# Explanation Hardening Evaluation on CIFAR-10 10% utility loss, ϵ = 8, δ = 10⁻⁵ (test)

Source: [Explanation Hardening Evaluation on CIFAR-10 10% utility loss, ϵ = 8, δ = 10⁻⁵ (test)](<https://www.sota2.com/research/sota/explanation-hardening-evaluation-on-cifar-10-10-percent-utility-loss-8-10-test>)

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

Data updated: 2026-02-26T05:49:55.725Z

Evaluation of model explanation hardening effectiveness on a specific configuration of CIFAR-10, measuring changes in Mean Logit Sensitivity (MLS) before and after hardening against specified utility loss constraints (ϵ=8, δ=10⁻⁵).

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 Hardening Evaluation |
| Dataset (including any stated split) | CIFAR-10 10% utility loss, ϵ = 8, δ = 10⁻⁵ (test) |
| Metric | MLS (Pre-Hardening, TPR@0.1% FPR) |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 30 of 30 |
| 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

- [MLS (Pre-Hardening, TPR@0.1% FPR)](<https://www.sota2.com/research/sota/explanation-hardening-evaluation-on-cifar-10-10-percent-utility-loss-8-10-test.md?metric=MLS%20(Pre-Hardening%2C%20TPR%400.1%25%20FPR)>): Higher is better; unit: SCALAR.

- [MLS (Post-Hardening, TPR@0.1% FPR)](<https://www.sota2.com/research/sota/explanation-hardening-evaluation-on-cifar-10-10-percent-utility-loss-8-10-test.md?metric=MLS%20(Post-Hardening%2C%20TPR%400.1%25%20FPR)>): Higher is better; unit: SCALAR.

- [Sensitivity Change (∆S)](<https://www.sota2.com/research/sota/explanation-hardening-evaluation-on-cifar-10-10-percent-utility-loss-8-10-test.md?metric=Sensitivity%20Change%20(%E2%88%86S)>): Lower is better; unit: PERCENT.

## Recorded results

| Method | MLS (Pre-Hardening, TPR@0.1% FPR) | Result date | Report paper |
| --- | --- | --- | --- |
| Anchors | 45.44 | Not recorded | 695f39e353fdd5c361743ce1 |
| Anchors | 45.44 | Not recorded | 695f39e353fdd5c361743ce1 |
| Deconvolution | 16.08 | Not recorded | 695f39e353fdd5c361743ce1 |
| Deconvolution | 16.08 | Not recorded | 695f39e353fdd5c361743ce1 |
| ProtoDash | 14.9 | Not recorded | 695f39e353fdd5c361743ce1 |
| ProtoDash | 14.9 | Not recorded | 695f39e353fdd5c361743ce1 |
| Guided BackProp | 10.94 | Not recorded | 695f39e353fdd5c361743ce1 |
| Guided BackProp | 10.94 | Not recorded | 695f39e353fdd5c361743ce1 |
| SmoothGrad | 10.38 | Not recorded | 695f39e353fdd5c361743ce1 |
| SmoothGrad | 10.38 | Not recorded | 695f39e353fdd5c361743ce1 |
| LIME | 10.26 | Not recorded | 695f39e353fdd5c361743ce1 |
| LIME | 10.26 | Not recorded | 695f39e353fdd5c361743ce1 |
| VarGrad | 6.78 | Not recorded | 695f39e353fdd5c361743ce1 |
| VarGrad | 6.78 | Not recorded | 695f39e353fdd5c361743ce1 |
| GradCAM++ | 6.7 | Not recorded | 695f39e353fdd5c361743ce1 |
| GradCAM++ | 6.7 | Not recorded | 695f39e353fdd5c361743ce1 |
| Saliency Map | 5.24 | Not recorded | 695f39e353fdd5c361743ce1 |
| Saliency Map | 5.24 | Not recorded | 695f39e353fdd5c361743ce1 |
| SHAP | 5.04 | Not recorded | 695f39e353fdd5c361743ce1 |
| SHAP | 5.04 | Not recorded | 695f39e353fdd5c361743ce1 |
| Integrated Gradients | 4.88 | Not recorded | 695f39e353fdd5c361743ce1 |
| Integrated Gradients | 4.88 | Not recorded | 695f39e353fdd5c361743ce1 |
| InputXGrad | 2.4 | Not recorded | 695f39e353fdd5c361743ce1 |
| InputXGrad | 2.4 | Not recorded | 695f39e353fdd5c361743ce1 |
| Occlusion | 1.1 | Not recorded | 695f39e353fdd5c361743ce1 |
| Occlusion | 1.1 | Not recorded | 695f39e353fdd5c361743ce1 |
| DeepLIFT | 1 | Not recorded | 695f39e353fdd5c361743ce1 |
| DeepLIFT | 1 | Not recorded | 695f39e353fdd5c361743ce1 |
| GradCAM | 0.95 | Not recorded | 695f39e353fdd5c361743ce1 |
| GradCAM | 0.95 | Not recorded | 695f39e353fdd5c361743ce1 |

