Article
Multidisciplinary Sciences
Xuxing Liu, Weize Gao, Rankang Li, Yu Xiong, Xiaoqin Tang, Shanxiong Chen
Summary: Ancient character recognition is crucial for the study of ancient history and the preservation of national culture. To address the challenges in ancient character research, we propose a Siamese similarity network that directly learns input similarity and establishes a one-shot classification task. The network features multi-scale fusion and an embedded structure to enhance feature extraction. We also introduce a soft similarity contrast loss function to optimize similar images while reducing overfitting. Our model is capable of rejecting unknown categories and has achieved superior performance compared to traditional deep learning and classic one-shot learning methods.
SCIENTIFIC REPORTS
(2022)
Article
Engineering, Electrical & Electronic
Carlos Iturrino-Garcia, Gabriele Patrizi, Alessandro Bartolini, Lorenzo Ciani, Libero Paolucci, Antonio Luchetta, Francesco Grasso
Summary: Power quality disturbances have become a concern for many due to the increasing number of nonlinear loads and renewable sources connected to the grid. This work presents a novel algorithm, SSPQDD, which outperforms other algorithms in terms of computational resources, accuracy, and layers. Experimental results using simulation and real measurement data validate the effectiveness of SSPQDD in detecting PQDs with an overall accuracy of 96.55%.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Computer Science, Artificial Intelligence
Linchao Zhu, Yi Yang
Summary: In this paper, the authors propose a method to leverage freely available unlabeled video data for few-shot video classification. By introducing a label independent memory and a multi-modality compound memory network, they address the challenge of imbalanced and unlabelled videos and achieve better classification performance.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Chemistry, Analytical
Dongri Shan, Yalu Xu, Peng Zhang, Xiaofang Wang, Dongmei He, Chenglong Zhang, Maohui Zhou, Guoqi Yu
Summary: This paper proposes a novel dual-path multi-scale object detection paradigm and designs a single-stage general object detection algorithm called DPSSD. The experimental results show that the algorithm outperforms other single-stage object detection algorithms in terms of average accuracy, reaching an advanced level.
Article
Optics
Zhenyu Ju, Zhenming Yu, Ziyi Meng, Ning Zhan, Lili Gui, Kun Xu
Summary: Researchers proposed a dual-function MMF imaging system that can transmit illumination light and images through the same fiber, and designed a deep learning-based network for image reconstruction. Experimental results show that the proposed network achieves the best reconstruction performance and the cropped speckle patterns can still be used for image reconstruction, reducing the computing complexity. The ability of cross-domain generalization of the proposed network is also demonstrated, showing the potential for more compact endoscopic imaging without external illumination.
Article
Computer Science, Artificial Intelligence
Junyu Gao, Tianzhu Zhang, Changsheng Xu
Summary: This study proposes a task-driven message passing process using a prototype-sample GNN to achieve zero-shot learning in video classification, successfully establishing relationships between categories and attributes, and achieving favorable performance on five popular video benchmarks.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Materials Science, Multidisciplinary
Dingchen Wang, Dingyao Liu, Yinan Lin, Anran Yuan, Woyu Zhang, Yaping Zhao, Shaocong Wang, Xi Chen, Hegan Chen, Yi Zhang, Yang Jiang, Shuhui Shi, Kam Chi Loong, Jia Chen, Songrui Wei, Qing Wang, Hongyu Yu, Renjing Xu, Dashan Shang, Han Zhang, Shiming Zhang, Zhongrui Wang
Summary: The bio-inspired material-algorithm co-design of the hydrogel-based optical Willshaw model (HOWM) enables simultaneous optical path configuration and deep learning model optimization thanks to opto-chemical reactions. HOWM serves as an all optical in-sensor AI processor and a ternary content addressable memory (TCAM) of an optical memory augmented neural network (MANN) for one-shot learning. HOWM achieves a 1000x boost in energy efficiency and a 10x boost in speed, paving the way for the next-generation autonomous, efficient, and affordable smart edge systems.
Article
Optics
Xianglei Liu, Joao Monteiro, Isabela Albuquerque, Yingming Lai, Cheng Jiang, Shian Zhang, Tiago H. Falk, Jinyang Liang
Summary: The paper introduces a snapshot-to-video autoencoder (S2V-AE) for machine-learning assisted real-time processing in compressed ultrahigh-speed imaging, achieving efficient reconstruction time and a large sequence depth.
