# Measuring Model Robustness via Fisher Information: Spectral Bounds, Theoretical Guarantees, and Practical Algorithms

Source: [Measuring Model Robustness via Fisher Information: Spectral Bounds, Theoretical Guarantees, and Practical Algorithms](<https://www.sota2.com/research/paper/6a210703c10484bac23f07e7>)

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

Data updated: 2026-06-11T06:00:04.628Z

Authors: Chong Zhang, Xiang Li, Jia Wang, Qiufeng Wang, Xiaobo Jin

Published: 2026-06-03T11:50:38.000Z

## Abstract

The robustness of deep neural networks is crucial for safety-critical deployments, yet existing evaluation methods are often attack-dependent and lack interpretability. We propose a principled, attack-agnostic robustness metric based on the spectral norm of the Fisher Information Matrix (FIM), which quantifies the worst-case sensitivity of the model's output distribution to input perturbations. Theoretically, we establish that the FIM equals the variance of the input Jacobian and derive closed-form spectral bounds for common architectures, including VGG, ResNet, DenseNet, and Transformer, providing the first theoretical robustness ranking. To enable scalable evaluation, we develop efficient algorithms, including power iteration and Hutchinson-based estimation, that support both white-box and black-box settings. Extensive experiments across multiple datasets, including CIFAR, ImageNet, and medical images, and across multiple architectures show a strong correlation between our metric and adversarial vulnerability. Our framework serves as an interpretable diagnostic tool that complements attack-based evaluations, offering insights into architectural sensitivity and guiding the design of more robust models. Code is available at: https://github.com/franz-chang/SRP/.

## Recorded benchmark results

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.

- [Robustness Evaluation on CIFAR10](<https://www.sota2.com/research/sota/robustness-evaluation-on-cifar10.md>): Robustness Evaluation; CIFAR10; L(x): not recorded.

- [Classification on Medical data](<https://www.sota2.com/research/sota/classification-on-medical-data.md>): Classification; Medical data; L(x) Distance: not recorded.

- [Image Classification on CIFAR100 (Adversarial Robustness Metrics)](<https://www.sota2.com/research/sota/image-classification-on-cifar100-adversarial-robustness-metrics.md>): Image Classification; CIFAR100; L(x) Value: not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2606.04767>)

[PDF](<https://arxiv.org/pdf/2606.04767.pdf>)

[Implementation](<https://github.com/franz-chang/SRP>)
