Article
Computer Science, Artificial Intelligence
Meng Zhang, Fei Liu, Dongpeng Weng
Summary: With the rapid development of CNN, its accuracy has greatly improved, but it also poses challenges for mobile terminals or embedded devices with limited resources. Recent achievements in compressing CNN through low-rank decomposition have been made. Our method proposes different decomposition forms and strategies without fine-tuning. The experimental results show that the weight parameter drop can exceed 50% and the FLOPs drop is about 20% without fine-tuning, while the accuracy loss is less than 1%.
JOURNAL OF REAL-TIME IMAGE PROCESSING
(2023)
Article
Mathematics
Konstantin Sobolev, Dmitry Ermilov, Anh-Huy Phan, Andrzej Cichocki
Summary: This paper proposes a Proxy-based Automatic tensor Rank Selection method (PARS) to find the best combination of ranks for neural network compression. Experimental results show that PARS improves the results of existing decomposition methods on multiple neural networks.
Article
Management
Robert Wiedmer, Stanley E. Griffis
Summary: The network structures in global supply chains influence firm and supply chain behavior, with research finding community, scale-free, and hierarchical structures within these networks.
JOURNAL OF BUSINESS LOGISTICS
(2021)
Article
Chemistry, Multidisciplinary
Gaoyuan Cai, Juhu Li, Xuanxin Liu, Zhibo Chen, Haiyan Zhang
Summary: Recently, the deep neural network (DNN) has been widely used in classification tasks due to its advanced capabilities. However, the large number of parameters in DNN models can be costly. To address this issue, this paper proposes a joint optimization framework that combines the loss function and compression cost function to compress DNN models. By using the CUR decomposition method, low-rank approximation matrices are obtained to narrow the gap between weight matrices and compression results, achieving higher accuracy and compression ratios in image classification tasks.
APPLIED SCIENCES-BASEL
(2023)
Article
Computer Science, Software Engineering
Guohua Liu, Fei Li
Summary: This paper proposes a fabric defect detection method based on low-rank decomposition with structural constraints. The method extracts energy features and constructs a fusion image to highlight defective regions, then builds a new low-rank decomposition model with structured sparsity-inducing norm introduced, and obtain the defect detection result through thresholding the sparse part. Experimental comparisons show the superiority of the proposed method over several state-of-the-art fabric defect detection methods.
Article
Engineering, Electrical & Electronic
Wei-Hao Wu, Ting-Zhu Huang, Hao Zhang, Jian-Li Wang, Xi-Le Zhao
Summary: This paper proposes an untrained low-rank neural network prior (ULRNNP) for multi-dimensional image recovery, which has powerful representation ability and stable behavior. By using a nonlinear Tucker decomposition module, ULRNNP can design a friendly stopping criteria without the need for a reference ground truth image.
IEEE SIGNAL PROCESSING LETTERS
(2023)
Article
Computer Science, Interdisciplinary Applications
F. Allmann-Rahn, R. Grauer, K. Kormann
Summary: This study presents a new parallel low-rank solver for the full six-dimensional electromagnetic Vlasov-Maxwell equations, which can reduce the computational cost by utilizing distributed memory architectures, while ensuring the conservation of mass and a good representation of Gauss's law.
JOURNAL OF COMPUTATIONAL PHYSICS
(2022)
Article
Mathematics, Applied
Wentao Qu, Xianchao Xiu, Haifei Zhang, Jun Fan
Summary: With the development of technology and the era of big data, matrix minimization has become important and necessary due to the generation of complex data structures. This paper proposes and studies a novel low-rank and sparse regularized matrix minimization method, and develops an efficient semi-proximal alternating direction method of multipliers (sPADMM) from the dual perspective. The theoretical analysis proves the convergence of the sPADMM and its statistical properties, and the extensive numerical experiments verify the superiority of the proposed method in various applications.
