Research map / Adversarial robustness and security / Certified and empirical robustness
Simple and Effective Stochastic Neural Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use activation-uncertainty regularization and entropy-based pruning | SE-SNN on MNIST with LeNet-5-Caffe · MNIST | FLOPs (Mil): 0.094 -> 0.061 | Helped |
| Use activation-uncertainty regularization and entropy-based pruning | SE-SNN on MNIST with LeNet-5-Caffe · MNIST | Memory (%): 15.55 -> 11.08 | Helped |
| Use activation-uncertainty regularization and entropy-based pruning | SE-SNN on MNIST with LeNet-5-Caffe · MNIST | Error (%): 1.0 -> 0.9 | Helped |
| Use activation-uncertainty regularization and entropy-based pruning | SE-SNN on CIFAR10 with VGG16 · CIFAR10 | FLOPs (Mil): 70.63 -> 53.61 | Helped |
| Use activation-uncertainty regularization and entropy-based pruning | SE-SNN on CIFAR10 with VGG16 · CIFAR10 | Error (%): 8.8 -> 8.0 | Helped |
| Use activation-uncertainty regularization and entropy-based pruning | SE-SNN on CIFAR100 with VGG16 variant · CIFAR100 | FLOPs (Mil): 203.10 -> 181.31 | Helped |
| Use activation-uncertainty regularization and entropy-based pruning | SE-SNN on CIFAR100 with VGG16 variant · CIFAR100 | Error (%): 25.9(25.7) -> 26.2 | Mixed |
| Learn data-dependent activation noise rather than using fixed-variance noise | SE-SNN adversarial defense · MNIST, untargeted FGS and CW-L2 attacks | accuracy | Helped |
| Train without adversarial attack samples | SE-SNN adversarial defense · MNIST, untargeted FGS | accuracy | Helped |
| Learn activation uncertainty to allocate high variance to hard or noisy instances | SE-SNN label-noise learning · MNIST with patterned label noise | test accuracy | Helped |
| Combine SE-SNN with Forward Correction | SE-SNN label-noise learning · MNIST with patterned label noise | test accuracy | Helped |
| Add a stochastic layer and temperature-scale the model | ResNet-110 with SE-SNN · CIFAR-10 | ECE (%): 1.39 -> 1.07 | Helped |
Closest papers by meaning
- Weight-covariance alignment for adversarially robust neural networks (ICML 2021 · similarity 0.88)
- Understanding and Exploring the Network with Stochastic Architectures (NeurIPS 2020 · similarity 0.87)
- How Sampling Impacts the Robustness of Stochastic Neural Networks (NeurIPS 2022 · similarity 0.86)
- Partially Stochastic Infinitely Deep Bayesian Neural Networks (ICML 2024 · similarity 0.86)
- Enhancing Adversarial Robustness in SNNs with Sparse Gradients (ICML 2024 · similarity 0.85)
- Stochastic Forward-Forward Learning through Representational Dimensionality Compression (NeurIPS 2025 · 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.