Research map / Adversarial robustness and security / Certified and empirical robustness
Efficient Certified Training and Robustness Verification of Neural ODEs
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 GAINS certified training instead of standard or adversarial training | NODE image classifiers · MNIST, ε=0.10 | certified accuracy: 0.0$^{\pm 0.0}$ -> 89.0$^{\pm 1.1}$ | Helped |
| Use GAINS certified training instead of standard or adversarial training | NODE image classifiers · FMNIST, ε=0.10 | certified accuracy: 0.0$^{\pm 0.0}$ -> 62.5$^{\pm 1.1}$ | Helped |
| Increase GAINS training perturbation from εt=0.11 to εt=0.22 (ablation) | NODE image classifiers · MNIST, clean | standard accuracy: 95.5$^{\pm 0.1}$ -> 91.8$^{\pm 1.3}$ | Hurt |
| Increase GAINS training perturbation from εt=0.11 to εt=0.22 (ablation) | NODE image classifiers · MNIST, ε=0.20 | certified accuracy: 0.2$^{\pm 0.2}$ -> 79.7$^{\pm 3.4}$ | Helped |
| Increase GAINS training perturbation from εt=0.11 to εt=0.16 (ablation) | NODE image classifiers · FMNIST, clean | standard accuracy: 75.1$^{\pm 1.2}$ -> 71.5$^{\pm 1.7}$ | Hurt |
| Increase GAINS training perturbation from εt=0.11 to εt=0.16 (ablation) | NODE image classifiers · FMNIST, ε=0.15 | certified accuracy: 13.3$^{\pm 3.1}$ -> 55.0$^{\pm 4.3}$ | Helped |
| Use GAINS certified training | Latent ODE forecasting models · Physio-Net 6h, ε=0.05 | certified robustness rate: 0.0$^{\pm 0.0}$ -> 93.0$^{\pm 2.7}$ | Helped |
| Increase GAINS training perturbation from εt=0.1 to εt=0.2 (ablation) | Latent ODE forecasting models · Physio-Net 6h, clean | Std. MAE [$\times 10^{-2}$]: 51.1$^{\pm 2.0}$ -> 57.6$^{\pm 2.5}$ | Hurt |
| Increase GAINS training perturbation from εt=0.1 to εt=0.2 (ablation) | Latent ODE forecasting models · Physio-Net 6h, ε=0.20 | certified robustness rate: 24.2$^{\pm 7.7}$ -> 70.5$^{\pm 18.9}$ | Helped |
| Increase GAINS training perturbation from εt=0.1 to εt=0.2 (ablation) | Latent ODE forecasting models · Physio-Net 12h, clean | Std. MAE [$\times 10^{-2}$]: 50.9$^{\pm 0.4}$ -> 52.9$^{\pm 0.1}$ | Hurt |
| Increase GAINS training perturbation from εt=0.1 to εt=0.2 (ablation) | Latent ODE forecasting models · Physio-Net 24h, clean | Std. MAE [$\times 10^{-2}$]: 51.5$^{\pm 0.1}$ -> 53.7$^{\pm 0.7}$ | Hurt |
| Train with GAINS rather than standard or adversarial training | NODE classifiers · MNIST, ε=0.10 trajectory attack | attack success: 98.9$^{\pm 0.3}$ -> 73.4$^{\pm 3.5}$ | Helped |
Closest papers by meaning
- Simulation-Free Training of Neural ODEs on Paired Data (NeurIPS 2024 · similarity 0.89)
- Semi-Implicit Neural Ordinary Differential Equations (AAAI 2025 · similarity 0.89)
- From Fourier to Neural ODEs: Flow Matching for Modeling Complex Systems (ICML 2024 · similarity 0.88)
- Efficient Training of Neural Stochastic Differential Equations by Matching Finite Dimensional Distributions (ICLR 2025 · similarity 0.88)
- How Deep Do We Need: Accelerating Training and Inference of Neural ODEs via Control Perspective (ICML 2024 · similarity 0.88)
- Coupling Graph Neural Networks with Fractional Order Continuous Dynamics: A Robustness Study (AAAI 2024 · similarity 0.87)
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.