Abstract Large machine learning models are revolutionary technologies of artificial intelligence whose bottlenecks include huge computational expenses, power, and time used both in the pre-training fine-tuning process. In this work, we show that fault-tolerant quantum computing could possibly provide provably efficient resolutions for generic (stochastic) gradient descent algorithms, scaling as $${{{{{{{\mathcal{O}}}}}}}}({T}^{2}\times {{{{{{{\rm{polylog}}}}}}}}(n))$$ <mml:math...

In this letter, a reduced-dimension multiple signal classification (MUSIC) algorithm for near-field source localization (i.e., elevation angle and range) with uniform linear arrays is proposed. By splitting the directional matrix in terms of range, spectrum function can be constructed, where 2-D spectral search avoided only over domain involved. Moreover, contrast to rank-reduced (RARE) several times 1-D searches, proposed obtain estimates by one search, then, paired range directly computed,...

As a generalized coprime array structure, the with displaced subarrays (CADiS) allows large minimum inter-element spacing by introducing specific displacement between two sparse subarrays. While this structure can effectively reduce mutual coupling, holes in its difference co-array greatly decrease achievable number of uniform degrees freedom (DOFs). In paper, we first provide complete characterization for hole locations generated tailored CADiS (tCADiS) as union four subsets related via...

Generally, massive multiple-input multiple-output (MIMO) system incorporates hundreds of antennas at the base station for various attractive merits, whereas severe mutual coupling arises in dense structure. In this paper, we introduce sparse array configuration, coprime array, into MIMO to alleviate coupling, increase degrees freedom (DOFs) and enhance spatial resolution. Specifically, propose extended structures through two systematical schemes. One scheme is slide one subarray typical...

In this letter, the issue of direction arrival (DOA) estimation for narrowband signals with two parallel linear arrays is discussed, and an extended DOA-Matrix method derived. Specifically, proposed makes full use auto-correlation cross-correlation information by constructing DOA matrix. Subsequently, auto-paired estimates can be calculated eigenvalues eigenvectors Moreover, outperforms traditional which partially neglects information. Simulation results are provided to validate superiority method.

In this letter, we investigate the issue of sparse array design for direction arrival (DOA) estimation non-circular (NC) signals, and propose transformed nested (TNA) which reduces overlapping virtual sensors between difference sum co-array generated by conventional (NA) can thus obtain a sum-difference with higher degrees freedom (DOFs). From redundancy perspective, construct TNA via two operations. The first is to exchange positions subarrays NA in order redistribute its separate sensors....

Abstract Our study evaluates the limitations and potentials of Quantum Random Access Memory (QRAM) within principles quantum physics relativity. QRAM is crucial for advancing algorithms in fields like linear algebra machine learning, purported to efficiently manage large data sets with $${{{\mathcal{O}}}}(\log N)$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mi>O</mml:mi> <mml:mo>(</mml:mo> <mml:mi>log</mml:mi> <mml:mi>N</mml:mi> </mml:mrow> <mml:mo>)</mml:mo>...

With the development of Internet, information carriers and communication methods have become diversified. As an important carrier for obtaining information, images are easily copied tampered with during process. Digital watermarking technology is effective to protect image copyrights. In recent years, artificial intelligence, scholars begun apply deep learning digital technology. This article proposes a based on residual neural networks, which consists host processing network, watermark...

A diffusion probabilistic model (DPM) is a generative renowned for its ability to produce high-quality outputs in tasks such as image and audio generation. However, training DPMs on large, high-dimensional datasets high-resolution images or incurs significant computational, energy, hardware costs. In this work, we introduce efficient quantum algorithms implementing through various ODE solvers. These highlight the potential of Carleman linearization diverse mathematical structures, leveraging...

Adequate records of non-stationary extreme winds (e.g., typhoons and downbursts) are crucial for the reliable design structures located in areas with frequent occurrence such winds. Affected by limited capabilities complex service environment anemometers, quality wind data is often difficult to guarantee, resulting scarcity effective data. To this end, a new record-based simulation method, which consists two critical parts, i.e., multivariate fast iterative filtering-based instantaneous...

Nested array has aroused remarkable attention due to the capability obtain both enhanced degrees of freedom (DOFs) and enlarged aperture, whereas only difference co-array is employed. Here, authors construct sum nested by utilising non-circular (NC) characteristic and, subsequently, sum–difference (SD) obtained careful crafting typical co-array, which doubles resulting aperture simultaneously, further increases achievable DOF compared with array. Moreover, circumvent time-consuming...

In the past decade, sparse arrays have aroused considerable attention for ability to provide larger array aperture, more degrees of freedom (DOFs) and better estimation performance compared with uniform linear (ULA). this paper, problem design is investigated direction arrival (DOA) non-circular (NC) signals, where both difference co-array (DCA) sum (SCA) physical configuration can be utilized. Specifically, we first introduce a sliding strategy which moves along axis as whole in order...

Large machine learning models are revolutionary technologies of artificial intelligence whose bottlenecks include huge computational expenses, power, and time used both in the pre-training fine-tuning process. In this work, we show that fault-tolerant quantum computing could possibly provide provably efficient resolutions for generic (stochastic) gradient descent algorithms, scaling as O(T^2 polylog(n)), where n is size T number iterations training, long sufficiently dissipative sparse, with...

In recent years, Ethereum has become a hotspot for criminal activities such as phishing scams that seriously compromise transaction security. However, existing methods cannot accurately model data and make full use of the temporal structure information basic account features. this paper, we propose an detection framework based on motif By designing sampling method, convert labeled addresses into multi-directed subgraphs with time amount to avoid losing attribute information. To learn...

In this paper, we investigate the issue of direction arrival (DOA) estimation non-circular (NC) signals, and propose a modified nested array with enhanced degrees freedom (DOFs) vectorized NC propagator method (VNCPM) algorithm which can utilize difference-sum co-array to perform DOA estimation. Specifically, modify conventional by switching positions two subarrays aiming obtain higher DOFs. order fully exploit provided DOFs accuracy, characteristic generate equivalent received signal...

Footprint recognition plays an important role in criminal investigation, security protection and other fields. In terms of the key problems feature extraction classification, a footprint algorithm is proposed based on optical images. The first transforms image into complex frequency domain through dual-tree wavelet transform (DTCWT), then extracts histogram oriented gradient (HOG) features domain. Finally, symmetric projection matrix obtained metric learning. used to project original new...

针对快速水平集算法用于图像分割时，存在水平集初始化和阈值设置的困难，该文提出一种融合金字塔模型、随机游走及水平集(PYR-RW-LS)的新算法。首先将多尺度分析引入随机游走算法，把分割结果作为快速水平集算法的初始化曲线，解决其初始化问题；接着把水平集演化看成对曲线上的点不断进行模式分类的过程，引入贝叶斯分类决策和最小距离分类决策交替工作，产生曲线演化所需的驱动力，同时将两种分类决策的失效条件作为新算法迭代停止的条件，解决了快速水平集算法阈值设置的困难。仿真实验结果表明：PYR-RW-LS算法比只采用模式分类思想的快速水平集算法拥有更高的计算效率，且在抗噪性方面亦优于随机游走算法，同时保留了随机游走算法对弱边缘不敏感的优点，尤其适用于大尺寸，高清晰度的图像处理。