arxiv: 2604.25502 · v1 · submitted 2026-04-28 · 🧮 math.NA · cs.NA

A Discrete-Time Random Feature Method for Nonlinear Evolution Equations with Implicit-Explicit Runge--Kutta Time Stepping

Haoran Zhou , Zhaohui Fu , Yangshuai Wang , Xinlong Feng This is my paper

Pith reviewed 2026-05-07 15:39 UTC · model grok-4.3

classification 🧮 math.NA cs.NA

keywords random feature methodIMEX Runge-Kuttanonlinear evolution equationslinear least-squareserror estimatesAllen-Cahn equationBurgers equation

0 comments

The pith

Nonlinear evolution equations are advanced in discrete time by representing solutions in random feature spaces and solving each IMEX-RK stage via linear least-squares after operator splitting.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a fully discrete numerical method for nonlinear time-dependent PDEs that treats time stepping explicitly rather than embedding time as an input variable. At each time level the solution is expanded in a random feature basis, the nonlinear operator is split into linear and nonlinear parts, and an implicit-explicit third-order Runge-Kutta scheme produces a sequence of linear least-squares problems. A global error bound is derived that isolates the contributions from the random-feature approximation, the least-squares coefficient perturbations, and the temporal discretization error. Experiments on the Allen-Cahn, Burgers, KdV and Cahn-Hilliard equations confirm that relative L2 errors reach order 10^{-6} and that observed convergence rates match the third-order time scheme.

Core claim

The method advances the solution from one time level to the next by projecting onto a random feature trial space and applying a four-stage third-order IMEX-RK integrator; after linear-nonlinear splitting each stage reduces to an unconstrained linear least-squares problem whose solution yields the coefficients for the next time level, with a global error estimate separating the spatial approximation, coefficient perturbation and time-discretization contributions.

What carries the argument

Stage-wise linear least-squares formulation obtained by splitting the nonlinear operator into linear and nonlinear terms inside a third-order IMEX-RK time discretization applied to a random-feature spatial expansion at each discrete time level.

If this is right

The fully discrete scheme attains relative L2 errors of order 10^{-6} on the Allen-Cahn, Burgers, KdV and Cahn-Hilliard equations.
Observed spatial-temporal convergence rates remain consistent with the underlying third-order IMEX-RK integrator.
The method produces higher accuracy at lower cost than a comparable IMEX-PINN formulation on the same test problems.
The global error estimate cleanly separates contributions from random-feature approximation, least-squares perturbations and time discretization.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The linear least-squares structure at each stage suggests straightforward parallelization across time steps or across feature blocks.
If a stable splitting exists for other time integrators, the same discrete-time random-feature framework could be reused without redesigning the spatial approximation.
Adaptive selection or refinement of the random feature set could further reduce the number of features needed while keeping the least-squares residual below the temporal error.

Load-bearing premise

The nonlinear operator must admit a splitting into linear and nonlinear parts that preserves the third-order accuracy of the IMEX-RK scheme, while the random feature space must be rich enough that the least-squares error at each time level remains controllable.

What would settle it

Numerical tests on a problem whose splitting destroys third-order temporal accuracy, or on a sequence of random-feature spaces whose least-squares residuals fail to decrease with increasing feature count, would violate the predicted global error bound.

Figures

Figures reproduced from arXiv: 2604.25502 by Haoran Zhou, Xinlong Feng, Yangshuai Wang, Zhaohui Fu.

**Figure 1.** Figure 1: Illustration of the proposed discrete-time IMEX-RFM method. Local random feature approximations are constructed on subdomains, assembled by the partition-of-unity mechanism, and then updated through stage-wise IMEX-RK time stepping. where ωij ∈ R are unknown coefficients and N = MJn is the total number of degrees of freedom. For vector-valued solutions with dimension du, each component is approximated by (… view at source ↗

**Figure 2.** Figure 2: 1D Allen–Cahn equation: (a) relative L 2 -error at the final time versus time step size; (b) pointwise errors between the reference and predicted solutions at t = 0.5. 4.2. 1D Burgers’ Equation. We next consider the one-dimensional Burgers’ equation, which provides a Type-II test case because the nonlinear term contains a spatial derivative. The equation is ∂tu + u∂xu = ν∂xxu, ν = 1 10π , (x, t) ∈ [−1, 1] … view at source ↗

