arxiv: 2605.12580 · v1 · submitted 2026-05-12 · 💻 cs.LG

Recognition: 2 theorem links

· Lean Theorem

CAWI: Copula-Aligned Weight Initialization for Randomized Neural Networks

Mushir Akhtar , M. Tanveer , Mohd. Arshad

Authors on Pith no claims yet

Pith reviewed 2026-05-14 21:59 UTC · model grok-4.3

classification 💻 cs.LG

keywords randomized neural networkscopulaweight initializationclassificationdependence modelingmachine learningbiomedical datasets

0 comments

The pith

CAWI samples randomized neural network weights from data-fitted copulas to capture inter-feature dependence and raise accuracy.

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

Randomized neural networks fix input-to-hidden weights at random so the output layer can be solved in closed form without backpropagation. Conventional random sampling ignores correlations, asymmetries, and tail dependence among features, which hurts the conditioning of the hidden representations. CAWI maps features to the unit interval with empirical CDFs, fits a multivariate copula to their rank dependence, samples each weight column from that copula, and applies a fixed inverse marginal transform. The rest of the training procedure stays exactly the same. If correct, the method produces better predictions on classification tasks while preserving the speed and simplicity of the original paradigm.

Core claim

CAWI maps each feature to the unit interval using empirical CDFs, fits a multivariate copula (Gaussian, t, Clayton, Frank, or Gumbel) to the resulting rank correlations, samples every column of the input-to-hidden weight matrix from the fitted copula, and scales the samples by a fixed inverse marginal transform; the resulting weight matrix respects the observed dependence structure, so the unchanged closed-form output-layer solution yields higher predictive performance.

What carries the argument

Copula-aligned weight sampling: empirical-CDF marginal mapping followed by sampling from a fitted multivariate copula (elliptical or Archimedean) and inverse-marginal scaling to produce dependence-aware input-to-hidden weights.

If this is right

Predictive performance improves consistently over conventional random initialization on 83 diverse classification benchmarks and two biomedical datasets.
Both shallow and deep randomized neural network architectures benefit without any change to the objective or solver.
Elliptical and Archimedean copula families allow the method to capture symmetric, asymmetric, and tail dependence.
The closed-form, backpropagation-free training property of randomized neural networks remains fully intact.

Where Pith is reading between the lines

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

The same dependence-aware sampling could be tested on regression tasks that also rely on random projections.
In high-dimensional settings the copula fit itself might become a bottleneck, suggesting a need for sparse or factor copulas.
Incremental or online updating of the copula parameters would let the initialization adapt when new data arrive after initial training.

Load-bearing premise

That a copula fitted to training-feature dependence will produce hidden-layer projections whose conditioning improves the closed-form output solution on unseen data.

What would settle it

On the same 83 classification benchmarks and two biomedical datasets, using identical architectures and solvers, CAWI produces no average accuracy gain (or produces lower accuracy) relative to standard Gaussian or uniform random initialization.

Figures

Figures reproduced from arXiv: 2605.12580 by Mohd. Arshad, M. Tanveer, Mushir Akhtar.

**Figure 1.** Figure 1: Schematic of a copula. Left: empirical sample (with its joint CDF inset). Middle: univariate CDFs F1 and F2 (mapping xj 7→ uj = Fj (xj )). Right: the copula surface C(u1, u2), i.e., the joint CDF on [0, 1]2 with uniform marginals. The equality sign indicates Sklar’s factorization F(x1, x2) = C(F1(x1), F2(x2)); the ⊕ highlights that C combines the marginal information into a dependence structure. Biases b m… view at source ↗

read the original abstract

Randomized neural networks (RdNNs) enable efficient, backpropagation-free training by freezing randomly initialized input-to-hidden weights, which permits a closed-form solution for the output layer. However, conventional random initialization is blind to inter-feature dependence, ignoring correlations, asymmetries, and tail dependence in the data, which degrades conditioning and predictive performance. To the best of our knowledge, this limitation remains unaddressed in the RdNN literature. To close this gap, we propose CAWI (Copula-Aligned Weight Initialization), a framework that draws input-to-hidden weights from a data-fitted copula that matches empirical dependence, ensuring the frozen projections respect inter-feature dependence without sacrificing the closed-form solution. CAWI (i) maps each feature to the unit interval using empirical CDFs, (ii) fits a multivariate copula that captures rank-based dependence among features, and (iii) samples each weight column w_j from the fitted copula and applies a fixed inverse marginal transform to set scale. The objective, solver, and "freeze-once" paradigm remain unchanged; only the sampling law for W becomes dependence-aware. For dependence modeling, we consider two copula families: elliptical (Gaussian, t) and Archimedean (Clayton, Frank, Gumbel). This enables CAWI to handle diverse dependence, including tail dependence. We evaluate CAWI across 83 diverse classification benchmarks (binary and multiclass) and two biomedical datasets, BreaKHis and the Schizophrenia dataset, using standard shallow and deep RdNN architectures. CAWI consistently delivers significant improvements in predictive performance over conventional random initialization. Code is available at: https://github.com/mtanveer1/CAWI

