Pith. sign in

REVIEW 8 cited by

Neural Stochastic Differential Equations: Deep Latent Gaussian Models in the Diffusion Limit

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 1905.09883 v2 pith:CBNKCAUG submitted 2019-05-23 cs.LG stat.ML

classification cs.LGstat.ML
keywords diffusionlatentneuralstochasticgaussianmodelsautomaticdeep
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

In deep latent Gaussian models, the latent variable is generated by a time-inhomogeneous Markov chain, where at each time step we pass the current state through a parametric nonlinear map, such as a feedforward neural net, and add a small independent Gaussian perturbation. This work considers the diffusion limit of such models, where the number of layers tends to infinity, while the step size and the noise variance tend to zero. The limiting latent object is an It\^o diffusion process that solves a stochastic differential equation (SDE) whose drift and diffusion coefficient are implemented by neural nets. We develop a variational inference framework for these \textit{neural SDEs} via stochastic automatic differentiation in Wiener space, where the variational approximations to the posterior are obtained by Girsanov (mean-shift) transformation of the standard Wiener process and the computation of gradients is based on the theory of stochastic flows. This permits the use of black-box SDE solvers and automatic differentiation for end-to-end inference. Experimental results with synthetic data are provided.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 8 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Pathwise Learning of Stochastic Dynamical Systems with Partial Observations

    math.OC 2026-01 unverdicted novelty 7.0 of 10

    A pathwise Zakai-equation control formulation is used to train conditional neural SDEs that amortize nonlinear filtering of partially observed stochastic dynamics.

  2. Data-to-Energy Stochastic Dynamics

    cs.LG 2025-09 conditional novelty 6.0 of 10

    A new data-to-energy iterative proportional fitting algorithm trains Schrödinger bridges when endpoint distributions are known only through unnormalised densities.

  3. Efficient Training of Neural SDEs Using Stochastic Optimal Control

    cs.LG 2025-05 conditional novelty 6.0 of 10

    A hybrid variational inference method for neural SDEs that initializes the control function with the closed-form optimal linear control, then learns a neural residual.

  4. Fourier Neural Operators for Non-Markovian Processes:Approximation Theorems and Experiments

    cs.LG 2025-07 conditional novelty 5.0 of 10

    A mirror-padded Fourier neural operator can approximate, with proven error bounds, solution maps of path-dependent SDEs and Lipschitz transformations of fractional Brownian motion.

  5. Generalization in VAE and Diffusion Models: A Unified Information-Theoretic Analysis

    cs.LG 2025-06 reject novelty 5.0 of 10

    The authors derive information-theoretic generalization bounds for VAEs and diffusion models that expose a trade-off in the diffusion time T, and propose using the computable bound to select T and regularize training.

  6. Numerical PDE solvers outperform neural PDE solvers

    math.NA 2025-07 conditional novelty 4.0 of 10

    DeepFDM, a differentiable finite-difference solver that learns PDE coefficients, achieves 10 to 40 times lower error than FNO, U-Net and ResNet on five scalar time-dependent PDEs in one to three dimensions.

  7. Quantum-Inspired Differentiable Integral Neural Networks (QIDINNs): A Feynman-Based Architecture for Continuous Learning Over Streaming Data

    cs.SE 2025-06 reject novelty 2.0 of 10

    QIDINNs define parameter updates as kernel-weighted integrals of past gradients, essentially continuous-time momentum, and claim superior streaming learning without providing the backpropagation cost they claim to avoid.

  8. Deep Neural Networks Inspired by Differential Equations

    cs.LG 2025-10 unverdicted

    A review of differential-equation-inspired neural networks that compiles known results into a taxonomy, with no new experiments or theory.

Pith tools