**Figure 3.** Figure 3: 1D Burgers’ equation: (a) relative L 2 -error at the final time versus time step size; (b) pointwise errors between the reference and predicted solutions at t = 0.5. The initial condition is u(x, 0) = cos(πx), and periodic boundary conditions are imposed: u(−1, t) = u(1, t), ∂xu(−1, t) = ∂xu(1, t), ∂xxu(−1, t) = ∂xxu(1, t). The splitting is L(u) = α 2∂xxxu and G(u) = −u∂xu, again a Type-II problem. Since t… view at source ↗

**Figure 4.** Figure 4: 1D Korteweg–De Vries equation: (a) relative L 2 -error at the final time versus time step size; (b) pointwise errors between the reference and predicted solutions at t = 0.5 view at source ↗

**Figure 5.** Figure 5: 1D Cahn–Hilliard equation: (a) relative L 2 -error at the final time versus time step size; (b) pointwise errors between the reference and predicted solutions at t = 0.5 view at source ↗

**Figure 6.** Figure 6: Comparison of the relative L 2 -errors of IMEX-RFM and IMEX-PINN for the 2D Allen–Cahn equation. due to random feature variance, though the overall trend is consistent with third-order convergence. The IMEX-RFM method again outperforms the IMEX-PINN variant view at source ↗

**Figure 7.** Figure 7: 1D Allen–Cahn equation with ∆t = 10−2 : reference solution (left), predicted solution (center), and absolute error (right) view at source ↗

**Figure 8.** Figure 8: 1D Allen–Cahn equation with ∆t = 5 × 10−3 : reference solution (left), predicted solution (center), and absolute error (right) view at source ↗

**Figure 9.** Figure 9: 1D Burgers’ equation with ∆t = 5 × 10−2 : reference solution (left), predicted solution (center), and absolute error (right). [2] Uri M Ascher, Steven J Ruuth, and Brian TR Wetton. Implicit-explicit methods for time-dependent partial differential equations. SIAM Journal on Numerical Analysis, 32(3):797–823, 1995. [3] Vittorio E Badalassi, Hector D Ceniceros, and Sanjoy Banerjee. Computation of multiphase s… view at source ↗

**Figure 10.** Figure 10: 1D Burgers’ equation with ∆t = 2 × 10−2 : reference solution (left), predicted solution (center), and absolute error (right) view at source ↗

**Figure 11.** Figure 11: 1D KdV equation with ∆t = 3 × 10−3 : reference solution (left), predicted solution (center), and absolute error (right) view at source ↗

**Figure 12.** Figure 12: 1D KdV equation with ∆t = 10−3 : reference solution (left), predicted solution (center), and absolute error (right). [8] Jingrun Chen, Weinan E, and Yifei Sun. Optimization of random feature method in the high-precision regime. Communications on Applied Mathematics and Computation, 6(2):1490–1517, 2024. [9] Jingrun Chen, Yixin Luo, et al. The random feature method for time-dependent problems. arXiv prepr… view at source ↗

**Figure 13.** Figure 13: 1D Cahn–Hilliard equation with ∆t = 10−2 : reference solution (left), predicted solution (center), and absolute error (right) view at source ↗

**Figure 14.** Figure 14: 1D Cahn–Hilliard equation with ∆t = 5 × 10−3 : reference solution (left), predicted solution (center), and absolute error (right) view at source ↗

**Figure 15.** Figure 15: 2D Allen–Cahn equation with ∆t = 3 × 10−2 : reference solution (left), predicted solution (center), and absolute error (right). [13] Kamal Choudhary, Brian DeCost, Chi Chen, Anubhav Jain, Francesca Tavazza, Ryan Cohn, Cheol Woo Park, Alok Choudhary, Ankit Agrawal, Simon JL Billinge, et al. Recent advances and applications of deep learning methods in materials science. npj Computational Materials, 8(1):59,… view at source ↗