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

CAWI swaps standard random weights for copula-sampled ones in RdNNs to match feature dependence and reports accuracy gains on 83 benchmarks, but supplies no conditioning or stability diagnostics to back the mechanism.

read the letter

CAWI takes the usual random draw for the input-to-hidden matrix in randomized neural networks and replaces it with columns sampled from a copula fitted to the empirical rank correlations of the input features. The closed-form output solve stays untouched. The paper tests Gaussian, t, Clayton, Frank, and Gumbel copulas, applies the usual empirical-CDF and inverse-transform steps, and shows higher accuracy than plain random initialization on 83 classification sets plus two medical ones, using both shallow and deeper RdNN variants. Code is public, which makes the claim easy to check on new data. That is the concrete advance: a dependence-aware initialization that still respects the freeze-and-solve paradigm. The evaluation breadth is the part that stands up. The gains hold across binary and multiclass tasks and across different copula families, so the result is not tied to one narrow choice. For anyone already running RdNNs in low-resource settings, this is a drop-in option worth trying. The soft spot is the missing link between the method and the claimed benefit. The authors say that ignoring dependence degrades conditioning of the hidden matrix, yet the manuscript gives only final accuracy numbers. There are no condition-number tables, eigenvalue spectra, or solver-residual comparisons between CAWI and baseline. It remains possible that injecting the same dependence structure into the weight columns raises the condition number of H or HᵀH in some regimes, and the accuracy lift comes from another source. Without those diagnostics the central explanation stays untested. This paper is for people who already work with randomized networks or who want a lightweight way to bring copulas into initialization. It is narrow in scope but the experimental footprint is wide enough to be useful. I would send it to peer review. The public code and the scale of the benchmarks give a referee something concrete to examine, even if the conditioning story needs more support in revision.

Referee Report

2 major / 2 minor

Summary. The paper proposes CAWI, a weight initialization scheme for randomized neural networks that fits a multivariate copula (Gaussian, t, Clayton, Frank or Gumbel) to the rank correlations of the training features, samples each input-to-hidden weight column from the fitted copula, and applies an inverse marginal transform; the output-layer weights are still obtained in closed form. The central empirical claim is that this dependence-aware sampling yields consistent accuracy gains over i.i.d. random initialization on 83 classification benchmarks plus two biomedical datasets while leaving the training paradigm unchanged.

Significance. If the performance gains prove robust and the mechanism is confirmed, CAWI would supply a lightweight, data-dependent initialization that respects feature dependence without sacrificing the computational advantages of RdNNs. The public code release strengthens reproducibility.

major comments (2)

[Experimental evaluation] Experimental results: the manuscript reports accuracy improvements but supplies no condition-number statistics, eigenvalue spectra of H or HᵀH, or solver-residual diagnostics comparing CAWI to baseline i.i.d. initialization. Because the motivating claim is that copula sampling improves conditioning of the closed-form solve, the absence of these diagnostics leaves the hypothesized mechanism unverified.
[CAWI framework] Methodology: the precise procedure for choosing among the five copula families and for estimating their parameters (including any model-selection criterion) is not stated explicitly; without this information it is impossible to determine whether the reported gains are driven by the copula construction itself or by an implicit hyper-parameter search.

minor comments (2)

[Abstract] The abstract states that 83 benchmarks were used but does not indicate their sources, dimensionality range, or class-balance characteristics; a brief summary table would improve clarity.
[CAWI framework] Notation for the inverse marginal transform (step (iii) of the algorithm) is introduced without an explicit equation; adding a numbered equation would aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the presentation of our work. We address each major comment below and will revise the manuscript accordingly to strengthen the empirical verification and methodological clarity.

read point-by-point responses

Referee: [Experimental evaluation] Experimental results: the manuscript reports accuracy improvements but supplies no condition-number statistics, eigenvalue spectra of H or HᵀH, or solver-residual diagnostics comparing CAWI to baseline i.i.d. initialization. Because the motivating claim is that copula sampling improves conditioning of the closed-form solve, the absence of these diagnostics leaves the hypothesized mechanism unverified.

