Research map / Probabilistic and scientific modeling / Variational and generative sampling
Differentiable Particle Filtering via Entropy-Regularized Optimal Transport
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 differentiable entropy-regularized transport resampling instead of standard PF resampling | VRNN · JSB test | ELBO: $-7.92_{\pm 0.13}$ -> $-7.67_{\pm 0.08}$ | Helped |
| Use differentiable entropy-regularized transport resampling instead of soft-resampling | VRNN · JSB test | ELBO: $-8.17_{\pm 0.07}$ -> $-7.67_{\pm 0.08}$ | Helped |
| Use differentiable entropy-regularized transport resampling instead of standard PF resampling | VRNN · MuseData test | ELBO: $-7.60_{\pm 0.06}$ -> $-\mathbf{7.59}_{\pm 0.01 }$ | No clear effect |
| Use differentiable entropy-regularized transport resampling instead of standard PF resampling | VRNN · Nottingham test | ELBO: $-3.81_{\pm 0.02}$ -> $\mathbf{-3.79}_{\pm 0.02}$ | No clear effect |
| Use differentiable entropy-regularized transport resampling instead of standard PF resampling | robot localization model · Maze 1 test | MSE: $10.71_{\pm 0.45}$ -> $\mathbf{3.55}_{\pm \mathbf{0.20}}$ | Helped |
| Use differentiable entropy-regularized transport resampling instead of soft-resampling | robot localization model · Maze 1 test | MSE: $9.14_{\pm 0.39}$ -> $\mathbf{3.55}_{\pm \mathbf{0.20}}$ | Helped |
| Use differentiable entropy-regularized transport resampling instead of standard PF resampling | robot localization model · Maze 2 test | MSE: $11.86_{\pm 0.57}$ -> $4.65_{\pm 0.50}$ | Helped |
| Use differentiable entropy-regularized transport resampling instead of standard PF resampling | robot localization model · Maze 3 test | MSE: $12.88_{\pm 0.65}$ -> $4.44_{\pm \mathbf{0.26}}$ | Helped |
| Train proposal parameters with DPF gradients instead of PF gradients that omit resampling terms | linear Gaussian SSM proposal · proposal learning (100 realizations) | RMSE of φ − 1: 0.22 -> 0.11 | Helped |
| Train proposal parameters with DPF gradients instead of PF gradients that omit resampling terms | linear Gaussian SSM proposal · proposal learning (100 realizations) | average ESS after convergence: 25\% -> around 60\% | Helped |
| Replace standard PF resampling with DET at ε=0.25 | linear Gaussian SSM likelihood estimation · θ=(0.25,0.25), 25 particles | mean of (estimated log-likelihood − true log-likelihood)/T: -1.13 -> -1.14 | No clear effect |
| Vary DET regularization to ε=0.25, ε=0.5, or ε=0.75 (ablation) | DPF likelihood estimation · linear Gaussian SSM; θ=(0.25,0.25), (0.5,0.5), and (0.75,0.75) | standard deviation of log-likelihood estimates: 0.20; 0.18; 0.19 (ε=0.25), 0.20; 0.18; 0.18 (ε=0.5), 0.20; 0.18; 0.18 (ε=0.75) | No clear effect |
Closest papers by meaning
- Nonlinear Filtering with Brenier Optimal Transport Maps (ICML 2024 · similarity 0.87)
- Momentum Particle Maximum Likelihood (ICML 2024 · similarity 0.85)
- Functional Gradient Flows for Constrained Sampling (NeurIPS 2024 · similarity 0.85)
- Partially Observed Trajectory Inference using Optimal Transport and a Dynamics Prior (ICLR 2025 · similarity 0.84)
- To discretize continually: Mean shift interacting particle systems for Bayesian inference (NeurIPS 2026 · similarity 0.84)
- Particle Semi-Implicit Variational Inference (NeurIPS 2024 · similarity 0.84)
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.