Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2102.08609 · Journal of the Physical Society of Japan

Teacher-Student Learning for a Binary Perceptron with Quantum Fluctuations

Shunta Arai, Masayuki Ohzeki, Kazuyuki Tanaka

Published Jun 10, 2021• 16 citations
arXivHTMLPDF

Abstract

We analysed the generalisation performance of a binary perceptron with quantum fluctuations using the replica method. An exponential number of local minima dominate the energy landscape of the binary perceptron. Local search algorithms often fail to identify the ground state of a binary perceptron. In this study, we considered the teacher-student learning method and computed the generalisation error of a binary perceptron with quantum fluctuations. Due to the quantum fluctuations, we can efficiently find robust solutions that have better generalisation performance than the classical model. We validated our theoretical results through quantum Monte Carlo simulations. We adopted the replica symmetry (RS) ansatz assumption and static approximation. The RS solutions are consistent with our simulation results, except for the relatively low strength of the transverse field and high pattern ratio. These deviations are caused by the violation of ergodicity and static approximation. After accounting for the deviation between the RS solutions and numerical results, the enhancement of generalisation performance with quantum fluctuations holds.

DOI
10.7566/jpsj.90.074002

Related papers

On quantum perceptron learning via quantum search

Xiaoyu Sun, Mathieu Roget, Giuseppe Di Molfetta +1 more

Abstract With the growing interest in quantum machine learning, the perceptron, a fundamental building block in traditional machine learning, has emerged as a valuable model for exploring the potential of quantum algorithms. In this work, we make two principal contributions. First, we revisit the quantum version space perceptron algorithm proposed by Kapoor et al. (2016), by identifying and correcting a flawed complexity assumption. We show that the query complexity of the algorithm is dimension-dependent, which has significant implications for its behaviour in high-dimensional regimes under worst-case scenarios. Second, we propose and analyse two quantum-enhanced cutting-plane algorithms for perceptron learning. Specifically, we leverage established quantum subroutines such as Grover’s search and quantum walk search , and provide detailed algorithmic constructions together with query and arithmetic complexity analyses. Our results establish improved complexity bounds under an idealised implementation framework and noise-free quantum computational models, offering insights into the trade-offs between margin dependence, dimensional dependence, and quantum resources. These findings provide a refined understanding of quantum perceptron models and their theoretical computational complexity properties.

arXiv:2503.17308•Quantum Machine Intelligence•Jun 19, 2026•0 citations

Exploring Quantum Perceptron and Quantum Neural Network structures with a teacher-student scheme

Aikaterini, Gratsea, Patrick Huembeli

Near-term quantum devices can be used to build quantum machine learning models, such as quantum kernel methods and quantum neural networks (QNN) to perform classification tasks. There have been many proposals how to use variational quantum circuits as quantum perceptrons or as QNNs. The aim of this work is to systematically compare different QNN architectures and to evaluate their relative expressive power with a teacher-student scheme. Specifically, the teacher model generates the datasets mapping random inputs to outputs which then have to be learned by the student models. This way, we avoid training on arbitrary data sets and allow to compare the learning capacity of different models directly via the loss, the prediction map, the accuracy and the relative entropy between the prediction maps. We focus particularly on a quantum perceptron model inspired by the recent work of Tacchino et. al. \cite{Tacchino1} and compare it to the data re-uploading scheme that was originally introduced by Pérez-Salinas et. al. \cite{data_re-uploading}. We discuss alterations of the perceptron model and the formation of deep QNN to better understand the role of hidden units and non-linearities in these architectures.

arXiv:2105.01477•May 4, 2021•2 citations

Quantum Hopfield Model

Yoshihiko Nonomura, Hidetoshi Nishimori

The Hopfield model in a transverse field is investigated in order to clarify how quantum fluctuations affect the macroscopic behavior of neural networks. Using the Trotter decomposition and the replica method, we find that the $α$ (the ratio of the number of stored patterns to the system size)-$Δ$ (the strength of the transverse field) phase diagram of this model in the ground state resembles the $α$-$T$ phase diagram of the Hopfield model quantitatively, within the replica-symmetric and static approximations. This fact suggests that quantum fluctuations play quite similar roles to thermal fluctuations in neural networks as long as macroscopic properties are concerned.

