Parallel Broyden Methods for Efficiently Evaluating Nonlinear State Space Models

DOI10.48550/ARXIV.2611.13523MaRDI QIDQ7427288arXivFDO

Authors Ian-Christopher Tanoh, Scott W. Linderman, Ian Christopher Tanoh

Publication date 10 November 2026

Published in arXiv

Tldr We introduce a parallel Broyden's method with block diagonal-plus-low-rank approximations to efficiently evaluate nonlinear state space models on modern hardware, capturing interactions while retaining favorable scaling and avoiding automatic differentiation; we show empirically that our method converges faster than other quasi-Newton methods in various settings.

Full work available at URL https://doi.org/10.48550/ARXIV.2611.13523

arXiv classification

cs.LGmath.NA

Cites work

• A Class of Methods for Solving Nonlinear Simultaneous Equations
• A parallel-in-time Newton's method-based ODE solver
• A stochastic estimator of the trace of the influence matrix for laplacian smoothing splines
• Attention Is All You Need
• Bayesian data analysis
• Equation of state calculations by fast computing machines
• Long Short-Term Memory
• Matrix Compression via Randomized Low Rank and Low Precision Factorization
• MCMC using Hamiltonian dynamics
• Monte Carlo sampling methods using Markov chains and their applications
• Optimal Scaling of Discrete Approximations to Langevin Diffusions
• Randomized algorithms for low-rank matrix factorizations: sharp performance bounds
• State
• Stochastic Relaxation, Gibbs Distributions, and the Bayesian Restoration of Images
• SYMMETRIC GAUGE FUNCTIONS AND UNITARILY INVARIANT NORMS
• The approximation of one matrix by another of lower rank.
• The Convergence of a Class of Double-rank Minimization Algorithms 1. General Considerations
• Two classes of multisecant methods for nonlinear acceleration
• Updating quasi-Newton matrices with limited storage

This page was built for publication: Parallel Broyden Methods for Efficiently Evaluating Nonlinear State Space Models

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7427288)