检索结果-内蒙古大学图书馆

Decoding sensorimotor information from superior parietal lobule of macaque via Convolutional neural networks

neural networkS 2022年第0期151卷 276-294页

作者： Filippini, Matteo Borra, Davide Ursino, Mauro Magosso, Elisa Fattori, Patrizia Univ Bologna Dept Biomed & Neuromotor Sci Piazza Porta San Donato 2 I-40126 Bologna Italy Univ Bologna Dept Elect Elect & Informat Engn Guglielmo Marcon Cesena Campus Cesena Italy Alma Mater Res Inst Human Ctr Artificial Intellig Bologna Italy

Despite the well-recognized role of the posterior parietal cortex (PPC) in processing sensory information to guide action, the differential encoding properties of this dynamic processing, as operated by different PPC brain areas, are scarcely known. Within the monkey's PPC, the superior parietal lobule hosts areas V6A, PEc, and PE included in the dorso-medial visual stream that is specialized in planning and guiding reaching movements. Here, a Convolutional neural network (CNN) approach is used to investigate how the information is processed in these areas. We trained two macaque monkeys to perform a delayed reaching task towards 9 positions (distributed on 3 different depth and direction levels) in the 3D peripersonal space. The activity of single cells was recorded from V6A, PEc, PE and fed to convolutional neural networks that were designed and trained to exploit the temporal structure of neuronal activation patterns, to decode the target positions reached by the monkey. Bayesian Optimization was used to define the main CNN hyper-parameters. In addition to discrete positions in space, we used the same network architecture to decode plausible reaching trajectories. We found that data from the most caudal V6A and PEc areas outperformed PE area in the spatial position decoding. In all areas, decoding accuracies started to increase at the time the target to reach was instructed to the monkey, and reached a plateau at movement onset. The results support a dynamic encoding of the different phases and properties of the reaching movement differentially distributed over a network of interconnected areas. This study highlights the usefulness of neurons' firing rate decoding via CNNs to improve our understanding of how sensorimotor information is encoded in PPC to perform reaching movements. The obtained results may have implications in the perspective of novel neuroprosthetic devices based on the decoding of these rich signals for faithfully carrying out patient's intentio

关键词： neural decoding Posterior parietal cortex Convolutional neural network Sensorymotor transformation Brain-computer interfaces Monkey

来源：评论

学校读者我要写书评

暂无评论

CUTTANA: Scalable Graph Partitioning for Faster distributed Graph Databases and Analytics

引用

PROCEEDINGS OF THE VLDB ENDOWMENT 2024年第1期18卷 14-27页

作者： Hajidehi, Milad Rezaei Sridhar, Sraavan Seltzer, Margo Univ British Columbia Vancouver BC Canada

Graph partitioning plays a pivotal role in various distributed graph processing applications, including graph analytics, graph neural network training, and distributed graph databases. A "good" graph partitioner reduces workload execution time, worker imbalance, and network overhead. Graphs that require distributed settings are often too large to fit in the main memory of a single machine. This challenge renders traditional in-memory graph partitioners infeasible, leading to the emergence of streaming solutions. Streaming partitioners produce lower-quality partitions, because they work from partial information and must make premature decisions before they have a complete view of a vertex's neighborhood. We introduce CUTTANA, a streaming graph partitioner that partitions massive graphs (Web/Twitter scale) with superior quality compared to existing streaming solutions. CUTTANA uses a novel buffering technique that prevents the premature assignment of vertices to partitions and a scalable coarsening and refinement technique that enables a complete graph view, improving the intermediate assignment made by a streaming partitioner. We implemented a parallel version for CUTTANA that offers nearly the same partitioning latency as existing streaming partitioners. Our experimental analysis shows that CUTTANA consistently yields better partitioning quality than state-of-the-art streaming vertex partitioners in terms of both edge-cut and communication volume metrics. We also evaluate the workload latencies that result from using CUTTANA and other partitioners in distributed graph analytics and databases. CUTTANA outperforms the other methods in most scenarios (algorithms, datasets). In analytics applications, CUTTANA improves runtime performance by up to 59% compared to various streaming partitioners (i.e., HDRF, Fennel, Ginger, HeiStream). In graph database tasks, CUTTANA results in higher query throughput by up to 23%, without hurting tail latency.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Does Worst-Performing Agent Lead the Pack? Analyzing Agent Dynamics in Unified distributed SGD 38

Does Worst-Performing Agent Lead the Pack? Analyzing Agent D...