arXiv:cond-mat/9512142•Dec 21, 1995•0 citations

Quantum Neuron with Separable-State Encoding

London Cavaletto, Luca Candelori, Alex Matos-Abiague

The use of advanced quantum neuron models for pattern recognition applications requires fault tolerance. Therefore, it is not yet possible to test such models on a large scale in currently available quantum processors. As an alternative, we propose a quantum perceptron (QP) model that uses a reduced number of multi-qubit gates and is therefore less susceptible to quantum errors in current actual quantum computers with limited tolerance. The proposed quantum algorithm is superior to its classical counterpart, although since it does not take full advantage of quantum entanglement, it provides a lower encoding power than other quantum algorithms using multiple qubit entanglement. However, the use of separable-sate encoding allows for testing the algorithm and different training schemes at a large scale in currently available non-fault tolerant quantum computers. We demonstrate the performance of the proposed model by implementing a few qubits version of the QP in a simulated quantum computer. The proposed QP uses an N-ary encoding of the binary input data characterizing the patterns. We develop a hybrid (quantum-classical) training procedure for simulating the learning process of the QP and test their efficiency.

arXiv:2202.08306•Feb 16, 2022•0 citations

Artificial neural network architecture selection in a quantum computer

Adenilton José da Silva

Miniaturisation of computers components is taking us from classical to quantum physics domain. Further reduction in computer components size eventually will lead to the development of computer systems whose components will be on such a small scale that quantum physics intrinsic properties must be taken into account. The expression quantum computation and a first formal model of a quantum computer were first employed in the eighties. With the discovery of a quantum algorithm for factoring exponentially faster than any known classical algorithm in 1997, quantum computing began to attract industry investments for the development of a quantum computer and the design of novel quantum algorithms. For instance, the development of learning algorithms for neural networks. Some artificial neural networks models can simulate an universal Turing machine, and together with learning capabilities have numerous applications in real life problems. One limitation of artificial neural networks is the lack of an efficient algorithm to determine its optimal architecture. The main objective of this work is to verify whether we can obtain some advantage with the use of quantum computation techniques in a neural network learning and architecture selection procedure. We propose a quantum neural network, named quantum perceptron over a field (QPF). QPF is a direct generalisation of a classical perceptron which addresses some drawbacks found in previous models for quantum perceptrons. We also present a learning algorithm named Superposition based Architecture Learning algorithm (SAL) that optimises the neural network weights and architectures. SAL searches for the best architecture in a finite set of neural network architectures and neural networks parameters in linear time over the number of examples in the training set. SAL is the first quantum learning algorithm to determine neural network architectures in linear time. This speedup is obtained by the use of quantum parallelism and a non linear quantum operator.

•Jun 26, 2015•0 citations

Efficient Population Transfer via Non-Ergodic Extended States in Quantum Spin Glass

Kostyantyn Kechedzhi, Vadim Smelyanskiy, Jarrod R. McClean +6 more

Quantum tunneling has been proposed as a physical mechanism for solving binary optimization problems on a quantum computer because it provides an alternative to simulated annealing by directly connecting deep local minima of the energy landscape separated by large Hamming distances. However, classical simulations using Quantum Monte Carlo (QMC) were found to efficiently simulate tunneling transitions away from local minima if the tunneling is effectively dominated by a single path. We analyze a new computational role of coherent multi-qubit tunneling that gives rise to bands of non-ergodic extended (NEE) quantum states each formed by a superposition of a large number of deep local minima with similar energies. NEE provide a coherent pathway for population transfer (PT) between computational states with similar energies. In this regime, PT cannot be efficiently simulated by QMC. PT can serve as a new quantum subroutine for quantum search, quantum parallel tempering and reverse annealing optimization algorithms. We study PT resulting from quantum evolution under a transverse field of an n-spin system that encodes the energy function E(z) of an optimization problem over the set of bit configurations z. Transverse field is rapidly switched on in the beginning of algorithm, kept constant for sufficiently long time and switched off at the end. Given an energy function of a binary optimization problem and an initial bit-string with atypically low energy, PT protocol searches for other bitstrings at energies within a narrow window around the initial one. We provide an analytical solution for PT in a simple yet nontrivial model: M randomly chosen marked bit-strings are assigned energies E(z) within a narrow strip [-n -W/2, n + W/2], while the rest of the states are assigned energy 0. The PT starts at a marked state and ends up in a superposition of L marked states inside the narrow energy window whose width is smaller than W. The best known classical algorithm for finding another marked state is the exhaustive search. We find that the scaling of a typical PT runtime with n and L is the same as that in the multi-target Grover's quantum search algorithm, except for a factor that is equal to exp(n /(2B^2)) for finite transverse field B >>1. Unlike the Hamiltonians used in analog quantum unstructured search algorithms known so far, the model we consider is non-integrable and the transverse field delocalizes the marked states. As a result, our PT protocol is not exponentially sensitive in n to the weight of the driver Hamiltonian and may be initialized with a computational basis state. We develop the microscopic theory of PT by constructing a down-folded dense Hamiltonian acting in the space of marked states of dimension M. It belongs to the class of preferred basis Levy matrices (PBLM) with heavy-tailed distribution of the off-diagonal matrix elements. Under certain conditions, the band of the marked states splits into minibands of non-ergodic delocalized states. We obtain an explicit form of the heavy-tailed distribution of PT times by solving cavity equations for the ensemble of down-folded Hamiltonians. We study numerically the PT subroutine as a part of quantum parallel tempering algorithm for a number of examples of binary optimization problems on fully connected graphs.