PHOTONICS RESEARCH
(2021)
Article
Engineering, Electrical & Electronic
Cheng-Xin Xue, Yen-Cheng Chiu, Ta-Wei Liu, Tsung-Yuan Huang, Je-Syu Liu, Ting-Wei Chang, Hui-Yao Kao, Jing-Hong Wang, Shih-Ying Wei, Chun-Ying Lee, Sheng-Po Huang, Je-Min Hung, Shih-Hsih Teng, Wei-Chen Wei, Yi-Ren Chen, Tzu-Hsiang Hsu, Yen-Kai Chen, Yun-Chen Lo, Tai-Hsing Wen, Chung-Chuan Lo, Ren-Shuo Liu, Chih-Cheng Hsieh, Kea-Tiong Tang, Mon-Shu Ho, Chin-Yi Su, Chung-Cheng Chou, Yu-Der Chih, Meng-Fan Chang
Summary: The development of small, energy-efficient artificial intelligence edge devices has been limited by data transfer requirements between the processor and memory in traditional computing architectures. Non-volatile compute-in-memory (nvCIM) architectures show potential to overcome these limitations, but challenges still remain in developing configurations for high-bit-precision dot-product operations.
NATURE ELECTRONICS
(2021)
Article
Optics
Vaishnavi Ravi, Rama Krishna Gorthi
Summary: 3D reconstruction by fringe projection is an ill-posed problem. In this study, a Circular Fringe-to-3D reconstruction Network (CF3DNet) is proposed, which can establish a one-to-one mapping between phase deformations and absolute phase shifts. It can reconstruct discontinuous object profiles with the help of radially symmetric circular fringes.
OPTICS AND LASERS IN ENGINEERING
(2023)
Article
Automation & Control Systems
Xuan-Thuy Vo, Kang-Hyun Jo
Summary: This article proposes a novel bounding box encoding algorithm integrated into the single-shot detector to consider the flexible distribution of bounding box localization. It introduces detection quality by combining the localization and classification quality to rank detection during nonmaximum suppression. The proposed BBENet achieves state-of-the-art performance in single-shot object detection on benchmark datasets.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2022)
Article
Computer Science, Artificial Intelligence
Xin Xu, Junping Du, Zhe Xue
Summary: This paper proposes a novel graph meta-learning framework called MSPN, which utilizes multi-level adaptive learning to highlight the most expressive nodes and features and rectify the prototype to obtain a more accurate classification. Experimental results demonstrate that MSPN outperforms other methods in few-shot node classification.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Environmental Sciences
Haizhu Pan, Moqi Liu, Haimiao Ge, Liguo Wang
Summary: This article proposes a novel one-shot dense network with polarized attention (OSDN) for hyperspectral image (HSI) classification. OSDN has two independent branches to extract spectral and spatial features and uses specially designed filtering methods to maintain high internal resolution in both the channel and spatial dimensions. Extensive experiments on benchmark HSI datasets demonstrate that OSDN can significantly reduce computational cost and parameters while maintaining high accuracy with a few training samples.
Article
Computer Science, Artificial Intelligence
Kai He, Nan Pu, Mingrui Lao, Michael S. S. Lew
Summary: This paper introduces a method called meta-learning, which effectively utilizes prior knowledge to guide the learning of new tasks by simulating the tasks that will be presented at inference. The paper provides a comprehensive overview and key insights into meta-learning methods, categorizing them into three branches based on their technical characteristics: metric-based, model-based, and optimization-based meta-learning. Additionally, the paper presents an overview of current widely used benchmarks and the performances of recent meta-learning methods on these datasets.
INTERNATIONAL JOURNAL OF MULTIMEDIA INFORMATION RETRIEVAL
(2023)
Article
Computer Science, Information Systems
Mauro Conti, Shubham Khandhar, P. Vinod
Summary: This paper proposes few-shot classification techniques for accurate malware classification without the need for re-training the classifier. It also introduces a novel malware visualization technique using image representation. The experiments show the effectiveness of these approaches.
COMPUTERS & SECURITY
(2022)
Article
Physics, Multidisciplinary
Daniel Szombati, Alejandro Gomez Frieiro, Clemens Mueller, Tyler Jones, Markus Jerger, Arkady Fedorov
PHYSICAL REVIEW LETTERS
(2020)
Article
Physics, Applied
Xiao Guo, Xin He, Zach Degnan, Bogdan C. Donose, Karl Bertling, Arkady Fedorov, Aleksandar D. Rakic, Peter Jacobson
Summary: Superconducting quantum circuits are a leading quantum computing platform, and it is crucial to identify and address material imperfections that lead to decoherence for practical advancements. The study used terahertz scanning near-field optical microscopy to probe the local properties of wet-etched aluminum resonators on silicon, revealing higher carrier concentration in silicon within the etched channel compared to buffer oxide etched silicon. The results demonstrate that near-field THz investigations can be used to quantitatively evaluate and identify inhomogeneities in quantum devices.
