Research map / Learning theory and algorithms / Prediction and decision losses
BNN-DP: Robustness Certification of Bayesian Neural Networks via Dynamic Programming
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 BNN-DP instead of FL | Fully connected BNNs on 2D Noisy Sine and Kin8nm · 2D Noisy Sine and Kin8nm | gamma-robustness: 0.041 vs 0.8; 0.044 vs 0.3 | Helped |
| Use BNN-DP instead of FL | Fully connected BNNs on MNIST and Fashion MNIST · MNIST and Fashion MNIST | epsilon-robustness: 0.0150 vs 0.0102; 0.0128 vs 0.008 | Mixed |
| Use BNN-DP instead of FL | Bayesian CNNs on Fashion MNIST and CIFAR-10 · Fashion MNIST and CIFAR-10 | epsilon-robustness: 0.00065 vs 0.00122; 0.00061 vs 0.00117; 0.00007 vs 0.00010; 0.00011 vs 0.00021 | Hurt |
| Use BNN-DP instead of FL | Classification BNNs · MNIST, Fashion MNIST, and CIFAR-10 | computation time: 15.2 vs 859.2; 3.7 vs 545.5; 1.8 vs 250.7 | Helped |
| Average layer uncertainty using the backward dynamic program rather than optimize worst-case parameter combinations | BNN robustness certification · 1D Noisy Sine | certified output interval width: [-1.68, 0.59] vs BNN-DP tighter bounds; [-7.07, 9.31] vs BNN-DP tighter bounds | Helped |
Closest papers by meaning
- Adv-BNN: Improved Adversarial Defense through Robust Bayesian Neural Network (ICLR 2019 · similarity 0.89)
- Robustness of Bayesian Neural Networks to Gradient-Based Attacks (NeurIPS 2020 · similarity 0.89)
- Robust Learning of Fixed-Structure Bayesian Networks (NeurIPS 2018 · similarity 0.88)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.87)
- Verifying Properties of Binary Neural Networks Using Sparse Polynomial Optimization (ICLR 2025 · similarity 0.86)
- Understanding Certified Training with Interval Bound Propagation (ICLR 2024 · similarity 0.86)
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.