引用

38th Conference on neural Information processing Systems, NeurIPS 2024

作者： Hu, Jie Ma, Yi-Ting Eun, Do Young Department of Electrical and Computer Engineering North Carolina State University United States

distributed learning is essential to train machine learning algorithms across heterogeneous agents while maintaining data privacy. We conduct an asymptotic analysis of Unified distributed SGD (UD-SGD), exploring a variety of communication patterns, including decentralized SGD and local SGD within Federated Learning (FL), as well as the increasing communication interval in the FL setting. In this study, we assess how different sampling strategies, such as i.i.d. sampling, shuffling, and Markovian sampling, affect the convergence speed of UD-SGD by considering the impact of agent dynamics on the limiting covariance matrix as described in the Central Limit Theorem (CLT). Our findings not only support existing theories on linear speedup and asymptotic network independence, but also theoretically and empirically show how efficient sampling strategies employed by individual agents contribute to overall convergence in UD-SGD. Simulations reveal that a few agents using highly efficient sampling can achieve or surpass the performance of the majority employing moderately improved strategies, providing new insights beyond traditional analyses focusing on the worst-performing agent. © 2024 neural information processing systems foundation. All rights reserved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Comparison of Deep-neural-network-Based Models for Estimating distributed Representations of Compound Words 25

Comparison of Deep-Neural-Network-Based Models for Estimatin...

引用

25th KES International Conference on Knowledge-Based and Intelligent Information & Engineering Systems (KES)

作者： Dao, An Kertkeidkachorn, Natthawut Ichise, Ryutaro Univ Tokyo Tokyo Japan Japan Adv Inst Sci & Technol Nomi Japan Natl Inst Informat Tokyo Japan Natl Inst Adv Ind Sci & Technol Tokyo Japan

Word embeddings or word vectors have become fundamental in language processing techniques, especially deep learning approaches. Although many languages have compound words (e.g., "robot arm" and "maple leaf"), such words have not received much attention from researchers. Most research on compound word embeddings considered only two-word compounds;there has been little detailed analysis on the learning representations of arbitrary-length compound words. This paper discusses the necessity for learning-based approaches for estimating the distributed representations of compound words instead of a simple average of the representations of constituents. An evaluation of two downstream tasks confirms the effectiveness of compositional models in encoding useful information into vector spaces. The experimental results suggest that complex architectures such as long short-term memory, gated recurrent units, and transformers learn better representations for long entities, whereas simpler models such as recurrent neural networks are more applicable for downstream tasks where there are only short compounds (two or three words in length), as in the noun compound interpretation task. (C) 2021 The Authors. Published by Elsevier B.V.

关键词： compound word distributional models of semantics multi-word expressions

来源：评论

学校读者我要写书评

暂无评论

Emergence of Shape Bias in Convolutional neural networks through Activation Sparsity 37

Emergence of Shape Bias in Convolutional Neural Networks thr...

引用

37th Conference on neural Information processing Systems (NeurIPS)

作者： Li, Tianqin Wen, Ziqi Li, Yangfan Lee, Tai Sing Carnegie Mellon Univ Pittsburgh PA 15213 USA Nortwestern Univ Evanston IL USA

ISBN: (纸本)9781713899921

Current deep-learning models for object recognition are known to be heavily biased toward texture. In contrast, human visual systems are known to be biased toward shape and structure. What could be the design principles in human visual systems that led to this difference? How could we introduce more shape bias into the deep learning models? In this paper, we report that sparse coding, a ubiquitous principle in the brain, can in itself introduce shape bias into the network. We found that enforcing the sparse coding constraint using a non-differential Top-K operation can lead to the emergence of structural encoding in neurons in convolutional neural networks, resulting in a smooth decomposition of objects into parts and subparts and endowing the networks with shape bias. We demonstrated this emergence of shape bias and its functional benefits for different network structures with various datasets. For object recognition convolutional neural networks, the shape bias leads to greater robustness against style and pattern change distraction. For the image synthesis generative adversary networks, the emerged shape bias leads to more coherent and decomposable structures in the synthesized images. Ablation studies suggest that sparse codes tend to encode structures, whereas the more distributed codes tend to favor texture. Our code is host at the github repository: https://***/

关键词： Object recognition

来源：评论

学校读者我要写书评

暂无评论

On the Role of Noise in the Sample Complexity of Learning Recurrent neural networks: Exponential Gaps for Long Sequences 37

On the Role of Noise in the Sample Complexity of Learning Re...