Leibniz international proceedings in informatics•Jan 1, 2018•5 citations
We analysed the generalisation performance of a binary perceptron with quantum fluctuations using the replica method. An exponential number of local minima dominate the energy landscape of the binary perceptron. Local search algorithms often fail to identify the ground state of a binary perceptron. In this study, we considered the teacher-student learning method and computed the generalisation error of a binary perceptron with quantum fluctuations. Due to the quantum fluctuations, we can efficiently find robust solutions that have better generalisation performance than the classical model. We validated our theoretical results through quantum Monte Carlo simulations. We adopted the replica symmetry (RS) ansatz assumption and static approximation. The RS solutions are consistent with our simulation results, except for the relatively low strength of the transverse field and high pattern ratio. These deviations are caused by the violation of ergodicity and static approximation. After accounting for the deviation between the RS solutions and numerical results, the enhancement of generalisation performance with quantum fluctuations holds.
Abstract With the growing interest in quantum machine learning, the perceptron, a fundamental building block in traditional machine learning, has emerged as a valuable model for exploring the potential of quantum algorithms. In this work, we make two principal contributions. First, we revisit the quantum version space perceptron algorithm proposed by Kapoor et al. (2016), by identifying and correcting a flawed complexity assumption. We show that the query complexity of the algorithm is dimension-dependent, which has significant implications for its behaviour in high-dimensional regimes under worst-case scenarios. Second, we propose and analyse two quantum-enhanced cutting-plane algorithms for perceptron learning. Specifically, we leverage established quantum subroutines such as Grover’s search and quantum walk search , and provide detailed algorithmic constructions together with query and arithmetic complexity analyses. Our results establish improved complexity bounds under an idealised implementation framework and noise-free quantum computational models, offering insights into the trade-offs between margin dependence, dimensional dependence, and quantum resources. These findings provide a refined understanding of quantum perceptron models and their theoretical computational complexity properties.
Near-term quantum devices can be used to build quantum machine learning models, such as quantum kernel methods and quantum neural networks (QNN) to perform classification tasks. There have been many proposals how to use variational quantum circuits as quantum perceptrons or as QNNs. The aim of this work is to systematically compare different QNN architectures and to evaluate their relative expressive power with a teacher-student scheme. Specifically, the teacher model generates the datasets mapping random inputs to outputs which then have to be learned by the student models. This way, we avoid training on arbitrary data sets and allow to compare the learning capacity of different models directly via the loss, the prediction map, the accuracy and the relative entropy between the prediction maps. We focus particularly on a quantum perceptron model inspired by the recent work of Tacchino et. al. \cite{Tacchino1} and compare it to the data re-uploading scheme that was originally introduced by Pérez-Salinas et. al. \cite{data_re-uploading}. We discuss alterations of the perceptron model and the formation of deep QNN to better understand the role of hidden units and non-linearities in these architectures.
The Hopfield model in a transverse field is investigated in order to clarify how quantum fluctuations affect the macroscopic behavior of neural networks. Using the Trotter decomposition and the replica method, we find that the ααα (the ratio of the number of stored patterns to the system size)-ΔΔΔ (the strength of the transverse field) phase diagram of this model in the ground state resembles the ααα-TTT phase diagram of the Hopfield model quantitatively, within the replica-symmetric and static approximations. This fact suggests that quantum fluctuations play quite similar roles to thermal fluctuations in neural networks as long as macroscopic properties are concerned.
The use of advanced quantum neuron models for pattern recognition applications requires fault tolerance. Therefore, it is not yet possible to test such models on a large scale in currently available quantum processors. As an alternative, we propose a quantum perceptron (QP) model that uses a reduced number of multi-qubit gates and is therefore less susceptible to quantum errors in current actual quantum computers with limited tolerance. The proposed quantum algorithm is superior to its classical counterpart, although since it does not take full advantage of quantum entanglement, it provides a lower encoding power than other quantum algorithms using multiple qubit entanglement. However, the use of separable-sate encoding allows for testing the algorithm and different training schemes at a large scale in currently available non-fault tolerant quantum computers. We demonstrate the performance of the proposed model by implementing a few qubits version of the QP in a simulated quantum computer. The proposed QP uses an N-ary encoding of the binary input data characterizing the patterns. We develop a hybrid (quantum-classical) training procedure for simulating the learning process of the QP and test their efficiency.
