Entanglement estimation of Werner states with a quantum extreme learning machine

Hajar Assil, Abderrahim El Allati, Gian Luca Giorgi · 2025

Read Entanglement estimation of Werner states with a quantum extreme learning machine by Hajar Assil, Abderrahim El Allati, Gian Luca Giorgi (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-01387.txt.

Introduction

Hajar Assil, Abderrahim El Allati, and Gian Luca Giorgi estimate entanglement with a quantum extreme learning machine. In that scheme a fixed quantum reservoir does the processing, and only a simple output layer is trained. Their target is the Werner states, a standard family of two qubit states mixing a maximally entangled state with noise.

The protocol generates a sequence of random Werner states, combines them with a reservoir state, and evolves the whole under an Ising Hamiltonian. Measurements in the Bloch basis supply the features used to train the readout to recognize entanglement in unseen states. The authors test robustness by adding noise to the inputs, and they study how the magnetic field parameter in the Hamiltonian affects the estimation accuracy.

Catalog record

CREATORHajar Assil, Abderrahim El Allati, Gian Luca Giorgi
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-01387

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Entanglement estimation of Werner states with a quantum extreme learning machine Preprint: APS/123-QED Hajar Assil, Abderrahim El Allati, Gian Luca Giorgi Abstract Quantum Extreme Learning Machines (QELMs) have emerged as a potent tool for various quantum information processing tasks. We present a QELM protocol for estimating the amount of entanglement in Werner states. The protocol requires the generation of a sequence of random Werner states, which are then combined with a reservoir state and evolved using an Ising Hamiltonian. A set of observables based on the Bloch basis is constructed and employed to train the system to recognize unseen features. To assess the protocol’s robustness, noise is introduced into the input states, and the system’s performance under these noisy conditions is analyzed. Additionally, the influence of the magnetic field parameter within the Ising Hamiltonian on the estimation accuracy is investigated. I Introduction Traditional machine learning algorithms face challenges as datasets become more complex and larger. Quantum computing presents a promising solution to these limitations with its inherent parallelism and exponential computational capabilities. By utilizing the distinct characteristics of quantum systems, Quantum Machine learning can reach new frontiers in speed, precision, and scalability for training neural networks Wittek 2014 ; Schuld et al. 2015 ; Biamonte et al. 2017 ; Dunjko and Briegel 2018 . A very successful machine learning paradigm is represented by reservoir computers (RCs) and extreme learning machines (ELMs), which are subclasses of recurrent neural networks (RNN) that do not require the tuning of the parameters of the reservoir, as during the learning process, only the weights of the output layer are adapted through linear regression Jaeger 2001 ; Maass and Markram 2004 ; Nakajima and Fischer 2021 ; Huang et al. 2011 . Besides the easiness of the training process, reservoir computers and extreme learning machines are known for their ability to solve multiple tasks simultaneously by training the output layer independently for each task. The primary distinguishing factor between RCs and ELMs is the reservoir’s internal memory utilization. While reservoir computers retain the memory of past inputs, making them appropriate for processing temporal data, extreme learning machines perform static tasks and are not required to possess internal memory. The extension to the quantum domain of the RC Fujii and Nakajima 2017 ; Mujal et al. 2021 ; Nokkala et al. 2021 ; Martínez-Peña et al. 2021 ; Bravo et al. 2022 ; Kornjača et al. 2024 ; Senanian et al. 2024 and ELM Ghosh et al. 2019a ; Ghosh et al. 2019b ; Innocenti et al. 2023 ; Suprano et al. 2024 ; Xiong et al. 2023 paradigms are motivated by the promise of the exponential size of the Hilbert space compared to the physical size of the system Martínez Peña et al. 2020 ; Kalfus et al. 2022 ; Dudas et al. 2023 ; Sakurai et al. 2022 , which can potentially lead to overcome classical algorithms, and with the intrinsic ability to process quantum inputs Nokkala 2023 ; Nokkala et al. 2024 without requiring any information pre-processing, which would be inefficient in classical