**Figure 16.** Figure 16: 2D Allen–Cahn equation with ∆t = 10−2 : reference solution (left), predicted solution (center), and absolute error (right). [18] Junpeng Hu, Shi Jin, Nana Liu, and Lei Zhang. Quantum random feature method for solving partial differential equations, 2025. [19] Guang-Bin Huang, Lei Chen, and Chee-Kheong Siew. Universal approximation using incremental constructive feedforward networks with random hidden nod… view at source ↗

read the original abstract

We study a discrete-time random feature method for nonlinear, time-dependent partial differential equations. In contrast to continuous-time formulations that treat time as an additional input variable, the method advances the solution step by step, with each time level computed from previously available states. The spatial solution at each step is represented in the random feature trial space, and the time discretization is given by an implicit-explicit Runge--Kutta (IMEX-RK, 4 stages, third-order) scheme. After splitting the operator into linear and nonlinear parts, each stage admits a linear least-squares formulation, which avoids nonlinear least-squares solves. We also derive a global error estimate for the fully discrete method, separating the contributions of the stage-wise RFM approximation, perturbations in the least-squares coefficients, and the temporal discretization. Numerical experiments for the Allen--Cahn, Burgers, Korteweg--De Vries, and Cahn--Hilliard equations show relative $L^2$-errors of order $10^{-6}$ and convergence rates consistent with the third-order IMEX scheme. A comparison with an IMEX-PINN variant shows that the proposed method achieves higher accuracy at substantially lower computational cost.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper gives a discrete-time random feature method paired with third-order IMEX-RK stepping for nonlinear PDEs, plus a global error bound that separates approximation, least-squares, and time errors.

read the letter

The core advance is moving random features to a discrete-time setting for time-dependent nonlinear equations. The solution advances step by step from prior states, the spatial field lives in a random feature space, and an IMEX-RK scheme (third-order, four stages) handles time. After splitting the operator into linear and nonlinear pieces, every stage reduces to a linear least-squares problem instead of a nonlinear one. They also supply a global error estimate that isolates the random-feature approximation error, the perturbation from the least-squares coefficients, and the temporal truncation error. Numerical tests on Allen-Cahn, Burgers, KdV, and Cahn-Hilliard recover the design order with relative L2 errors near 10^{-6} and beat a comparable IMEX-PINN run on both accuracy and cost. That combination of discrete stepping, linear solves, and separated error bound is not in the earlier continuous-time or PINN literature they cite. The splitting assumption is load-bearing but appears chosen to preserve the IMEX order, and the experiments support it on the tested problems. A minor soft spot is that the number of random features remains a free parameter whose effect on the constants in the bound is not fully quantified in the abstract; if the feature space is too small or the splitting fails to stay stable, the practical error could exceed the predicted rate. Overall the work is for computational mathematicians who want efficient, non-neural alternatives for evolution PDEs. It is grounded enough in both analysis and numerics to deserve a serious referee, even if some details of the proof and implementation will need checking.

Referee Report

2 major / 2 minor

Summary. The paper introduces a discrete-time random feature method (RFM) for nonlinear time-dependent PDEs that advances the solution step-by-step in a random feature trial space using a 4-stage third-order IMEX Runge-Kutta scheme. After splitting the spatial operator into linear and nonlinear parts, each stage reduces to a linear least-squares problem. The manuscript derives a global error estimate that separates the contributions from stage-wise RFM approximation error, perturbations in the least-squares coefficients, and temporal discretization error. Numerical experiments on the Allen-Cahn, Burgers, KdV, and Cahn-Hilliard equations report relative L² errors of order 10^{-6} with observed convergence rates matching the design order of the IMEX scheme, and the method is compared favorably to an IMEX-PINN baseline in both accuracy and cost.