Authors: We agree that these diagnostics are necessary to verify the conditioning hypothesis. In the revised manuscript we will add condition-number statistics, eigenvalue spectra of H and HᵀH, and solver-residual norms for both CAWI and the i.i.d. baseline on representative datasets from the benchmark suite. These additions will directly confirm whether the dependence-aware sampling improves the numerical properties of the closed-form solve. revision: yes
Referee: [CAWI framework] Methodology: the precise procedure for choosing among the five copula families and for estimating their parameters (including any model-selection criterion) is not stated explicitly; without this information it is impossible to determine whether the reported gains are driven by the copula construction itself or by an implicit hyper-parameter search.

Authors: We thank the referee for highlighting this omission. The full manuscript evaluates all five families (Gaussian, Student-t, Clayton, Frank, Gumbel) and selects, for each dataset, the family that maximizes the log-likelihood of the fitted copula on the training features; parameters are obtained via standard maximum-likelihood estimation for the chosen family. We will explicitly document this selection rule and estimation procedure in the revised methodology section, making clear that the only data-driven choice is the copula family itself and that no additional hyper-parameter tuning is performed. revision: yes

Circularity Check

0 steps flagged

No significant circularity; CAWI is an explicit data-dependent sampling construction with empirical validation.

full rationale

The paper defines CAWI as a three-step procedure: (i) map features via empirical CDFs, (ii) fit a multivariate copula to rank correlations, (iii) sample weight columns from the fitted copula and apply inverse marginal transform. This is a direct algorithmic construction, not a derivation in which any claimed result (e.g., improved conditioning or accuracy) is forced by the paper's own equations to equal its inputs. No load-bearing self-citations, uniqueness theorems, or ansatzes imported from prior work appear. Performance gains are reported via cross-benchmark experiments rather than by-construction predictions. The method therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that a fitted copula accurately represents the dependence structure that matters for the linear projection performed by the frozen weights. No new physical entities are introduced. The copula parameters themselves are free parameters estimated from data.

free parameters (1)

copula parameters (theta for each family)
Fitted to the rank-transformed training features for each dataset; the specific values are data-dependent and chosen to maximize the copula likelihood.

axioms (2)

standard math Empirical CDF transform produces uniform marginals suitable for copula modeling
Standard property of probability integral transform invoked when mapping features to the unit interval.
domain assumption Multivariate copula can be sampled to produce weight vectors whose joint distribution matches observed feature dependence
Core modeling assumption stated in the three-step procedure.

pith-pipeline@v0.9.0 · 5613 in / 1502 out tokens · 22986 ms · 2026-05-14T21:59:54.957462+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We consider two copula families: elliptical (Gaussian, t) and Archimedean (Clayton, Frank, Gumbel)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

105 extracted references · 105 canonical work pages · 3 internal anchors

[1]

Langley , title =

P. Langley , title =. Proceedings of the 17th International Conference on Machine Learning (ICML 2000) , address =. 2000 , pages =

work page 2000
[2]

T. M. Mitchell. The Need for Biases in Learning Generalizations. 1980

work page 1980
[3]

M. J. Kearns , title =

work page
[4]

Machine Learning: An Artificial Intelligence Approach, Vol. I. 1983

work page 1983
[5]

R. O. Duda and P. E. Hart and D. G. Stork. Pattern Classification. 2000

work page 2000
[6]

Suppressed for Anonymity , author=

work page
[7]

Newell and P

A. Newell and P. S. Rosenbloom. Mechanisms of Skill Acquisition and the Law of Practice. Cognitive Skills and Their Acquisition. 1981

work page 1981
[8]

A. L. Samuel. Some Studies in Machine Learning Using the Game of Checkers. IBM Journal of Research and Development. 1959

work page 1959
[9]

2009 IEEE Control Applications,(CCA) & Intelligent Control,(ISIC) , pages=

Feasibility of random basis function approximators for modeling and control , author=. 2009 IEEE Control Applications,(CCA) & Intelligent Control,(ISIC) , pages=. 2009 , organization=

work page 2009
[10]

Information Sciences , volume=

Insights into randomized algorithms for neural networks: Practical issues and common pitfalls , author=. Information Sciences , volume=. 2017 , publisher=

work page 2017
[11]