JOURNAL OF COMPUTATIONAL AND APPLIED MATHEMATICS
(2023)
Article
Mathematics, Applied
Bo Feng, Gang Wu
Summary: The paper explores the eigenproblem on large and low-rank matrices, focusing on the relations between the Jordan decomposition and the Schur decomposition of small and large matrices. The proposed construction methods are not only theoretical but also practical, as demonstrated by numerical experiments.
APPLIED MATHEMATICS LETTERS
(2021)
Article
Neurosciences
Farras Abdelnour, Michael Dayan, Orrin Devinsky, Thomas Thesen, Ashish Raj
Summary: The study found that the relationship between brain function and structure in patients with temporal lobe epilepsy and in normal individuals is similar, suggesting that the brain reconfigures and rewires fine-scale connectivity under temporal lobe epilepsy conditions.
Article
Computer Science, Information Systems
Farhan Amin, Gyu Sang Choi
Summary: The Internet of Things has the potential to be applied to social networks due to innovative characteristics and solutions. Social network analysis has gained scientific attention and offers revolutionary theories in interconnected networks. This study aims to understand and capture clustering properties in large networks and social networks, proposing a network growth model and a scale-free artificial social network with controllable clustering coefficients.
CMC-COMPUTERS MATERIALS & CONTINUA
(2022)
Article
Computer Science, Information Systems
Jiang Liu, Xinyuan Zhang, Ran Zhang, Tao Huang, F. Richard Yu
Summary: Low earth orbit (LEO) small satellites have attracted great interests in civilian and military applications. This article introduces a hierarchical satellite network structure and clustering to enhance satellite network capabilities. A coalition game-theoretic framework and a distributed coalition formation algorithm are proposed to solve the small satellite clustering problem.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Environmental Studies
Giulia Reggiani, Trivik Verma, Winnie Daamen, Serge Hoogendoorn
Summary: Bicycle networks consist of various types of infrastructure, and understanding the different definitions and their impact is crucial for policymakers. This study analyzes the definitions of bicycle networks in 47 cities, examining scaling effects and network metrics to enhance knowledge for design interventions.
ENVIRONMENT AND PLANNING B-URBAN ANALYTICS AND CITY SCIENCE
(2023)
Article
Computer Science, Information Systems
Qian Sun, Gongxue Cheng, Xiaoyi Wang, Jiping Xu, Li Wang, Huiyan Zhang, Jiabin Yu, Ning Cao, Ruichao Wang
Summary: The paper proposes a routing optimization algorithm based on a small-world network model for water quality sensor networks to extend their lifecycle. The algorithm utilizes a short average path and large clustering coefficient to reduce energy consumption and improve fault-tolerance. Energy threshold and non-uniform clustering are constructed to enhance the network's lifecycle.
CMC-COMPUTERS MATERIALS & CONTINUA
(2021)
Article
Thermodynamics
Jun Zhao, Hassan Nasir Mangi, Zhenyue Zhang, Ru'an Chi, Haochen Zhang, Mengyu Xian, Hong Liu, Haibin Zuo, Guangwei Wang, Zhigao Xu, Ming Wu
Summary: This study aimed to utilize low-rank coal and biomass residue in the coking process, and investigated the performance and structural characteristics of the prepared cokes. The results showed that modified coal could improve the quality and caking performance of coke, as well as enhance the gasification performance.
