Research map / Probabilistic and scientific modeling / Dynamical systems
Neural Hybrid Automata: Learning Dynamics With Multiple Modes and Stochastic Transitions
Michael Poli, Stefano Massaroli, Luca Scimeca, Sanghyuk Chun, Seong Joon Oh, Atsushi Yamashita, Hajime Asama, Jinkyoo Park, Animesh Garg
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 |
|---|---|---|---|
| Use NHA mode encoder instead of DBSCAN | NHA · TCP mode recovery, m=3 | v-measure: 0.66 -> 0.86 | Improved |
| Use NHA mode encoder instead of k-means++ | NHA · TCP mode recovery, m=10 | v-measure: 0.30 -> 0.96 | Improved |
| Increase number of latent modes m from 3 to 10 (ablation) | NHA · TCP mode recovery | v-measure: 0.86 -> 0.96 | Improved |
| Increase number of supervising trajectories n from 1 to 10 (ablation) | NHA event module · TCP event module | flow NLL: 2.761 -> 2.313 | Improved |
| Increase number of supervising trajectories n from 1 to 10 (ablation) | NHA event module · TCP event module | jump map MSE (1e-3): 1.435 -> 0.003 | Improved |
| Increase n from 3 to 10 (ablation) | NHA event module · TCP event module | flow NLL: 2.375 -> 2.313 | No clear effect |
| Add segmentation noise p=0.5 (ablation) | NHA-5 · TCP mode recovery | v-measure: 0.91 -> 0.43 | Hurt |
| Add segmentation noise p=0.3 (ablation) | NHA-3 · TCP mode recovery | v-measure: 0.86 -> 0.73 | Hurt |
| Add segmentation noise p=0.5 (ablation) | DBSCAN · TCP mode recovery | v-measure: 0.69 -> 0.23 | Hurt |
| Add segmentation noise p=0.1 (ablation) | NHA-3 · TCP mode recovery | v-measure: +0.02 | No clear effect |
| Use softmax latents instead of categorical sampling (ablation) | NHA encoder · Switching linear system | mode recovery quality | Hurt |
| Use one extra latent mode beyond system modes (overclustering) (ablation) | NHA · Training stability | training stability | Improved |
Closest papers by meaning
- Amortized Equation Discovery in Hybrid Dynamical Systems (ICML 2024 · similarity 0.86)
- Identifying latent state transitions in non-linear dynamical systems (ICLR 2025 · similarity 0.86)
- Automatically Learning Hybrid Digital Twins of Dynamical Systems (NeurIPS 2024 · similarity 0.86)
- Learning Hybrid Dynamics Models with Simulator-Informed Latent States (AAAI 2024 · similarity 0.85)
- Safely Learning Controlled Stochastic Dynamics (NeurIPS 2025 · similarity 0.85)
- Deep learning for continuous-time stochastic control with jumps (NeurIPS 2025 · 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.