IEEE transactions on cybernetics , volume=

Stochastic configuration networks: Fundamentals and algorithms , author=. IEEE transactions on cybernetics , volume=. 2017 , publisher=

work page 2017
[12]

Information Sciences , volume=

Voting based extreme learning machine , author=. Information Sciences , volume=. 2012 , publisher=

work page 2012
[13]

Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery , volume=

Randomness in neural networks: an overview , author=. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery , volume=. 2017 , publisher=

work page 2017
[14]

Computational Intelligence and Neuroscience , volume=

Deep learning for computer vision: A brief review , author=. Computational Intelligence and Neuroscience , volume=. 2018 , publisher=

work page 2018
[15]

and Medina, Julian R

Otter, Daniel W. and Medina, Julian R. and Kalita, Jugal K. , journal=. A Survey of the Usages of Deep Learning for Natural Language Processing , year=

work page
[16]

Nature medicine , volume=

A guide to deep learning in healthcare , author=. Nature medicine , volume=. 2019 , publisher=

work page 2019
[17]

Nature , volume=

Deep learning , author=. Nature , volume=. 2015 , publisher=

work page 2015
[18]

Information Sciences , volume=

A survey of randomized algorithms for training neural networks , author=. Information Sciences , volume=. 2016 , publisher=

work page 2016
[19]

Neurocomputing , volume=

A review on neural networks with random weights , author=. Neurocomputing , volume=. 2018 , publisher=

work page 2018
[20]

Neurocomputing , volume=

Extreme learning machine: theory and applications , author=. Neurocomputing , volume=. 2006 , publisher=

work page 2006
[21]

Multimedia Tools and Applications , volume=

A review on extreme learning machine , author=. Multimedia Tools and Applications , volume=. 2022 , publisher=

work page 2022
[22]

Neurocomputing , volume=

Learning and generalization characteristics of the random vector functional-link net , author=. Neurocomputing , volume=. 1994 , publisher=

work page 1994
[23]

Applied Soft Computing , volume=

Random vector functional link network: recent developments, applications, and future directions , author=. Applied Soft Computing , volume=. 2023 , publisher=

work page 2023
[24]

Information sciences , volume=

A comprehensive evaluation of random vector functional link networks , author=. Information sciences , volume=. 2016 , publisher=

work page 2016
[25]

IEEE Transactions on Neural Networks , volume=

Stochastic choice of basis functions in adaptive function approximation and the functional-link net , author=. IEEE Transactions on Neural Networks , volume=. 1995 , publisher=

work page 1995
[26]

IEEE transactions on neural networks and learning systems , volume=

Broad learning system: An effective and efficient incremental learning system without the need for deep architecture , author=. IEEE transactions on neural networks and learning systems , volume=. 2017 , publisher=

work page 2017
[27]

2017 32nd youth academic annual conference of Chinese association of automation (YAC) , pages=

Broad learning system: A new learning paradigm and system without going deep , author=. 2017 32nd youth academic annual conference of Chinese association of automation (YAC) , pages=. 2017 , organization=

work page 2017
[28]

IEEE transactions on neural networks and learning systems , volume=

Universal approximation capability of broad learning system and its structural variations , author=. IEEE transactions on neural networks and learning systems , volume=. 2018 , publisher=

work page 2018
[29]

IEEE Transactions on Industrial Electronics , volume=

Broad convolutional neural network based industrial process fault diagnosis with incremental learning capability , author=. IEEE Transactions on Industrial Electronics , volume=. 2019 , publisher=

work page 2019
[30]

IEEE Transactions on Cybernetics , volume=

Research review for broad learning system: Algorithms, theory, and applications , author=. IEEE Transactions on Cybernetics , volume=. 2021 , publisher=

work page 2021
[31]

Neurocomputing , volume=

Regularized robust broad learning system for uncertain data modeling , author=. Neurocomputing , volume=. 2018 , publisher=

work page 2018
[32]

IEEE transactions on cybernetics , volume=

Fuzzy broad learning system: A novel neuro-fuzzy model for regression and classification , author=. IEEE transactions on cybernetics , volume=. 2018 , publisher=

work page 2018
[33]

IEEE Transactions on Fuzzy Systems , year=

Intuitionistic fuzzy broad learning system: Enhancing robustness against noise and outliers , author=. IEEE Transactions on Fuzzy Systems , year=

work page
[34]