### Anchors

MLS (Pre-Hardening, TPR@0.1% FPR): 45.44. Result ID: 6996f5205698665ec441cd1e.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### Anchors

MLS (Pre-Hardening, TPR@0.1% FPR): 45.44. Result ID: 6996f5205698665ec441cd1f.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### Deconvolution

MLS (Pre-Hardening, TPR@0.1% FPR): 16.08. Result ID: 6996f5205698665ec441cd18.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### Deconvolution

MLS (Pre-Hardening, TPR@0.1% FPR): 16.08. Result ID: 6996f5205698665ec441cd19.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### ProtoDash

MLS (Pre-Hardening, TPR@0.1% FPR): 14.9. Result ID: 6996f5205698665ec441cd1c.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### ProtoDash

MLS (Pre-Hardening, TPR@0.1% FPR): 14.9. Result ID: 6996f5205698665ec441cd1d.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### Guided BackProp

MLS (Pre-Hardening, TPR@0.1% FPR): 10.94. Result ID: 6996f5205698665ec441cd12.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### Guided BackProp

MLS (Pre-Hardening, TPR@0.1% FPR): 10.94. Result ID: 6996f5205698665ec441cd13.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### SmoothGrad

MLS (Pre-Hardening, TPR@0.1% FPR): 10.38. Result ID: 6996f5205698665ec441cd02.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### SmoothGrad

MLS (Pre-Hardening, TPR@0.1% FPR): 10.38. Result ID: 6996f5205698665ec441cd03.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### LIME

MLS (Pre-Hardening, TPR@0.1% FPR): 10.26. Result ID: 6996f5205698665ec441cd0e.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### LIME

MLS (Pre-Hardening, TPR@0.1% FPR): 10.26. Result ID: 6996f5205698665ec441cd0f.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### VarGrad

MLS (Pre-Hardening, TPR@0.1% FPR): 6.78. Result ID: 6996f5205698665ec441cd04.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### VarGrad

MLS (Pre-Hardening, TPR@0.1% FPR): 6.78. Result ID: 6996f5205698665ec441cd05.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### GradCAM++

MLS (Pre-Hardening, TPR@0.1% FPR): 6.7. Result ID: 6996f5205698665ec441cd0a.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### GradCAM++

MLS (Pre-Hardening, TPR@0.1% FPR): 6.7. Result ID: 6996f5205698665ec441cd0b.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### Saliency Map

MLS (Pre-Hardening, TPR@0.1% FPR): 5.24. Result ID: 6996f5205698665ec441cd10.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### Saliency Map

MLS (Pre-Hardening, TPR@0.1% FPR): 5.24. Result ID: 6996f5205698665ec441cd11.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### SHAP

MLS (Pre-Hardening, TPR@0.1% FPR): 5.04. Result ID: 6996f5205698665ec441cd0c.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### SHAP

MLS (Pre-Hardening, TPR@0.1% FPR): 5.04. Result ID: 6996f5205698665ec441cd0d.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### Integrated Gradients

MLS (Pre-Hardening, TPR@0.1% FPR): 4.88. Result ID: 6996f5205698665ec441cd06.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### Integrated Gradients

MLS (Pre-Hardening, TPR@0.1% FPR): 4.88. Result ID: 6996f5205698665ec441cd07.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### InputXGrad

MLS (Pre-Hardening, TPR@0.1% FPR): 2.4. Result ID: 6996f5205698665ec441cd16.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### InputXGrad

MLS (Pre-Hardening, TPR@0.1% FPR): 2.4. Result ID: 6996f5205698665ec441cd17.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### Occlusion

MLS (Pre-Hardening, TPR@0.1% FPR): 1.1. Result ID: 6996f5205698665ec441cd1a.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### Occlusion

MLS (Pre-Hardening, TPR@0.1% FPR): 1.1. Result ID: 6996f5205698665ec441cd1b.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### DeepLIFT

MLS (Pre-Hardening, TPR@0.1% FPR): 1. Result ID: 6996f5205698665ec441cd14.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### DeepLIFT

MLS (Pre-Hardening, TPR@0.1% FPR): 1. Result ID: 6996f5205698665ec441cd15.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

### GradCAM

MLS (Pre-Hardening, TPR@0.1% FPR): 0.95. Result ID: 6996f5205698665ec441cd08.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DeepLeak"}. This record does not assert independent reproduction.

### GradCAM

MLS (Pre-Hardening, TPR@0.1% FPR): 0.95. Result ID: 6996f5205698665ec441cd09.

[Reporting paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39e353fdd5c361743ce1.md>)

Recorded evaluation configuration: {"Hardening Method":"DP-SGD"}. This record does not assert independent reproduction.

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