Significance. If the global error bound is rigorous and the linear-nonlinear splitting preserves the IMEX order under the stated conditions, the work supplies a computationally attractive, theoretically supported alternative to physics-informed neural networks for evolution equations. The separation of error sources, avoidance of nonlinear optimization, and demonstrated order recovery constitute clear strengths that could enable efficient long-time or high-dimensional simulations once the method is extended or optimized further.

major comments (2)

[Error Analysis section] The global error estimate (abstract and Error Analysis section) separates stage-wise RFM approximation, least-squares perturbations, and temporal discretization, but the proof must explicitly verify that the chosen linear-nonlinear splitting does not introduce order reduction for the specific third-order IMEX-RK scheme; this is load-bearing for the claim that the fully discrete method retains design order.
[Numerical Experiments section] Numerical Experiments section: the reported L² errors of order 10^{-6} and third-order rates are encouraging, yet the dependence of the least-squares perturbation term on the number of random features should be quantified (e.g., via an additional table or plot) to confirm that this contribution remains controllable and does not dominate the temporal error.

minor comments (2)

Clarify the precise Butcher tableau and stability function of the 4-stage third-order IMEX-RK scheme in the method formulation to allow direct reproduction of the stage equations.
In the comparison with IMEX-PINN, report the number of parameters, training epochs, and wall-clock times for both methods on the same hardware to make the computational-cost claim quantitative.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment and the constructive comments on the error analysis and numerical experiments. We address each major comment below.

read point-by-point responses

Referee: [Error Analysis section] The global error estimate (abstract and Error Analysis section) separates stage-wise RFM approximation, least-squares perturbations, and temporal discretization, but the proof must explicitly verify that the chosen linear-nonlinear splitting does not introduce order reduction for the specific third-order IMEX-RK scheme; this is load-bearing for the claim that the fully discrete method retains design order.

Authors: We agree that an explicit verification is required to confirm order preservation. In the revised manuscript we have added a new remark in Section 3 (Error Analysis) that directly checks the linear-nonlinear splitting against the order conditions of the specific 4-stage third-order IMEX-RK tableau. The remark verifies that the splitting satisfies the requisite consistency and stability requirements under the smoothness hypotheses already stated in the paper, thereby ensuring the global error bound retains the design order without reduction. revision: yes
Referee: [Numerical Experiments section] Numerical Experiments section: the reported L² errors of order 10^{-6} and third-order rates are encouraging, yet the dependence of the least-squares perturbation term on the number of random features should be quantified (e.g., via an additional table or plot) to confirm that this contribution remains controllable and does not dominate the temporal error.

Authors: We appreciate the suggestion to quantify the perturbation term. In the revised Numerical Experiments section we have added a new figure (Figure 7) that plots the least-squares perturbation error versus the number of random features for the Allen-Cahn and Burgers equations (representative cases). The figure shows that the perturbation decreases monotonically with feature count and remains at least one order of magnitude below the temporal discretization error for the feature counts used in the reported experiments, confirming it does not dominate. revision: yes

Circularity Check

0 steps flagged

Error bound derived by independent separation of contributions

full rationale

The paper's central derivation is a global error estimate obtained by decomposing the total error into three explicitly separated terms: stage-wise random-feature approximation error, perturbations arising from the least-squares solves, and the truncation error of the third-order IMEX-RK scheme. Each term is bounded using standard approximation theory for random features and stability properties of the IMEX splitting; none is obtained by fitting to the target solution or by renaming an input quantity. The numerical experiments serve only as consistency checks and do not enter the proof. No self-citation is invoked as a uniqueness theorem or load-bearing premise, and the linear-nonlinear splitting is stated as an assumption rather than derived from the result itself. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the random feature space being able to approximate solutions at each time level and on the validity of the linear-nonlinear operator splitting for the chosen PDEs.

free parameters (1)

Number of random features
Hyperparameter controlling the dimension of the trial space; chosen by the user to balance accuracy and cost.

axioms (1)

domain assumption The solution is sufficiently smooth for the error estimates of the random feature approximation and the IMEX-RK scheme to hold.
Standard assumption invoked when deriving the global error bound.

pith-pipeline@v0.9.0 · 5524 in / 1333 out tokens · 41074 ms · 2026-05-07T15:39:10.605125+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

43 extracted references · 1 canonical work pages

[1]

Christoffel adaptive sampling for sparse random feature expansions, 2026

Ben Adcock, Khiem Can, and Xuemeng Wang. Christoffel adaptive sampling for sparse random feature expansions, 2026. DISCRETE-TIME RFM FOR PDES 18 Figure 7. 1D Allen–Cahn equation with ∆t = 10 −2: reference solution (left), predicted solution (center), and absolute error (right). Figure 8. 1D Allen–Cahn equation with ∆t = 5 × 10−3: reference solution (left)...