Annals of Data Science , pages=

A comprehensive survey of loss functions in machine learning , author=. Annals of Data Science , pages=. 2020 , publisher=

work page 2020
[35]

Neurocomputing , volume=

Recent advances on loss functions in deep learning for computer vision , author=. Neurocomputing , volume=. 2022 , publisher=

work page 2022
[36]

Pattern Recognition , pages=

Advancing supervised learning with the wave loss function: A robust and smooth approach , author=. Pattern Recognition , pages=. 2024 , publisher=

work page 2024
[37]

Advances in Neural Information Processing Systems , volume=

This looks like that: deep learning for interpretable image recognition , author=. Advances in Neural Information Processing Systems , volume=

work page
[38]

Neurocomputing , volume=

An introduction to deep learning in natural language processing: Models, techniques, and tools , author=. Neurocomputing , volume=. 2022 , publisher=

work page 2022
[39]

2016 , publisher=

Deep learning , author=. 2016 , publisher=

work page 2016
[40]

Proceedings of the 30th International Conference on Machine Learning , pages =

On the difficulty of training recurrent neural networks , author =. Proceedings of the 30th International Conference on Machine Learning , pages =. 2013 , volume =

work page 2013
[41]

Proceedings of the thirteenth International Conference on Artificial Intelligence and Statistics , pages=

Understanding the difficulty of training deep feedforward neural networks , author=. Proceedings of the thirteenth International Conference on Artificial Intelligence and Statistics , pages=. 2010 , organization=

work page 2010
[42]

Exact solutions to the nonlinear dynamics of learning in deep linear neural networks

Exact solutions to the nonlinear dynamics of learning in deep linear neural networks , author=. arXiv preprint arXiv:1312.6120 , year=

work page internal anchor Pith review Pith/arXiv arXiv
[43]

Spectral Normalization for Generative Adversarial Networks

Spectral normalization for generative adversarial networks , author=. arXiv preprint arXiv:1802.05957 , year=

work page internal anchor Pith review Pith/arXiv arXiv
[44]

International journal of machine learning and cybernetics , volume=

Extreme learning machines: a survey , author=. International journal of machine learning and cybernetics , volume=. 2011 , publisher=

work page 2011
[45]

Acta Numerica , volume=

Random matrix theory , author=. Acta Numerica , volume=. 2005 , publisher=

work page 2005
[46]

American Mathematical Society , author=

Topics in random matrix theory. American Mathematical Society , author=. Graduate Studies in Mathematics , volume=. 2012 , url=

work page 2012
[47]

2004 , publisher=

Random matrices , author=. 2004 , publisher=

work page 2004
[48]

2018 , publisher=

High-dimensional probability: An introduction with applications in data science , author=. 2018 , publisher=

work page 2018
[49]

2010 , publisher=

Spectral analysis of large dimensional random matrices , author=. 2010 , publisher=

work page 2010
[50]

2010 , publisher=

An introduction to random matrices , author=. 2010 , publisher=

work page 2010
[51]

Physics Reports , volume=

Random-matrix theories in quantum physics: common concepts , author=. Physics Reports , volume=. 1998 , publisher=

work page 1998
[52]

Journal of Statistical Planning and Inference , volume=

Random matrix theory in statistics: A review , author=. Journal of Statistical Planning and Inference , volume=. 2014 , publisher=

work page 2014
[53]

Foundations and Trends

Random matrix theory and wireless communications , author=. Foundations and Trends. 2004 , publisher=

work page 2004
[54]

Language-Driven Cross-Modal Classifier for Zero-Shot Multi-Label Image Recognition , author=

work page
[55]

Proceedings of the 41st International Conference on Machine Learning , pages =

Language-Driven Cross-Modal Classifier for Zero-Shot Multi-Label Image Recognition , author =. Proceedings of the 41st International Conference on Machine Learning , pages =. 2024 , volume =

work page 2024
[56]

Proceedings of the 41st International Conference on Machine Learning , pages =

Characterizing Truthfulness in Large Language Model Generations with Local Intrinsic Dimension , author =. Proceedings of the 41st International Conference on Machine Learning , pages =. 2024 , volume =

work page 2024
[57]

Old can be Gold: Better Gradient Flow can Make Vanilla-GCNs Great Again , volume =

Jaiswal, Ajay and Wang, Peihao and Chen, Tianlong and Rousseau, Justin and Ding, Ying and Wang, Zhangyang , booktitle =. Old can be Gold: Better Gradient Flow can Make Vanilla-GCNs Great Again , volume =

work page
[58]

