arXiv：cs.LG· Fredrik Cumlin, Saikat Chatterjee·· 3 小时前
可微系统重采样（DSR）为变分序列蒙特卡洛实现全梯度流
Differentiable Systematic Resampling for Variational Sequential Monte Carlo
AI 导读
研究人员提出可微系统重采样（DSR），一种受温度控制的系统重采样松弛方法，在保留系统重采样 CDF 有序带状结构的同时实现全梯度流。DSR 在温度趋于零时收敛到精确的系统重采样，并证明了诱导偏差的逐点指数收敛速率。相比基于最优传输的可微重采样，DSR 无需迭代求解器、计算开销显著更低，在随机动力系统和真实手写数据上取得相当或更优的滤波与动力学学习表现，已被 NeurIPS 2026 接收。
正文
Abstract:Particle filters are a standard tool for nonlinear state estimation, but their resampling step is discrete, preventing gradient-based learning in variational sequential Monte Carlo. We introduce Differentiable Systematic Resampling (DSR), a temperature-controlled relaxation of systematic resampling, that preserves the CDF-ordered, banded structure of systematic resampling while enabling full gradient flow. DSR converges to exact systematic resampling as the temperature vanishes, and we prove a pointwise exponential convergence rate for the induced bias. Compared to optimal-transport-based differentiable resampling, DSR avoids iterative solvers and has substantially lower computational overhead. Experiments on stochastic dynamical systems and real-world handwriting data show that DSR achieves comparable or superior filtering and dynamics learning performance.
| Comments: | Accepted to NeurIPS 2026 |
| Subjects: | Machine Learning (stat.ML); Machine Learning (cs.LG); Signal Processing (eess.SP) |
| Cite as: | arXiv:2610.12094 [stat.ML] |
| (or arXiv:2610.12094v1 [stat.ML] for this version) | |
| https://doi.org/10.48550/arXiv.2610.12094 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Fredrik Cumlin Mr [view email]
[v1]
Thu, 8 Oct 2026 15:01:45 UTC (826 KB)
来源：arXiv：cs.LG · arxiv.org