# Explanation Evaluation on Gemma Vision Encoder SAE 3 (test)

Source: [Explanation Evaluation on Gemma Vision Encoder SAE 3 (test)](<https://www.sota2.com/research/sota/explanation-evaluation-on-gemma-vision-encoder-sae-3-test>)

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

Data updated: 2026-03-25T14:26:50.822Z

Evaluation of explanation quality derived from the later layers of a Sparse Autoencoder (SAE) applied to the vision encoder of the Gemma model, assessed using metrics such as IoU, AUROC, and CLIP Score for both mask and heatmap representations.

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 | Explanation Evaluation |
| Dataset (including any stated split) | Gemma vision encoder later layer SAE 3 (test) |
| Metric | IoU (Masks) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 3 of 3 |
| 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

- [IoU (Masks)](<https://www.sota2.com/research/sota/explanation-evaluation-on-gemma-vision-encoder-sae-3-test.md?metric=IoU%20(Masks)>): Higher is better; unit: PERCENT.

- [IoU (Heatmaps)](<https://www.sota2.com/research/sota/explanation-evaluation-on-gemma-vision-encoder-sae-3-test.md?metric=IoU%20(Heatmaps)>): Higher is better; unit: PERCENT.

- [AUROC (Masks)](<https://www.sota2.com/research/sota/explanation-evaluation-on-gemma-vision-encoder-sae-3-test.md?metric=AUROC%20(Masks)>): Higher is better; unit: PERCENT.

- [AUROC (Heatmaps)](<https://www.sota2.com/research/sota/explanation-evaluation-on-gemma-vision-encoder-sae-3-test.md?metric=AUROC%20(Heatmaps)>): Higher is better; unit: PERCENT.

- [Synth. Act. Score (Masks)](<https://www.sota2.com/research/sota/explanation-evaluation-on-gemma-vision-encoder-sae-3-test.md?metric=Synth.%20Act.%20Score%20(Masks)>): Higher is better; unit: SCALAR.

- [Synth. Act. Score (Heatmaps)](<https://www.sota2.com/research/sota/explanation-evaluation-on-gemma-vision-encoder-sae-3-test.md?metric=Synth.%20Act.%20Score%20(Heatmaps)>): Higher is better; unit: SCALAR.

- [CLIP Score (Masks)](<https://www.sota2.com/research/sota/explanation-evaluation-on-gemma-vision-encoder-sae-3-test.md?metric=CLIP%20Score%20(Masks)>): Higher is better; unit: PERCENT.

- [CLIP Score (Heatmaps)](<https://www.sota2.com/research/sota/explanation-evaluation-on-gemma-vision-encoder-sae-3-test.md?metric=CLIP%20Score%20(Heatmaps)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | IoU (Masks) | Result date | Report paper |
| --- | --- | --- | --- |
| Steering | 20.4 | Not recorded | 69c36bfa0636e5bf4afc6406 |
| Steering-informed Top-k | 19.6 | Not recorded | 69c36bfa0636e5bf4afc6406 |
| Top-k | 19.4 | Not recorded | 69c36bfa0636e5bf4afc6406 |

### Steering

IoU (Masks): 20.4. Result ID: 69c3efd747f34ce924adc041.

[Reporting paper](<https://www.sota2.com/research/paper/69c36bfa0636e5bf4afc6406.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c36bfa0636e5bf4afc6406.md>)

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

### Steering-informed Top-k

IoU (Masks): 19.6. Result ID: 69c3efd747f34ce924adc043.

[Reporting paper](<https://www.sota2.com/research/paper/69c36bfa0636e5bf4afc6406.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c36bfa0636e5bf4afc6406.md>)

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

### Top-k

IoU (Masks): 19.4. Result ID: 69c3efd747f34ce924adc042.

[Reporting paper](<https://www.sota2.com/research/paper/69c36bfa0636e5bf4afc6406.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa5f861e209131ce9eaa87.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.