Article
Neurosciences
Victor J. Barranca, Han Huang, Sida Li
COGNITIVE NEURODYNAMICS
(2019)
Article
Mathematical & Computational Biology
Victor J. Barranca, Han Huang, Genji Kawakita
JOURNAL OF COMPUTATIONAL NEUROSCIENCE
(2019)
Article
Mathematical & Computational Biology
Qing-long L. Gu, Songting Li, Wei P. Dai, Douglas Zhou, David Cai
FRONTIERS IN COMPUTATIONAL NEUROSCIENCE
(2019)
Article
Mechanics
Shixiao W. Jiang, Gregor Kovacic, Douglas Zhou, David Cai
JOURNAL OF FLUID MECHANICS
(2019)
Article
Neurosciences
Zhi-Qin John Xu, Xiaowei Gu, Chengyu Li, David Cai, Douglas Zhou, David W. McLaughlin
EUROPEAN JOURNAL OF NEUROSCIENCE
(2020)
Article
Behavioral Sciences
Talia Borofsky, Victor J. Barranca, Rebecca Zhou, Dora von Trentini, Robert L. Broadrup, Christopher Mayack
Article
Mathematics, Applied
Hong Cheng, David Cai, Douglas Zhou
Article
Neurosciences
Pamela B. Pyzza, Katherine A. Newhall, Gregor Kovacic, Douglas Zhou, David Cai
Summary: Research has shown that there are similar dynamical behaviors in the early olfactory pathway responses across different species when exposed to odors, which may be influenced by the time scales of fast excitation and fast and slow inhibition. By designing an ideal model and conducting numerical simulations, this hypothesis can be verified, and a firing-rate model can be derived to extract the structure of slow transition.
COGNITIVE NEURODYNAMICS
(2021)
Article
Computer Science, Artificial Intelligence
Victor J. Barranca
Summary: This study introduces a neural network framework for learning improved CS sampling based on the intrinsic structure present in classes of training signals, resulting in better CS signal reconstructions compared to uniformly random sampling. The learning methodology is purely data-driven and does not assume knowledge of any specific signal statistics.
Article
Neurosciences
Victor J. Barranca, Asha Bhuiyan, Max Sundgren, Fangzhou Xing
Summary: This study investigates the functional implications of network model dynamics that violate Dale's law, which states that a neuron transmits the same set of neurotransmitters at all of its post-synaptic connections. The results show that a single population network violating Dale's law can maintain balanced dynamics and produce effective decision-making dynamics in two competing pools of neurons. The study suggests that the one-population network exhibits more robust balanced activity for systems with fewer computational units, while the two-population network responds more rapidly to temporal variations in network inputs.
FRONTIERS IN NEUROSCIENCE
(2022)
Article
Mathematical & Computational Biology
Victor J. Barranca
Summary: This study develops a novel method for reverse-engineering the connectivity matrix of neuronal networks by utilizing the sparsity of neuronal connections. The researchers efficiently reconstruct the network connectivity and recover high dimensional natural stimuli from neuronal dynamics.
JOURNAL OF COMPUTATIONAL NEUROSCIENCE
(2023)
Article
Physics, Multidisciplinary
Katelyn Plaisier Leisman, Douglas Zhou, J. W. Banks, Gregor Kovacic, David Cai
Summary: A robust and spatiotemporally disordered family of waves has been found, in which waves with increasing amplitudes gradually evolve into weakly coupled collections of plane waves over long timescales. The amount of energy contained in their coupling decays to zero as the wave amplitude increases.
PHYSICAL REVIEW RESEARCH
(2022)
Article
Mathematics, Applied
Victor J. Barranca, Yolanda Hu, Zoe Porterfield, Samuel Rothstein, Alex Xuan
Summary: This research explores the mechanism for information preservation in neuronal networks across downstream layers of the brain by fitting a linear input-output mapping based on the widespread linearity of individual neuronal responses to strong ramped artificial inputs. Through analyzing different dynamics of neuronal network models and applying compressive sensing theory, sparse stimuli can be reconstructed efficiently using downstream neuronal firing rates, even in cases where theoretical analysis is challenging or governing equations are unknown.
SIAM JOURNAL ON APPLIED DYNAMICAL SYSTEMS
(2021)
Article
Mathematics, Applied
Victor J. Barranca, Gregor Kovacic, Douglas Zhou
COMMUNICATIONS IN MATHEMATICAL SCIENCES
(2019)
Article
Physics, Fluids & Plasmas
Zhi-Qin John Xu, Jennifer Crodelle, Douglas Zhou, David Cai