hardware. Furthermore, the popularity of QRC and QELM is motivated by their immediate applicability in noisy intermediate-scale quantum (NISQ) devices Preskill 2018 ; Bharti et al. 2022 . The concept of ELM was introduced within the broader framework of reservoir computing, which evolved from echo state networks (ESN) and liquid state machines (LSM) in the early 2000s ( Hauser 2021 ) , ( Martínez Peña et al. 2020 ) . The RC system consists of a reservoir denoted as \mathcal{R} and a trainable readout layer w . The reservoir retains its shape over time and is instrumental in processing information. Various reservoir types, such as traditional RNNs, ESNs, LSMs, and physical systems like octopus arms or photonic reservoirs, can be employed. The reservoir serves as an information transducer, transforming input information into the system’s state. Notably, from ELM, the state \mathbf{x}_{k} of the reservoir is uniquely determined by the corresponding input u_{k} , as expressed by the equation \mathbf{x}_{k}=f\left(u_{k}\right). (1) This design choice differs from reservoir computing, where the state evolution depends on both current and past inputs, which makes ELM useful in classification tasks. Consider a classification problem that aims to determine whether a picture contains a car or not. The presence of a car in a previous picture has no bearing on the presence of a car in the current picture ( Mujal et al. 2021 ) . Quantum Extreme Learning Machine (QELM) stands as an interesting form within the context of Quantum Reservoir Computing (QRC), a paradigm that seamlessly integrates quantum computing principles with physical reservoir computing for real-time machine learning applications ( Fujii and Nakajima 2021 ; Fujii and Nakajima 2017 ) . While QRC processes input data over time, capturing temporal dependencies and correlations among previous inputs, QELM introduces a distinctive departure from this temporal paradigm. Compared to QRC, QELM does not rely on temporal dependencies and does not store information about previous inputs in the quantum reservoir. Instead, QELM performs a static, one-shot transformation of the input data using the quantum reservoir. This characteristic renders QELM particularly suitable for tasks such as classification, where the inherent temporal dynamics of sequential data may be less crucial. The novel approach of QELM capitalizes on the unique features of quantum systems, including the exponential growth of degrees of freedom in a Hilbert space (d.o.f \sim 4^{N} for a spin-1/2 system). The temporal independence of QELM makes it an intriguing model for specific applications within the broader landscape of Quantum Reservoir Computing. In this paper, we investigate an extreme learning machine technique that employs a quantum network consisting of randomly coupled spins, similar to the proposals described in Fujii et al. (2017). Our goal is to accurately determine the Werner state parameter p using a machine learning algorithm that relies solely on individual qubit measurements. The protocol’s resilience to noise in input states is evaluated, shedding light on its efficacy under noisy conditions. Furthermore, we explore the concept of generalizability in our QELM. We train the network using data from various domains and then assess its performance on a target domain not included in the training set. This analysis provides insight into the system’s capability to learn and apply its knowledge to unseen data, a crucial aspect for practical applications. The structure of this paper is outlined as follows: In the initial section, we articulate our proposed model, utilizing the Transverse Ising model as a reservoir, illuminating each step in the process. Subsequently, we assess the robustness of the reservoir by introducing noise to the input states. The results section is dedicated to an in-depth discussion and analysis of the various findings. II Protocol Reservoir - Our reservoir is given by a Transverse Ising Model Fujii and Nakajima 2017 ; Martínez Peña et al. 2020 ; Martínez-Peña et al. 2021 , whose Hamiltonian reads H=\sum_{i<j}^{N}J_{ij}\sigma_{i}^{x}\sigma_{j}^{x}+h\sum_{i=1}^{N}\sigma_{i}^{z} (2) where i,j label the sites of the network, \sigma_{i}^{a}(a=x,y,z) are the Pauli matrices acting on the i- th site, h is the value of the external magnetic field, J_{ij} is the spin-spin coupling, randomly selected from a uniform distribution in the range \{-J_{s}/2,J_{s}/2\} with J_{s}=1 and N is the number of qubits. As

Read the complete text