# Image Classification on FMNIST (test) (Epsilon-Weighted Accuracy)

Source: [Image Classification on FMNIST (test) (Epsilon-Weighted Accuracy)](<https://www.sota2.com/research/sota/image-classification-on-fmnist-test-epsilon-weighted-accuracy>)

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

Data updated: 2026-06-04T11:23:53.955Z

Evaluation of Image Classification performance on the test split of the Fashion MNIST dataset, reporting accuracy metrics based on specific epsilon thresholds (1, 3, and 8).

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) | FMNIST (test) |
| Metric | Accuracy (epsilon=1) |
| Unit | PERCENT |
| Direction | Higher 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

- [Accuracy (epsilon=1)](<https://www.sota2.com/research/sota/image-classification-on-fmnist-test-epsilon-weighted-accuracy.md?metric=Accuracy%20(epsilon%3D1)>): Higher is better; unit: PERCENT.

- [Accuracy (epsilon=3)](<https://www.sota2.com/research/sota/image-classification-on-fmnist-test-epsilon-weighted-accuracy.md?metric=Accuracy%20(epsilon%3D3)>): Higher is better; unit: PERCENT.

- [Accuracy (epsilon=8)](<https://www.sota2.com/research/sota/image-classification-on-fmnist-test-epsilon-weighted-accuracy.md?metric=Accuracy%20(epsilon%3D8)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy (epsilon=1) | Result date | Report paper |
| --- | --- | --- | --- |
| TP-TopK | 85.28 | Not recorded | 6a2106d2c10484bac23f0470 |
| H-CNN + RS | 85.01 | Not recorded | 6a2106d2c10484bac23f0470 |
| DP-SGD | 84.06 | Not recorded | 6a2106d2c10484bac23f0470 |
| AUTO-S | 83.35 | Not recorded | 6a2106d2c10484bac23f0470 |
| SA-DP-SGD | 82.08 | Not recorded | 6a2106d2c10484bac23f0470 |

### TP-TopK

Accuracy (epsilon=1): 85.28. Result ID: 6a216036fbc373592c1e193c.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106d2c10484bac23f0470.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a2106d2c10484bac23f0470.md>)

Recorded evaluation configuration: {"Category":"Ours"}. This record does not assert independent reproduction.

### H-CNN + RS

Accuracy (epsilon=1): 85.01. Result ID: 6a216036fbc373592c1e193b.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106d2c10484bac23f0470.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6fe71e209131cea4803a.md>)

Recorded evaluation configuration: {"Category":"Random"}. This record does not assert independent reproduction.

### DP-SGD

Accuracy (epsilon=1): 84.06. Result ID: 6a216036fbc373592c1e1938.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106d2c10484bac23f0470.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65301e209131cea08bf5.md>)

Recorded evaluation configuration: {"Category":"Baseline"}. This record does not assert independent reproduction.

### AUTO-S

Accuracy (epsilon=1): 83.35. Result ID: 6a216036fbc373592c1e1939.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106d2c10484bac23f0470.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70c91e209131cea4d14e.md>)

[Implementation](<https://github.com/awslabs/fast-differential-privacy>)

Recorded evaluation configuration: {"Category":"Adaptive clipping"}. This record does not assert independent reproduction.

### SA-DP-SGD

Accuracy (epsilon=1): 82.08. Result ID: 6a216036fbc373592c1e193a.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106d2c10484bac23f0470.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e2a1e209131cea3e3f5.md>)

Recorded evaluation configuration: {"Category":"Adaptive clipping"}. This record does not assert independent reproduction.

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