2026
[2]

Implicit-explicit methods for time-dependent partial differential equations

Uri M Ascher, Steven J Ruuth, and Brian TR Wetton. Implicit-explicit methods for time-dependent partial differential equations. SIAM Journal on Numerical Analysis , 32(3):797–823, 1995

1995
[3]

Computation of multiphase systems with phase field models

Vittorio E Badalassi, Hector D Ceniceros, and Sanjoy Banerjee. Computation of multiphase systems with phase field models. Journal of computational physics , 190(2):371–397, 2003

2003
[4]

Physics-informed neural networks (pinns) for fluid mechanics: A review

Shengze Cai, Zhiping Mao, Zhicheng Wang, Minglang Yin, and George Em Karniadakis. Physics-informed neural networks (pinns) for fluid mechanics: A review. Acta Mechanica Sinica , 37(12):1727–1738, 2021

2021
[5]

Deep learning in computer vision: A critical review of emerging techniques and application scenarios

Junyi Chai, Hao Zeng, Anming Li, and Eric WT Ngai. Deep learning in computer vision: A critical review of emerging techniques and application scenarios. Machine Learning with Applications , 6:100134, 2021

2021
[6]

Domain decomposition algorithms

Tony F Chan and Tarek P Mathew. Domain decomposition algorithms. Acta numerica, 3:61–143, 1994

1994
[7]

Bridging traditional and machine learning-based algo- rithms for solving pdes: the random feature method

Jingrun Chen, Xurong Chi, Zhouwang Yang, et al. Bridging traditional and machine learning-based algo- rithms for solving pdes: the random feature method. J Mach Learn , 1:268–98, 2022. DISCRETE-TIME RFM FOR PDES 19 Figure 10. 1D Burgers’ equation with ∆t = 2 × 10−2: reference solution (left), predicted solution (center), and absolute error (right). Figure...

2022
[8]

Optimization of random feature method in the high-precision regime

Jingrun Chen, Weinan E, and Yifei Sun. Optimization of random feature method in the high-precision regime. Communications on Applied Mathematics and Computation , 6(2):1490–1517, 2024

2024
[9]

The random feature method for time-dependent problems

Jingrun Chen, Yixin Luo, et al. The random feature method for time-dependent problems. arXiv preprint arXiv:2304.06913, 2023

work page arXiv 2023
[10]

A micro-macro decomposition-based asymptotic-preserving random feature method for multiscale radiative transfer equations

Jingrun Chen, Zheng Ma, and Keke Wu. A micro-macro decomposition-based asymptotic-preserving random feature method for multiscale radiative transfer equations. Journal of Computational Physics , 537:114103, 2025

2025
[11]

Phase-field models for microstructure evolution

Long-Qing Chen. Phase-field models for microstructure evolution. Annual review of materials research , 32(1):113–140, 2002

2002
[12]

The random feature method for solving interface problems

Xurong Chi, Jingrun Chen, and Zhouwang Yang. The random feature method for solving interface problems. Computer Methods in Applied Mechanics and Engineering , 420:116719, 2024. DISCRETE-TIME RFM FOR PDES 20 Figure 13. 1D Cahn–Hilliard equation with ∆t = 10 −2: reference solution (left), predicted solution (center), and absolute error (right). Figure 14. 1...

