Research map / Architectures and efficiency / Pruning and efficient tuning
MoreauGrad: Sparse and Robust Interpretation of Neural Networks via Moreau Envelope
Jingwei Zhang, Farzan Farnia
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| increase Moreau envelope coefficient rho from 0 to 1 (rho=0 is simple gradient) (ablation) | Vanilla MoreauGrad · CIFAR-10 / ImageNet saliency maps | saliency map quality (qualitative) | Improved |
| make rho too large (ablation) | Vanilla MoreauGrad · CIFAR-10 / ImageNet saliency maps | saliency map quality and optimization | Hurt |
| increase L1 penalty coefficient eta in Sparse MoreauGrad (ablation) | Sparse MoreauGrad · CIFAR-10 / ImageNet saliency maps | sparsity of interpretation map | Improved |
| use L2,1 group-sparse penalty on adjacent pixel groups (ablation) | Group-Sparse MoreauGrad · CIFAR-10 / ImageNet saliency maps | structure of saliency map | Improved |
| add sparsity (L1) to MoreauGrad (ablation) | MoreauGrad · CIFAR-10 / ImageNet saliency maps | visual quality | Improved |
| replace simple gradient, IG, DeepLIFT, SmoothGrad baselines with MoreauGrad | interpretation methods · CIFAR-10, Tiny-ImageNet, ImageNet under L2 interpretation attack | robustness (L2 distance, top-k intersection, SSIM) | Improved |
| add Gaussian smoothing to enforce weak convexity before Moreau envelope (regularized mode) | MoreauGrad · theory | weak convexity of score function | Improved |
Closest papers by meaning
- Structured Gradient-based Interpretations via Norm-Regularized Adversarial Training (CVPR 2024 · similarity 0.86)
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.85)
- From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit (NeurIPS 2025 · similarity 0.85)
- Global Minimizers of $\ell^p$-Regularized Objectives Yield the Sparsest ReLU Neural Networks (NeurIPS 2025 · similarity 0.84)
- Deeper or Wider: A Perspective from Optimal Generalization Error with Sobolev Loss (ICML 2024 · similarity 0.84)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
