Anderson Melchor Hernandez
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Anderson Melchor Hernandez has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Davide Pastorello
University of Bologna
4 shared publications - Giacomo De Palma
Department of Mathematics, University of Bologna, Bologna, Italy
4 shared publications - Valeria Chiadò Piat
Dip. di Scienze Matematiche "G. Lagrange", Politecnico di Torino, Torino, Italy
3 shared publications - Virginia De Cicco
Department of Basic and Applied Sciences for Engineering, Sapienza University of Rome, Via A. Scarpa 10, 00185 Roma, Italy
3 shared publications - Filippo Girardi
Scuola Normale Superiore Piazza dei Cavalieri 7 56126 Pisa Italy
2 shared publications - Diógenes de Jesus Ramirez-Ramirez
Departamento de Matemáticas y Estadística, Universidad Nacional de Colombia, Sede Manizales, Kilómetro 7 Vía al Aeropuerto, Manizales 170003, Colombia
1 shared publication
Latest Publications
- Open Access
- 27 March 2026
- Published by Springer Nature in Letters in Mathematical Physics
- Vol. 116 (2) , 42
Mean-field limit from general mixtures of experts to quantum neural networks
In this work, we study the asymptotic behavior of mixture of experts (MoE) trained via gradient flow on supervised learning problems. Our main result establishes the propagation of chaos for a MoE as the number of experts diverges. We demonstrate that the corresponding empirical measure of their parameters is close to a probability measure that solves a nonlinear continuity equation, and we provide an explicit convergence rate that depends solely on the number of experts and on the dimensionality of the parameter space. We apply our results to a MoE generated by a quantum neural network.
- 27 March 2026
- Published by Springer Nature in Letters in Mathematical Physics
- Vol. 116 (2) , 42
Open AccessMean-field limit from general mixtures of experts to quantum neural networks
In this work, we study the asymptotic behavior of mixture of experts (MoE) trained via gradient flow on supervised learning problems. Our main result establishes the propagation of chaos for a MoE as the number of experts diverges. We demonstrate that the corresponding empirical measure of their parameters is close to a probability measure that solves a nonlinear continuity equation, and we provide an explicit convergence rate that depends solely on the number of experts and on the dimensionality of the parameter space. We apply our results to a MoE generated by a quantum neural network.
- Open Access
A Modified Bayesian Criterion for Model Selection in Mixed and Hierarchical Frameworks
In this work, we propose a modified Bayesian Information Criterion (BIC) specifically designed for mixture models and hierarchical structures. This criterion incorporates the determinant of the Hessian matrix of the log-likelihood function, thereby refining the classical Bayes Factor by accounting for the curvature of the likelihood surface. Such geometric information introduces a more nuanced penalization for model complexity. The proposed approach improves model selection, particularly under small-sample conditions or in the presence of noise variables. Through theoretical derivations and extensive simulation studies-including both linear and linear mixed models-we show that our criterion consistently outperforms traditional methods such as BIC, Akaike Information Criterion (AIC), and related variants. The results suggest that integrating curvature-based information from the likelihood landscape leads to more robust and accurate model discrimination in complex data environments.
Open AccessA Modified Bayesian Criterion for Model Selection in Mixed and Hierarchical Frameworks
In this work, we propose a modified Bayesian Information Criterion (BIC) specifically designed for mixture models and hierarchical structures. This criterion incorporates the determinant of the Hessian matrix of the log-likelihood function, thereby refining the classical Bayes Factor by accounting for the curvature of the likelihood surface. Such geometric information introduces a more nuanced penalization for model complexity. The proposed approach improves model selection, particularly under small-sample conditions or in the presence of noise variables. Through theoretical derivations and extensive simulation studies-including both linear and linear mixed models-we show that our criterion consistently outperforms traditional methods such as BIC, Akaike Information Criterion (AIC), and related variants. The results suggest that integrating curvature-based information from the likelihood landscape leads to more robust and accurate model discrimination in complex data environments.
- 1 January 2026
- Published by Banach Press LLC in Journal of Convex Analysis
- Vol. 33 (1&2) , 335-360
Relaxation for a Degenerate Functional with Linear Growth in the Onedimensional Case
We study the relaxation of a degenerate functional with linear growth, depending on a weight w that does not exhibit doubling or Muckenhoupt-type conditions. In order to obtain an explicit representation of the relaxed functional and its domain, our main tools for are Sobolev inequalities with double weight.
- 1 January 2026
- Published by Banach Press LLC in Journal of Convex Analysis
- Vol. 33 (1&2) , 335-360
Relaxation for a Degenerate Functional with Linear Growth in the Onedimensional Case
We study the relaxation of a degenerate functional with linear growth, depending on a weight w that does not exhibit doubling or Muckenhoupt-type conditions. In order to obtain an explicit representation of the relaxed functional and its domain, our main tools for are Sobolev inequalities with double weight.