引用

37th Conference on neural Information processing Systems (NeurIPS)

作者： Pour, Alireza F. Ashtiani, Hassan Univ Waterloo Cheriton Sch Comp Sci Waterloo ON Canada McMaster Univ Dept Comp & Software Hamilton ON Canada

ISBN: (纸本)9781713899921

We consider the class of noisy multi-layered sigmoid recurrent neural networks with w (unbounded) weights for classification of sequences of length T, where independent noise distributed according to N(0, sigma(2)) is added to the output of each neuron in the network. Our main result shows that the sample complexity of PAC learning this class can be bounded by O(w log(T/sigma)). For the non-noisy version of the same class (i.e., sigma = 0), we prove a lower bound of Omega(wT) for the sample complexity. Our results indicate an exponential gap in the dependence of sample complexity on T for noisy versus non-noisy networks. Moreover, given the mild logarithmic dependence of the upper bound on 1/sigma, this gap still holds even for numerically negligible values of sigma.(1)

关键词： Recurrent neural networks

来源：评论

学校读者我要写书评

暂无评论

A new health indicator extracted by unsupervised learning using autoencoder in tandem with t-sne and multi-kernel CNN to enhance the early detection and classification of bearings multi-faults

引用

JOURNAL OF THE BRAZILIAN SOCIETY OF MECHANICAL SCIENCES AND ENGINEERING 2023年第7期45卷 1-14页

作者： Zair, Mohamed Rahmoune, Chemseddine Imane, Moussaoui Amine, Mahami Benazzouz, Djamel Univ Mhamed Bougara Boumerdes Solid Mech & Syst Lab LMSS Boumerdes 35000 Algeria

The conventional process diagnostic scheme comprises data acquisition, feature extraction, and fault classification. However, traditional feature extraction uses signal processing technologies that are deeply reliant on both subjectivity and foreknowledge. However, these techniques have shown some limitations that are affecting efficiency and effectiveness, especially for varying fault and noisy environments. To address these issues, this paper proposes a practical and effective unsupervised deep learning methodology based on autoencoder (AE) in tandem with t-distributed stochastic neighbor embedding (t-sne) and multi-kernel convolutional neural network. In this approach, the raw data are first extracted in the time domain, normalizing and segmenting the vibration signal into small portions by sliding window to keep more information data. Subsequently, during autoencoder (AE) training, the dropout smoothing and batch normalization are used to avoid overfitting and to extract the deep features of the retread form from the normalized data set in the time domain. Then, the nonlinear mapping obtained in the high-dimensional data is reduced using the t-sne algorithm by deleting redundant and insignificant parameters that can confuse the classification. Finally, the measurement with low-dimensional feature vectors is selected as inputs of the deep structure of multi-kernel CNN for automatic fault detection and classification. Comparative studies were implemented with several techniques including CNN, KNN and SVM to diagnose and differentiate different types of defects in bearings. The results showed that the proposed approach is effective for bearings in terms of predictive accuracy with higher accuracy 99.46% compared to conventional methods.

关键词： Rolling bearings Autoencoder Visualization data Faults diagnosis t-distributed stochastic neighbor embedding (t-sne) Fault classification Multi-kernel conventional neural network (CNN)

来源：评论

学校读者我要写书评

暂无评论

Pipeline Leakage Event Recognition in distributed Optical Fiber Sensing Using One-Dimensional Convolutional neural network with Fusion Input of Features and Multi-Parametric Signals

引用

Guangxue Xuebao/Acta Optica Sinica 2025年第10期45卷

作者： Song, Muping Jia, Ning Cui, Enxue College of Information Science and Electronic Engineering Zhejiang University Zhejiang Hangzhou310027 China

Objective distributed optical fiber sensors (DOFSs) have attracted significant attention in recent years due to their capabilities for real-time, distributed, and long-distance sensing. They can be applied to pipelines, cables, tunnels, and other scenarios. DOFSs detect Rayleigh, Raman, and Brillouin scattered light to sense various parameters sitch as Vibration, temperature, and strain. In practical event recognition applications, multiple parameters are often affected simultaneously. The signals collected by different DOFS Systems have different features in the time and frequency domains. Vibration signals are dynamic, change rapidly, and require high sampling rates, while temperature signals are relatively static and change slowly over longer time. Therefore, appropriate signal processing methods are essential for accurate event recognition. Conventional signal processing methods typically rely on a single sensing parameter. When the signal-to-noise ratio (SNR) decreases or event features change, such methods may confuse events with interference signals of similar characteristics, leading to misidentification and reduced recognition accuracy. To address this, combimng multiple DOFS Systems and multi-parametric signals is an effective method. However, there are some technical challenges, such as time scale mismatches between multi-parametric signal types and the extraction of appropriate features. To overcome these issues, we propose a one-dimensional convolutional neural network (ld-CNN) that takes both extracted features and dynamic/static multi-parametric signals as input. This approach enables effective signal fusion and improves event recognition accuracy compared to traditional *** We propose a multi-parametric event recognition method based on ld-CNN, integrating both extracted features and multi-parametric signals. The Vibration signals are pre-processed and used as dynamic inputs, while the temperature Signals are pre-processed and used as sta

