Research map / Probabilistic and scientific modeling / Variational and generative sampling
Neural Markov Jump Processes
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 irregular observation grids rather than regular grids | NeuralMJP on Discrete Flashing Ratchet · Discrete Flashing Ratchet parameter inference | parameter recovery | Helped |
| Train NeuralMJP on the full ion-channel dataset instead of a one-second window | NeuralMJP · Switching ion channel stationary distribution | stationary probabilities: Bottom: $0.17672$ -> $0.19631$; Middle: $0.09472 $ -> $0.07153 $; Top: $0.72856$ -> $0.73217$ | Mixed |
| Use two-step optimization, training encoder-decoder and prior parameters separately (ablation) | NeuralMJP · Ground-truth MJP experiments | parameter recovery | Helped |
| Train encoder-decoder and prior parameters together in one back-propagation step (ablation) | NeuralMJP · Ground-truth MJP experiments | parameter recovery | Hurt |
| Replace explicit prior-rate parameters with a neural generator of rates (ablation) | NeuralMJP · Prior transition-rate inference | training stability | Helped |
| Use direct trainable prior-rate parameters instead of a neural generator (ablation) | NeuralMJP · Prior transition-rate inference | training stability | Hurt |
| Add a Mercer time embedding to the posterior-rate network (ablation) | NeuralMJP · Posterior dynamics | learned dynamics | Helped |
| Use augmented posterior dynamics to approximate the KL integral instead of Gaussian quadrature (ablation) | NeuralMJP · KL approximation | computational speed | Hurt |
| Use Tanh-Sinh quadrature instead of Gaussian quadrature (ablation) | NeuralMJP · KL approximation | computational efficiency | No clear effect |
| Add an emission-covariance scheduler and gradually increase the number of observations during training (ablation) | NeuralMJP with Gaussian emissions · Training stability | index collapse occurrence | Helped |
| Allow the KL regularization to jointly smooth posterior rates during training (ablation) | NeuralMJP · Ground-truth MJP inference | transition-rate recovery | Hurt |
| Fit homogeneous prior MJPs to the optimized inhomogeneous posterior process in a separate step (ablation) | NeuralMJP · Ground-truth MJP inference | transition-rate recovery | Helped |
Closest papers by meaning
- Foundation Inference Models for Markov Jump Processes (NeurIPS 2024 · similarity 0.91)
- Piecewise deterministic generative models (NeurIPS 2024 · similarity 0.88)
- Variational Inference for SDEs Driven by Fractional Noise (ICLR 2024 · similarity 0.86)
- Latent Representation and Simulation of Markov Processes via Time-Lagged Information Bottleneck (ICLR 2024 · similarity 0.86)
- Bridging discrete and continuous state spaces: Exploring the Ehrenfest process in time-continuous diffusion models (ICML 2024 · similarity 0.86)
- LEAPS: A discrete neural sampler via locally equivariant networks (ICML 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.