2024
[13]

Recent advances and applications of deep learning methods in materials science

Kamal Choudhary, Brian DeCost, Chi Chen, Anubhav Jain, Francesca Tavazza, Ryan Cohn, Cheol Woo Park, Alok Choudhary, Ankit Agrawal, Simon JL Billinge, et al. Recent advances and applications of deep learning methods in materials science. npj Computational Materials , 8(1):59, 2022

2022
[14]

Runge–kutta random feature method for solving multiphase flow problems of cells

Yangtao Deng and Qiaolin He. Runge–kutta random feature method for solving multiphase flow problems of cells. Physics of Fluids , 37(2):023350, 2025

2025
[15]

Adaptive feature capture method for solving partial differential equations with near singular solutions, 2025

Yangtao Deng, Qiaolin He, and Xiaoping Wang. Adaptive feature capture method for solving partial differential equations with near singular solutions, 2025

2025
[16]

Local extreme learning machines and domain decomposition for solving lin- ear and nonlinear partial differential equations

Suchuan Dong and Zongwei Li. Local extreme learning machines and domain decomposition for solving lin- ear and nonlinear partial differential equations. Computer Methods in Applied Mechanics and Engineering , 387:114129, 2021

2021
[17]

Algorithms for the solution of the nonlinear least-squares problem

Philip E Gill and Walter Murray. Algorithms for the solution of the nonlinear least-squares problem. SIAM Journal on Numerical Analysis , 15(5):977–992, 1978. DISCRETE-TIME RFM FOR PDES 21 Figure 16. 2D Allen–Cahn equation with ∆t = 10 −2: reference solution (left), predicted solution (center), and absolute error (right)

1978
[18]

Quantum random feature method for solving partial differential equations, 2025

Junpeng Hu, Shi Jin, Nana Liu, and Lei Zhang. Quantum random feature method for solving partial differential equations, 2025

2025
[19]

Universal approximation using incremental construc- tive feedforward networks with random hidden nodes

Guang-Bin Huang, Lei Chen, and Chee-Kheong Siew. Universal approximation using incremental construc- tive feedforward networks with random hidden nodes. IEEE Transactions on Neural Networks , 17(4):879– 892, 2006

2006
[20]

Extended physics-informed neural networks (xpinns): A generalized space-time domain decomposition based deep learning framework for nonlinear partial differ- ential equations

Ameya D Jagtap and George Em Karniadakis. Extended physics-informed neural networks (xpinns): A generalized space-time domain decomposition based deep learning framework for nonlinear partial differ- ential equations. Communications in Computational Physics , 28(5), 2020

2020
[21]

Physics- informed machine learning

George Em Karniadakis, Ioannis G Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. Physics- informed machine learning. Nature Reviews Physics , 3(6):422–440, 2021

2021
[22]

Weak random feature method for solving partial differential equations, 2025

Mikhail Kuvakin, Zijian Mei, and Jingrun Chen. Weak random feature method for solving partial differential equations, 2025

2025
[23]

Towards a unified analysis of random fourier features

Zhu Li, Jean-Francois Ton, Dino Oglic, and Dino Sejdinovic. Towards a unified analysis of random fourier features. In International conference on machine learning , pages 3905–3914. PMLR, 2019

2019
[24]

Dae-pinn: a physics-informed neural network model for simulating dif- ferential algebraic equations with application to power networks

Christian Moya and Guang Lin. Dae-pinn: a physics-informed neural network model for simulating dif- ferential algebraic equations with application to power networks. Neural Computing and Applications , 35(5):3789–3804, 2023

2023
[25]

Reservoir computing

Kohei Nakajima and Ingo Fischer. Reservoir computing. Springer, 2021

2021
[26]

A survey of the usages of deep learning for natural language processing

Daniel W Otter, Julian R Medina, and Jugal K Kalita. A survey of the usages of deep learning for natural language processing. IEEE transactions on neural networks and learning systems , 32(2):604–624, 2020

2020
[27]

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equa- tions

Maziar Raissi, Paris Perdikaris, and George E Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equa- tions. Journal of Computational physics , 378:686–707, 2019

2019
[28]

An introduction to the finite element method, 1993

JN Reddy. An introduction to the finite element method, 1993

1993
[29]

Hierarchical deep learning neural network (hidenn): an artifi- cial intelligence (ai) framework for computational science and engineering

Sourav Saha, Zhengtao Gan, Lin Cheng, Jiaying Gao, Orion L Kafka, Xiaoyu Xie, Hengyang Li, Mahsa Tajdari, H Alicia Kim, and Wing Kam Liu. Hierarchical deep learning neural network (hidenn): an artifi- cial intelligence (ai) framework for computational science and engineering. Computer Methods in Applied Mechanics and Engineering , 373:113452, 2021