- Open Access
- 8 October 2025
- Published by Springer Nature in Annales Henri Poincare
- Vol. 27 (11) , 4011-4067
Quantitative Convergence of Trained Quantum Neural Networks to a Gaussian Process
We study quantum neural networks where the generated function is the expectation value of the sum of single-qubit observables across all qubits. In (Girardi et al., CMP (2025), it is proven that the probability distributions of such generated functions converge in distribution to a Gaussian process in the limit of infinite width for both untrained networks with randomly initialized parameters and trained networks. In this paper, we provide a quantitative proof of this convergence in terms of the Wasserstein distance of order 1. First, we establish an upper bound on the distance between the probability distribution of the function generated by any untrained network with finite width and the Gaussian process with the same covariance. This proof utilizes Stein’s method to estimate the Wasserstein distance of order 1. Next, we analyze the training dynamics of the network via gradient flow, proving an upper bound on the distance between the probability distribution of the function generated by the trained network and the corresponding Gaussian process. This proof is based on a quantitative upper bound on the maximum variation of a parameter during training. This bound implies that for sufficiently large widths, training occurs in the lazy regime, i.e., each parameter changes only by a small amount. While the convergence result of Girardi et al., CMP (2025) holds at a fixed training time, our upper bounds are uniform in time and hold even as $$t \rightarrow \infty $$ .- 8 October 2025
- Published by Springer Nature in Annales Henri Poincare
- Vol. 27 (11) , 4011-4067
Open AccessQuantitative Convergence of Trained Quantum Neural Networks to a Gaussian Process
We study quantum neural networks where the generated function is the expectation value of the sum of single-qubit observables across all qubits. In (Girardi et al., CMP (2025), it is proven that the probability distributions of such generated functions converge in distribution to a Gaussian process in the limit of infinite width for both untrained networks with randomly initialized parameters and trained networks. In this paper, we provide a quantitative proof of this convergence in terms of the Wasserstein distance of order 1. First, we establish an upper bound on the distance between the probability distribution of the function generated by any untrained network with finite width and the Gaussian process with the same covariance. This proof utilizes Stein’s method to estimate the Wasserstein distance of order 1. Next, we analyze the training dynamics of the network via gradient flow, proving an upper bound on the distance between the probability distribution of the function generated by the trained network and the corresponding Gaussian process. This proof is based on a quantitative upper bound on the maximum variation of a parameter during training. This bound implies that for sufficiently large widths, training occurs in the lazy regime, i.e., each parameter changes only by a small amount. While the convergence result of Girardi et al., CMP (2025) holds at a fixed training time, our upper bounds are uniform in time and hold even as $$t \rightarrow \infty $$ . - Open Access
Efficient classical computation of the neural tangent kernel of quantum neural networks
We propose an efficient classical algorithm to estimate the Neural Tangent Kernel (NTK) associated with a broad class of quantum neural networks. These networks consist of arbitrary unitary operators belonging to the Clifford group interleaved with parametric gates given by the time evolution generated by an arbitrary Hamiltonian belonging to the Pauli group. The proposed algorithm leverages a key insight: the average over the distribution of initialization parameters in the NTK definition can be exactly replaced by an average over just four discrete values, chosen such that the corresponding parametric gates are Clifford operations. This reduction enables an efficient classical simulation of the circuit. Combined with recent results establishing the equivalence between wide quantum neural networks and Gaussian processes [Girardi \emph{et al.}, Comm. Math. Phys. 406, 92 (2025); Melchor Hernandez \emph{et al.}, arXiv:2412.03182], our method enables efficient computation of the expected output of wide, trained quantum neural networks, and therefore shows that such networks cannot achieve quantum advantage.
Open AccessEfficient classical computation of the neural tangent kernel of quantum neural networks
We propose an efficient classical algorithm to estimate the Neural Tangent Kernel (NTK) associated with a broad class of quantum neural networks. These networks consist of arbitrary unitary operators belonging to the Clifford group interleaved with parametric gates given by the time evolution generated by an arbitrary Hamiltonian belonging to the Pauli group. The proposed algorithm leverages a key insight: the average over the distribution of initialization parameters in the NTK definition can be exactly replaced by an average over just four discrete values, chosen such that the corresponding parametric gates are Clifford operations. This reduction enables an efficient classical simulation of the circuit. Combined with recent results establishing the equivalence between wide quantum neural networks and Gaussian processes [Girardi \emph{et al.}, Comm. Math. Phys. 406, 92 (2025); Melchor Hernandez \emph{et al.}, arXiv:2412.03182], our method enables efficient computation of the expected output of wide, trained quantum neural networks, and therefore shows that such networks cannot achieve quantum advantage.