关键词： Cochannel interference

来源：评论

学校读者我要写书评

暂无评论

FedGraphMRI-net: A federated graph neural network framework for robust MRI reconstruction across non-IID data

引用

BIOMEDICAL SIGNAL processing AND CONTROL 2025年 102卷

作者： Ahmed, Shahzad Jinchao, Feng Manan, Malik Abdul Yaqub, Muhammad Ali, Muhammad Usman Raheem, Abdul Beijing Univ Technol Fac Informat Technol Beijing Key Lab Computat Intelligence & Intelligen Beijing 100124 Peoples R China Hunan Univ Sch Biomed Sci Changsha Peoples R China COMSATS Univ Islamabad Dept Comp Sci Sahiwal Campus Sahiwal Pakistan Beijing Univ Technol Fac Informat Technol Beijing 100124 Peoples R China

Accurate MRI reconstruction from undersampled k-space data is essential in medical imaging. Still, it is often dependent on conditional models closely tied to specific imaging operators, which limits their adaptability to different imaging protocols and equipment. This dependence leads to suboptimal performance under varying conditions. Centralized approaches also pose data privacy concerns, as they require data sharing across institutions. To address these challenges, we introduce FedGraphMRI-Net, a federated learning framework specifically designed for MRI reconstruction in non-Independent and Identically distributed (non-IID) settings. Our approach leverages graph-based neural networks to capture both local and global anatomical correlations, ensuring patient privacy and adaptability to diverse, site-specific data distributions. FedGraphMRI-Net employs a graph clustering strategy via the Louvain algorithm to partition global MRI data into sub-graphs, each representing localized anatomical features and spatial relationships. Experimental results demonstrate that FedGraphMRI-Net achieves superior MRI reconstruction performance, obtaining PSNR scores of 43.8 f 1.1, 44.1 f 1.0, and 45.0 f 1.1 dB, and SSIM values of 98.5 f 0.2 %, 98.3 f 0.2 %, and 98.8 f 0.1 % for T1, T2, and PDweighted scans on the IXI dataset. On the fastMRI dataset with 4x acceleration, the model achieved PSNR scores of 42.0 f 1.5, 40.8 f 1.4, and 43.2 f 1.6 dB, along with SSIM values of 98.5 f 0.4 %, 97.9 f 0.3 %, and 98.8 f 0.1 % for T1, T2, and FLAIR scans. FedGraphMRI-Net outperforms state-of-the-art models in cross-domain generalization and high acceleration scenarios, offering a robust, scalable, and privacy-preserving MRI reconstruction solution for varied clinical environments.

关键词： Federated learning MRI reconstruction Graph neural network Deep learning Privacy-preserving Medical imaging

来源：评论

学校读者我要写书评

暂无评论

Post-processing of hydrological model simulations using the convolutional neural network and support vector regression

引用

HYDROLOGY RESEARCH 2022年第4期53卷 605-621页

作者： Liu, Songnan Wang, Jun Wang, Huijun Wu, Yuetao Nanjing Univ Informat Sci & Technol Collaborat Innovat Ctr Forecast & Evaluat Meteoro Minist Educ Nanjing 210044 Jiangsu Peoples R China Nanjing Univ Informat Sci & Technol Key Lab Meteorol Disaster Minist Educ Nanjing 210044 Jiangsu Peoples R China Chinese Acad Sci Nansen Zhu Int Res Ctr Inst Atmospher Phys Beijing 100029 Peoples R China

Post-processing methods can be used to reduce the biases of hydrological models. In this research, six post-processing methods are compared: quantile mapping (QM) methods, which include four kinds of transformations, and two newly established machine learning frameworks [support vector regression (SVR) and convolutional neural network (CNN)] based on meteorological data and variation mode decomposition (VMD)-decomposed streamflow. These post-processing methods are applied to a distributed model (WRF-Hydro), and the evaluation is carried out over five watersheds with different areas in South China. The post-processing methods are separately applied to calibrated and uncalibrated models. The results show that the SVR- and CNN-based post-processing methods perform better than the QM methods in terms of daily streamflow simulations in different areas with different topographies in the Xijiang River basin. There are large uncertainties in the QM post-processing methods. The CNN-based post-processing performs slightly better than the SVR-based post-processing, but both methods can markedly improve the simulated streamflow. The CNN- and SVR-based post-processing frameworks are suitable for both calibration and test periods. The differences between post-processing with uncalibrated and calibrated models are quite small for SVR- and CNN-based post-processing, but large for QM post-processing. For WRF-Hydro, the CNN- and SVR-based post-processing methods consume much less time and computational resources than model calibration.

关键词： convolutional neural network machine learning post-processing support vector regression WRF-Hydro

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：