APPLIED PHYSICS LETTERS
(2021)
Article
Physics, Applied
Tyler Jones, Kaiah Steven, Xavier Poncini, Matthew Rose, Arkady Fedorov
Summary: Classical simulations of time-dependent quantum systems are widely used in quantum control research, but the selection of models plays a crucial role in the effectiveness of optimized control protocols applied to hardware. Experimental characterization of different models and benchmarking of control protocols on quantum devices reveal error amplification and the emergence of uncorrectable errors due to underlying mistreatment of noncomputational dynamics. Despite substantial variance in numerical predictions across models, the complexity of discovering local optimal control protocols remains invariant in the simple control scheme setting.
PHYSICAL REVIEW APPLIED
(2021)
Article
Physics, Multidisciplinary
Rohit Navarathna, Dat Thanh Le, Andres Rosario Hamann, Hien Duy Nguyen, Thomas M. Stace, Arkady Fedorov
Summary: This study reports the first proof-of-principle realization of a passive on-chip circulator that is made from a superconducting loop interrupted by three notionally identical Josephson junctions and is tuned with only dc control fields. The experimental results show evidence for nonreciprocal scattering and excellent agreement with theoretical simulations. By reducing the junction asymmetry and utilizing the known methods of protection from quasiparticles, the Josephson-loop circulator is expected to become ubiquitous in superconducting circuits.
PHYSICAL REVIEW LETTERS
(2023)
Article
Nanoscience & Nanotechnology
Xiao Guo, Xin He, Zachary Degnan, Chun-Ching Chiu, Bogdan C. Donose, Karl Bertling, Arkady Fedorov, Aleksandar D. Rakic, Peter Jacobson
Summary: This study demonstrates the use of nanoscale THz plasmon polaritons as an indicator of surface quality in prototypical quantum devices properties. By utilizing THz near-field hyperspectral measurements, polaritonic features in doped silicon near a metal-semiconductor interface were observed. A multilayer extraction procedure was applied to quantitatively probe the doped surface layer and determine its thickness and complex permittivity. The results matched the dielectric conditions necessary to support the THz surface plasmon polariton. Additionally, it was shown that etching of this doped layer leads to an increase in the quality factor of superconducting resonators as determined by cryogenic measurements.
Proceedings Paper
Computer Science, Hardware & Architecture
Tina Moghaddam, Minjune Kim, Jin-Hee Cho, Hyuk Lim, Terrence J. Moore, Frederica F. Nelson, Dan Dongseong Kim
Summary: This paper presents a practical evaluation of the effectiveness of virtual IP-shuffling MTD technique in a SDN testbed. The results validate the effectiveness of the MTD technique and show that attackers can adjust their approach if they are aware of the MTD technique being used, in order to increase their success rate.
52ND ANNUAL IEEE/IFIP INTERNATIONAL CONFERENCE ON DEPENDABLE SYSTEMS AND NETWORKS WORKSHOP VOLUME (DSN-W 2022)
(2022)
Article
Materials Science, Multidisciplinary
Zach Degnan, Xin He, Alejandro Gomez Frieiro, Yauhen P. Sachkou, Arkady Fedorov, Peter Jacobson
Summary: The study explores the impact of two new substrate materials, spinel and lanthanum aluminate, on the performance of superconducting quantum devices, finding that devices fabricated on lanthanum aluminate have quality factors three times higher than earlier devices, while MgAl2O4 consistently outperforms lanthanum aluminate. These results underscore the importance of material exploration, substrate preparation, and characterization for high-performance superconducting quantum circuitry.
MATERIALS FOR QUANTUM TECHNOLOGY
(2022)
Article
Physics, Multidisciplinary
Dat Thanh Le, Clemens Mueller, Rohit Navarathna, Arkady Fedorov, T. M. Stace
Summary: This study focuses on two operational issues of a superconducting circulator based on Josephson-junction rings, optimizing performance through numerical methods and discussing the impact of quasiparticle tunneling on signal circulation.
PHYSICAL REVIEW RESEARCH
(2021)
Article
Materials Science, Multidisciplinary
Lukas Homeier, Christian Schweizer, Monika Aidelsburger, Arkady Fedorov, Fabian Grusdt
Summary: This work presents a building block for Z(2) lattice gauge theories coupled to dynamical matter, enabling the implementation of the toric code ground state and its topological excitations. The proposal is realized in the second-order coupling regime and is well suited for superconducting qubit implementations. The study outlines a pathway for preparing topologically nontrivial initial states and experimental signatures of the ground-state wave function.