- Open Access
Minimization of Degenerate Nonlinear Functionals under Radial Symmetry
In this work, we study the minimization of nonlinear functionals in dimension $d\geq 1$ that depend on a degenerate radial weight $w$. Our goal is to prove the existence of minimizers in a suitable functional class here introduced and to establish that the minimizers of such functionals, which exhibit $p$-growth with $1 < p < +\infty$, are radially symmetric. In our analysis, we adopt the approach developed in [Chiadò Piat, De Cicco and Melchor Hernandez, NoDEA $2025$, De Cicco and Serra Cassano, ESAIM:COCV $2024$], where $w$ does not satisfy classical assumptions such as doubling or Muckenhoupt conditions. The core of our method relies on proving the validity of a weighted Poincaré inequality involving a suitably constructed auxiliary weight.Open AccessMinimization of Degenerate Nonlinear Functionals under Radial Symmetry
In this work, we study the minimization of nonlinear functionals in dimension $d\geq 1$ that depend on a degenerate radial weight $w$. Our goal is to prove the existence of minimizers in a suitable functional class here introduced and to establish that the minimizers of such functionals, which exhibit $p$-growth with $1 < p < +\infty$, are radially symmetric. In our analysis, we adopt the approach developed in [Chiadò Piat, De Cicco and Melchor Hernandez, NoDEA $2025$, De Cicco and Serra Cassano, ESAIM:COCV $2024$], where $w$ does not satisfy classical assumptions such as doubling or Muckenhoupt conditions. The core of our method relies on proving the validity of a weighted Poincaré inequality involving a suitably constructed auxiliary weight. - Open Access
- 12 May 2025
- Published by Springer Nature in Nonlinear Differential Equations and Applications NoDEA
- Vol. 32 (4) , 1-19
Relaxation for degenerate nonlinear functionals in the one-dimensional case
In this study, we approach the analysis of a degenerate nonlinear functional in one dimension, accommodating a degenerate weight w. Our investigation focuses on establishing an explicit relaxation formula for a functional exhibiting p-growth for $$1< p<+\infty $$ . We adopt the approach developed in [6], where some assumptions like doubling or Muckenhoupt conditions are dropped. Our main tools consist of proving the validity of a weighted Poincaré inequality involving an auxiliary weight.- 12 May 2025
- Published by Springer Nature in Nonlinear Differential Equations and Applications NoDEA
- Vol. 32 (4) , 1-19
Open AccessRelaxation for degenerate nonlinear functionals in the one-dimensional case
In this study, we approach the analysis of a degenerate nonlinear functional in one dimension, accommodating a degenerate weight w. Our investigation focuses on establishing an explicit relaxation formula for a functional exhibiting p-growth for $$1< p<+\infty $$ . We adopt the approach developed in [6], where some assumptions like doubling or Muckenhoupt conditions are dropped. Our main tools consist of proving the validity of a weighted Poincaré inequality involving an auxiliary weight. - Open Access
Quantitative convergence of trained quantum neural networks to a Gaussian process
We study quantum neural networks where the generated function is the expectation value of the sum of single-qubit observables across all qubits. In [Girardi \emph{et al.}, arXiv:2402.08726], it is proven that the probability distributions of such generated functions converge in distribution to a Gaussian process in the limit of infinite width for both untrained networks with randomly initialized parameters and trained networks. In this paper, we provide a quantitative proof of this convergence in terms of the Wasserstein distance of order $1$. First, we establish an upper bound on the distance between the probability distribution of the function generated by any untrained network with finite width and the Gaussian process with the same covariance. This proof utilizes Stein's method to estimate the Wasserstein distance of order $1$. Next, we analyze the training dynamics of the network via gradient flow, proving an upper bound on the distance between the probability distribution of the function generated by the trained network and the corresponding Gaussian process. This proof is based on a quantitative upper bound on the maximum variation of a parameter during training. This bound implies that for sufficiently large widths, training occurs in the lazy regime, \emph{i.e.}, each parameter changes only by a small amount. While the convergence result of [Girardi \emph{et al.}, arXiv:2402.08726] holds at a fixed training time, our upper bounds are uniform in time and hold even as $t \to \infty$.Open AccessQuantitative convergence of trained quantum neural networks to a Gaussian process
We study quantum neural networks where the generated function is the expectation value of the sum of single-qubit observables across all qubits. In [Girardi \emph{et al.}, arXiv:2402.08726], it is proven that the probability distributions of such generated functions converge in distribution to a Gaussian process in the limit of infinite width for both untrained networks with randomly initialized parameters and trained networks. In this paper, we provide a quantitative proof of this convergence in terms of the Wasserstein distance of order $1$. First, we establish an upper bound on the distance between the probability distribution of the function generated by any untrained network with finite width and the Gaussian process with the same covariance. This proof utilizes Stein's method to estimate the Wasserstein distance of order $1$. Next, we analyze the training dynamics of the network via gradient flow, proving an upper bound on the distance between the probability distribution of the function generated by the trained network and the corresponding Gaussian process. This proof is based on a quantitative upper bound on the maximum variation of a parameter during training. This bound implies that for sufficiently large widths, training occurs in the lazy regime, \emph{i.e.}, each parameter changes only by a small amount. While the convergence result of [Girardi \emph{et al.}, arXiv:2402.08726] holds at a fixed training time, our upper bounds are uniform in time and hold even as $t \to \infty$.