Applied Soft Computing , volume=

On the origins of randomization-based feedforward neural networks , author=. Applied Soft Computing , volume=. 2021 , publisher=

work page 2021
[59]

Self-Distillation for Randomized Neural Networks , year=

Hu, Minghui and Gao, Ruobin and Suganthan, Ponnuthurai Nagaratnam , journal=. Self-Distillation for Randomized Neural Networks , year=

work page
[60]

Pattern Recognition , volume=

Random vector functional link neural network based ensemble deep learning , author=. Pattern Recognition , volume=. 2021 , publisher=

work page 2021
[61]

An Ensemble Broad Learning Scheme for Semisupervised Vehicle Type Classification , year=

Guo, Li and Li, Runze and Jiang, Bin , journal=. An Ensemble Broad Learning Scheme for Semisupervised Vehicle Type Classification , year=

work page
[62]

2025 , publisher=

Sajid, M and Quadir, A and Tanveer, M and for the Alzheimer’s Disease Neuroimaging Initiative , journal=. 2025 , publisher=

work page 2025
[63]

and Malik, A

Sajid, M. and Malik, A. K. and Tanveer, M. and Suganthan, Ponnuthurai N. , journal=. Neuro-Fuzzy Random Vector Functional Link Neural Network for Classification and Regression Problems , year=

work page
[64]

and Tanveer, M

Sajid, M. and Tanveer, M. and Suganthan, Ponnuthurai N. , journal=. Ensemble Deep Random Vector Functional Link Neural Network Based on Fuzzy Inference System , year=

work page
[65]

Neural Networks , volume=

An unsupervised parameter learning model for RVFL neural network , author=. Neural Networks , volume=. 2019 , publisher=

work page 2019
[66]

IEEE Transactions on Neural Networks and Learning Systems , volume=

Broad learning system: An effective and efficient incremental learning system without the need for deep architecture , author=. IEEE Transactions on Neural Networks and Learning Systems , volume=. 2017 , publisher=

work page 2017
[67]

Advances in Neural Information Processing Systems , volume=

Nonlinear random matrix theory for deep learning , author=. Advances in Neural Information Processing Systems , volume=

work page
[68]

and Malik, A

Sajid, M. and Malik, A. K. and Tanveer, M. , journal=. Intuitionistic Fuzzy Broad Learning System: Enhancing Robustness Against Noise and Outliers , year=

work page
[69]

2017 , publisher=

Dua, Dheeru and Graff, Casey , journal=. 2017 , publisher=

work page 2017
[70]

Derrac, J and Garcia, S and Sanchez, L and Herrera, F , journal=. K

work page
[71]

Random matrix theory analysis of neural network weight matrices , author=

work page
[72]

Communications in Statistics-Theory and Methods , volume=

Approximations of the critical region of the fbietkan statistic , author=. Communications in Statistics-Theory and Methods , volume=. 1980 , publisher=

work page 1980
[73]

The Journal of Machine Learning Research , volume=

Statistical comparisons of classifiers over multiple data sets , author=. The Journal of Machine Learning Research , volume=. 2006 , publisher=

work page 2006
[74]

Proceedings of the AAAI Conference on Artificial Intelligence , volume=

Learning deep hierarchical features with spatial regularization for one-class facial expression recognition , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

work page
[75]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

RCV2023 challenges: Benchmarking model training and inference for resource-constrained deep learning , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

work page
[76]

IEEE Transactions on Neural Networks and Learning Systems , year=

Random orthogonal additive filters: a solution to the vanishing/exploding gradient of deep neural networks , author=. IEEE Transactions on Neural Networks and Learning Systems , year=

work page
[77]

Frontiers in Applied Mathematics and Statistics , volume=

Random vector functional link networks for function approximation on manifolds , author=. Frontiers in Applied Mathematics and Statistics , volume=. 2024 , publisher=

work page 2024
[78]

Advances in Neural Information Processing Systems , volume=

High-dimensional asymptotics of feature learning: How one gradient step improves the representation , author=. Advances in Neural Information Processing Systems , volume=

work page
[79]

2013 , publisher=

Matrix computations , author=. 2013 , publisher=

work page 2013
[80]

International Conference on Machine Learning , pages=

Stabilizing gradients for deep neural networks via efficient svd parameterization , author=. International Conference on Machine Learning , pages=. 2018 , organization=

work page 2018

Showing first 80 references.