Miniaturisation of computers components is taking us from classical to quantum physics domain. Further reduction in computer components size eventually will lead to the development of computer systems whose components will be on such a small scale that quantum physics intrinsic properties must be taken into account. The expression quantum computation and a first formal model of a quantum computer were first employed in the eighties. With the discovery of a quantum algorithm for factoring exponentially faster than any known classical algorithm in 1997, quantum computing began to attract industry investments for the development of a quantum computer and the design of novel quantum algorithms. For instance, the development of learning algorithms for neural networks. Some artificial neural networks models can simulate an universal Turing machine, and together with learning capabilities have numerous applications in real life problems. One limitation of artificial neural networks is the lack of an efficient algorithm to determine its optimal architecture. The main objective of this work is to verify whether we can obtain some advantage with the use of quantum computation techniques in a neural network learning and architecture selection procedure. We propose a quantum neural network, named quantum perceptron over a field (QPF). QPF is a direct generalisation of a classical perceptron which addresses some drawbacks found in previous models for quantum perceptrons. We also present a learning algorithm named Superposition based Architecture Learning algorithm (SAL) that optimises the neural network weights and architectures. SAL searches for the best architecture in a finite set of neural network architectures and neural networks parameters in linear time over the number of examples in the training set. SAL is the first quantum learning algorithm to determine neural network architectures in linear time. This speedup is obtained by the use of quantum parallelism and a non linear quantum operator.
Quantum tunneling has been proposed as a physical mechanism for solving binary optimization problems on a quantum computer because it provides an alternative to simulated annealing by directly connecting deep local minima of the energy landscape separated by large Hamming distances. However, classical simulations using Quantum Monte Carlo (QMC) were found to efficiently simulate tunneling transitions away from local minima if the tunneling is effectively dominated by a single path. We analyze a new computational role of coherent multi-qubit tunneling that gives rise to bands of non-ergodic extended (NEE) quantum states each formed by a superposition of a large number of deep local minima with similar energies. NEE provide a coherent pathway for population transfer (PT) between computational states with similar energies. In this regime, PT cannot be efficiently simulated by QMC. PT can serve as a new quantum subroutine for quantum search, quantum parallel tempering and reverse annealing optimization algorithms. We study PT resulting from quantum evolution under a transverse field of an n-spin system that encodes the energy function E(z) of an optimization problem over the set of bit configurations z. Transverse field is rapidly switched on in the beginning of algorithm, kept constant for sufficiently long time and switched off at the end. Given an energy function of a binary optimization problem and an initial bit-string with atypically low energy, PT protocol searches for other bitstrings at energies within a narrow window around the initial one. We provide an analytical solution for PT in a simple yet nontrivial model: M randomly chosen marked bit-strings are assigned energies E(z) within a narrow strip [-n -W/2, n + W/2], while the rest of the states are assigned energy 0. The PT starts at a marked state and ends up in a superposition of L marked states inside the narrow energy window whose width is smaller than W. The best known classical algorithm for finding another marked state is the exhaustive search. We find that the scaling of a typical PT runtime with n and L is the same as that in the multi-target Grover's quantum search algorithm, except for a factor that is equal to exp(n /(2B^2)) for finite transverse field B >>1. Unlike the Hamiltonians used in analog quantum unstructured search algorithms known so far, the model we consider is non-integrable and the transverse field delocalizes the marked states. As a result, our PT protocol is not exponentially sensitive in n to the weight of the driver Hamiltonian and may be initialized with a computational basis state. We develop the microscopic theory of PT by constructing a down-folded dense Hamiltonian acting in the space of marked states of dimension M. It belongs to the class of preferred basis Levy matrices (PBLM) with heavy-tailed distribution of the off-diagonal matrix elements. Under certain conditions, the band of the marked states splits into minibands of non-ergodic delocalized states. We obtain an explicit form of the heavy-tailed distribution of PT times by solving cavity equations for the ensemble of down-folded Hamiltonians. We study numerically the PT subroutine as a part of quantum parallel tempering algorithm for a number of examples of binary optimization problems on fully connected graphs.