2021
[30]

Randomness in neural networks: an overview

Simone Scardapane and Dianhui Wang. Randomness in neural networks: an overview. Wiley Interdisci- plinary Reviews: Data Mining and Knowledge Discovery , 7(2):e1200, 2017

2017
[31]

Feed forward neural networks with random weights

Wouter F Schmidt, Martin A Kraaijveld, Robert PW Duin, et al. Feed forward neural networks with random weights. In International conference on pattern recognition , pages 1–1. IEEE Computer Society Press, 1992

1992
[32]

Domain decomposition methods for partial differential equations

Barry F Smith. Domain decomposition methods for partial differential equations. In Parallel numerical algorithms, pages 225–243. Springer, 1997

1997
[33]

Learning without data: Physics-informed neural networks for fast time-domain simulation

Jochen Stiasny, Samuel Chevalier, and Spyros Chatzivasileiadis. Learning without data: Physics-informed neural networks for fast time-domain simulation. In 2021 IEEE International Conference on Communi- cations, Control, and Computing Technologies for Smart Grids (SmartGridComm) , pages 438–443. IEEE, 2021

2021
[34]

Robust and eﬀicient solvers for nonlinear partial differential equations based on random feature method, 2025

Longze Tan. Robust and eﬀicient solvers for nonlinear partial differential equations based on random feature method, 2025

2025
[35]

High-precision randomized preconditioned iterative methods for the random feature method

Longze Tan and Jingrun Chen. High-precision randomized preconditioned iterative methods for the random feature method. Journal of Computational and Applied Mathematics , 481:117255, 2026. DISCRETE-TIME RFM FOR PDES 22

2026
[36]

Deep learning for computer vision: A brief review

Athanasios Voulodimos, Nikolaos Doulamis, Anastasios Doulamis, and Eftychios Protopapadakis. Deep learning for computer vision: A brief review. Computational intelligence and neuroscience, 2018(1):7068349, 2018

2018
[37]

An extreme learning machine-based method for computational pdes in higher dimensions

Yiran Wang and Suchuan Dong. An extreme learning machine-based method for computational pdes in higher dimensions. Computer Methods in Applied Mechanics and Engineering , 418:116578, 2024

2024
[38]

Pdenneval: A com- prehensive evaluation of neural network methods for solving pdes

Ping Wei, Menghan Liu, Jianhuan Cen, Ziyang Zhou, Liao Chen, and Qingsong Zou. Pdenneval: A com- prehensive evaluation of neural network methods for solving pdes. In Kate Larson, editor, Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, IJCAI-24 , pages 5181–5189. International Joint Conferences on Artificial Intel...

2024
[39]

Training neural networks on high-dimensional data using random projection

Piotr Iwo Wójcik and Marcin Kurdziel. Training neural networks on high-dimensional data using random projection. Pattern Analysis and Applications , 22:1221–1231, 2019

2019
[40]

Recent trends in deep learning based natural language processing

Tom Young, Devamanyu Hazarika, Soujanya Poria, and Erik Cambria. Recent trends in deep learning based natural language processing. IEEE Computational intelligenCe magazine , 13(3):55–75, 2018

2018
[41]

The deep ritz method: a deep learning-based numerical algorithm for solving variational problems

Bing Yu et al. The deep ritz method: a deep learning-based numerical algorithm for solving variational problems. Communications in Mathematics and Statistics , 6(1):1–12, 2018

2018
[42]

Weak adversarial networks for high-dimensional partial differential equations

Yaohua Zang, Gang Bao, Xiaojing Ye, and Haomin Zhou. Weak adversarial networks for high-dimensional partial differential equations. Journal of Computational Physics , 411:109409, 2020

2020
[43]

A new learning paradigm for random vector functional-link network: Rvfl+

Peng-Bo Zhang and Zhi-Xin Yang. A new learning paradigm for random vector functional-link network: Rvfl+. Neural Networks , 122:94–105, 2020

2020