Research map / Probabilistic and scientific modeling / Dynamical systems
ODE2VAE: Deep generative second order ODEs with Bayesian neural networks
Cagatay Yildiz, Markus Heinonen, Harri Lahdesmaki
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 |
|---|---|---|---|
| add KL penalty between encoder and ODE-flow distributions (ODE2VAE-KL vs ODE2VAE) (ablation) | ODE2VAE · CMU Mocap-1 | MSE on future frames: 93.07 -> 15.99 | Improved |
| add KL penalty between encoder and ODE-flow distributions (ablation) | ODE2VAE · CMU Mocap-2 | MSE on future frames: 10.06 -> 8.09 | Improved |
| add KL penalty between encoder and ODE-flow distributions (ablation) | ODE2VAE · Rotating MNIST test angle | prediction error: 0.0194 -> 0.0188 | Improved |
| second-order Bayesian latent ODE instead of first-order deterministic Neural ODE | latent ODE VAE · CMU Mocap-1 | MSE on future frames: 87.23 -> 15.99 | Improved |
| second-order Bayesian latent ODE instead of first-order deterministic Neural ODE | latent ODE VAE · CMU Mocap-2 | MSE on future frames: 22.49 -> 8.09 | Improved |
| continuous-time ODE latent dynamics instead of RNN-based DTSBN-S | deep generative sequence model · CMU Mocap-1 | MSE on future frames: 80.21 -> 15.99 | Improved |
| plain ODE2VAE vs black-box npODE on small dataset | ODE2VAE · CMU Mocap-1 | MSE on future frames: 45.74 -> 93.07 | Hurt |
| ODE2VAE-KL vs GPPVAE-joint on rotating MNIST | ODE2VAE-KL · Rotating MNIST test angle | prediction error: 0.0288 -> 0.0188 | Improved |
| CNN encoder on channel-stacked frames instead of RNN encoder (ablation) | ODE2VAE · Bouncing balls | prediction MSE | Improved |
| ODE2VAE vs DDPAE and DTSBN-S on bouncing balls | ODE2VAE · Bouncing balls | per-pixel MSE on 10 future frames | Improved |
| larger second dataset (Mocap-2) vs smaller (Mocap-1) | all models · CMU Mocap | MSE on future frames: 93.07 -> 10.06 | Improved |
Later papers that compared with this method
3 later accepted papers report ODE2VAE as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Meta-iLaD: Identifiable Latent Dynamics via Meta-Learning of Dynamics Environments (ICML 2026 · better on 6 of 6 shared results)
- Equivariant Neural Simulators for Stochastic Spatiotemporal Dynamics (NeurIPS 2023 · better on 3 of 3 shared results)
- Scalable Inference in SDEs by Direct Matching of the Fokker–Planck–Kolmogorov Equation (NeurIPS 2021 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ODEWorld: A Continuous Predictive Architecture via Physical-Time Flow (NeurIPS 2026 · similarity 0.87)
- Balanced Neural ODEs: nonlinear model order reduction and Koopman operator approximations (ICLR 2025 · similarity 0.86)
- Improving Neural ODE Training with Temporal Adaptive Batch Normalization (NeurIPS 2024 · similarity 0.86)
- Simulation-Free Training of Neural ODEs on Paired Data (NeurIPS 2024 · similarity 0.85)
- CSG-ODE: ControlSynth Graph ODE For Modeling Complex Evolution of Dynamic Graphs (ICML 2025 · similarity 0.85)
- How Deep Do We Need: Accelerating Training and Inference of Neural ODEs via Control Perspective (ICML 2024 · similarity 0.85)
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.
