fbc373592c01b792.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"Sign Flip"}. This record does not assert independent reproduction.

### QUBO-based client selection

Detection Accuracy: 60. Result ID: 6a0c0fb1fbc373592c01b795.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee88dbc836b87d82c388.md>)

Recorded evaluation configuration: {"Attack":"Shuffle"}. This record does not assert independent reproduction.

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