quEEGNet Quantum AI for Biosignal Processing Toshiaki Koike-Akino Ye Wang Mitsubishi Electric Research Laboratories MERL

2025-04-29 0 0 851.65KB 4 页 10玖币

侵权投诉

quEEGNet: Quantum AI for Biosignal Processing

Toshiaki Koike-Akino, Ye Wang

Mitsubishi Electric Research Laboratories (MERL)

201 Broadway, Cambridge, MA 02139, USA

{koike, yewang}@merl.com

Abstract—In this paper, we introduce an emerging quantum

machine learning (QML) framework to assist classical deep

learning methods for biosignal processing applications. Specif-

ically, we propose a hybrid quantum-classical neural network

model that integrates a variational quantum circuit (VQC)

into a deep neural network (DNN) for electroencephalogram

(EEG), electromyogram (EMG), and electrocorticogram (ECoG)

analysis. We demonstrate that the proposed quantum neural

network (QNN) achieves state-of-the-art performance while the

number of trainable parameters is kept small for VQC.

Index Terms—Quantum computing, deep neural network

(DNN), quantum machine learning (QML), electroencephalo-

gram (EEG), electromyogram (EMG), biosignal processing

I. INTRODUCTION

The great advancement of artiﬁcial intelligence (AI) tech-

niques based on deep neural networks (DNN) has enabled

practical development of human-machine interfaces (HMI)

including brain-computer interfaces (BCI) through the analysis

of the user’s physiological data [1], such as electroencephalo-

gram (EEG) [2] and electromyogram (EMG) [3]. However,

such biosignals are highly prone to variation depending on

the biological states of each subject [4]. Hence, frequent

calibration is often required in typical HMI systems. Toward

resolving this issue, subject-invariant methods [5]–[11], em-

ploying domain generalization and transfer learning, have been

proposed to reduce user calibration for HMI systems.

In this paper, we introduce an emerging framework “quan-

tum machine learning (QML)” [12]–[31] into biosignal pro-

cessing applications for the ﬁrst time in the literature, envision-

ing future era of quantum supremacy [32], [33]. Quantum com-

puters have the potential to realize computationally efﬁcient

signal processing compared to traditional digital computers

by exploiting quantum mechanisms, e.g., superposition and

entanglement, in terms of not only execution time but also

energy consumption. In the past few years, several vendors

have successfully manufactured commercial quantum process-

ing units (QPUs). For instance, IBM released 127-qubit QPUs

in 2021, and plans to produce 1121-qubit QPUs by 2023.

It is thus no longer far future when QML will be widely

used for real applications. Recently, hybrid quantum-classical

algorithms based on the variational principle [34]–[37] were

proposed to deal with quantum noise.

The main contributions of this paper are summarized below:

•We introduce the emerging QML framework for biosignal

processing;

•We propose a hybrid quantum-classic DNN model called

quEEGNet;

•We demonstrate the proof-of-concept study on QML for

various physiological datasets.

To the best of our knowledge, this is the very ﬁrst research on

QML applied to HMI and BCI ﬁelds. Although there exist

a few literature [38], [39] discussing the potential use of

quantum computing for BCI, no practical demonstration on

QML-assisted HMI systems is found to date. Note that our

QNN is different from a recurrent QNN (RQNN) employing

quantum stochastic ﬁltering based on the Schr¨

odinger equa-

tion [40]–[43], which is motivated by quantum physics but

does not need real QPUs. In addition, our work is tangential

to quantum sensing technologies such as superconducting

quantum interference devices (SQUID) [44].

II. QUANTUM ARTIFICIAL INTELLIGENCE (QAI) FOR HMI

A. Quantum Bit (Qubit)

In quantum systems, a qubit is expressed as the following

state superposing bases of |0iand |1i:|φi=α0|0i+α1|1i,

where α1and α2are complex numbers subject to |α0|2+

|α1|2= 1. When qubits are measured, the classical bit 0or 1is

observed with a probability of |α0|2or |α1|2, respectively. The

above ket-notation corresponds to column-vector operations of

the two basis states |0i= [1,0]Tand |1i= [0,1]T, whereas

the bra-notation is used for row-vector operations corresponds

to its Hermitian transpose; i.e., hφ|=|φi†= [α∗

0, α∗

1].

Here, [·]†,[·]∗and [·]Tdenote Hermitian transpose, complex

conjugate and transpose, respectively. Note that a multi-qubit

state is represented by sum of Kronecker products of basis

vectors such as |000i=|0i⊗3.

B. Quantum Gates

The basic operations on a qubit is deﬁned as a unitary

matrix, which is called gate. Some of the most common gates

are associated with Pauli matrices: I= [ 1 0

0 1 ],X= [ 0 1

1 0 ],

Y=0−

0, and Z=1 0

0−1, where is the imaginary unit

satisfying 2=−1. The X gate is bit-ﬂip (i.e., NOT operation),

Z gate is phase-ﬂip, and Y gate ﬂips both bit and phase. The

Hadamard (H) gate is used to generate a superposition state

|+i=1

√2|0i+1

√2|1i:H=1

√21 1

1−1. A controlled-NOT

(CNOT or CX) gate is a multi-qubit gate that ﬂips the target

qubit if and only if the control qubit is |1i.

C. Quantum Machine Learning (QML)

A number of modern DNN methods have been already

migrated into the quantum domain, e.g., convolutional lay-

ers [12], autoencoders [13], graph neural networks [17], and

arXiv:2210.00864v1 [quant-ph] 29 Sep 2022

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

quEEGNet:QuantumAIforBiosignalProcessingToshiakiKoike-Akino,YeWangMitsubishiElectricResearchLaboratories(MERL)201Broadway,Cambridge,MA02139,USAfkoike,yewangg@merl.comAbstractInthispaper,weintroduceanemergingquantummachinelearning(QML)frameworktoassistclassicaldeeplearningmethodsforbiosignalprocessi...

展开>> 收起<<

quEEGNet Quantum AI for Biosignal Processing Toshiaki Koike-Akino Ye Wang Mitsubishi Electric Research Laboratories MERL.pdf

共4页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

quEEGNet Quantum AI for Biosignal Processing Toshiaki Koike-Akino Ye Wang Mitsubishi Electric Research Laboratories MERL

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: