# Image Classification on CIFAR-10 (test) (Accuracy and Retention)

Source: [Image Classification on CIFAR-10 (test) (Accuracy and Retention)](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-test-accuracy-and-retention>)

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

Data updated: 2026-02-26T05:52:55.385Z

Evaluation of Image Classification performance on the CIFAR-10 test set, reporting standard accuracy alongside model retention 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 | Image Classification |
| Dataset (including any stated split) | CIFAR-10 (test) |
| Metric | Model Retention |
| Unit | PERCENT |
| Direction | Higher 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

- [Model Retention](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-test-accuracy-and-retention.md?metric=Model%20Retention>): Higher is better; unit: PERCENT.

- [Accuracy](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-test-accuracy-and-retention.md?metric=Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Model Retention | Result date | Report paper |
| --- | --- | --- | --- |
| Teacher (ResNet-56) | 1 | Not recorded | 693a50306cfd488761a54c3b |
| Direct Training | 0.859 | Not recorded | 693a50306cfd488761a54c3b |
| HPM-KD | 0.854 | Not recorded | 693a50306cfd488761a54c3b |
| TAKD | 0.851 | Not recorded | 693a50306cfd488761a54c3b |
| Traditional KD | 0.847 | Not recorded | 693a50306cfd488761a54c3b |
| Attention Transfer | 0.838 | Not recorded | 693a50306cfd488761a54c3b |
| FitNets | 0.79 | Not recorded | 693a50306cfd488761a54c3b |

### Teacher (ResNet-56)

Model Retention: 1. Result ID: 6999735a5698665ec4e1b7b3.

[Reporting paper](<https://www.sota2.com/research/paper/693a50306cfd488761a54c3b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa62321e209131ce9f8328.md>)

Recorded evaluation configuration: {"Backbone":"ResNet-56","Time (s)":"595.2","CR":"1.0x"}. This record does not assert independent reproduction.

### Direct Training

Model Retention: 0.859. Result ID: 6999735a5698665ec4e1b7b4.

[Reporting paper](<https://www.sota2.com/research/paper/693a50306cfd488761a54c3b.md>)

Recorded evaluation configuration: {"Time (s)":"593.7 ± 10.9","CR":"3.1x"}. This record does not assert independent reproduction.

### HPM-KD

Model Retention: 0.854. Result ID: 6999735a5698665ec4e1b7b9.

[Reporting paper](<https://www.sota2.com/research/paper/693a50306cfd488761a54c3b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/693a50306cfd488761a54c3b.md>)

Recorded evaluation configuration: {"Time (s)":"603.1 ± 52.0","CR":"3.1x"}. This record does not assert independent reproduction.

### TAKD

Model Retention: 0.851. Result ID: 6999735a5698665ec4e1b7b8.

[Reporting paper](<https://www.sota2.com/research/paper/693a50306cfd488761a54c3b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66d21e209131cea129c3.md>)

Recorded evaluation configuration: {"Time (s)":"655.5 ± 66.0","CR":"3.1x"}. This record does not assert independent reproduction.

### Traditional KD

Model Retention: 0.847. Result ID: 6999735a5698665ec4e1b7b5.

[Reporting paper](<https://www.sota2.com/research/paper/693a50306cfd488761a54c3b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa61b81e209131ce9f5b87.md>)

Recorded evaluation configuration: {"Time (s)":"578.0 ± 6.2","CR":"3.1x"}. This record does not assert independent reproduction.

### Attention Transfer

Model Retention: 0.838. Result ID: 6999735a5698665ec4e1b7b7.

[Reporting paper](<https://www.sota2.com/research/paper/693a50306cfd488761a54c3b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa631f1e209131ce9fd01c.md>)

[Implementation](<https://github.com/szagoruyko/attention-transfer>)

Recorded evaluation configuration: {"Time (s)":"588.3 ± 6.1","CR":"3.1x"}. This record does not assert independent reproduction.

### FitNets

Model Retention: 0.79. Result ID: 6999735a5698665ec4e1b7b6.

[Reporting paper](<https://www.sota2.com/research/paper/693a50306cfd488761a54c3b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa61ca1e209131ce9f5ebe.md>)

Recorded evaluation configuration: {"Time (s)":"592.4 ± 5.9","CR":"3.1x"}. This record does not assert independent reproduction.

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