检索结果-内蒙古大学图书馆

Contrastive explanations in neural networks

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Prabhushankar, Mohit Kwon, Gukyeong Temel, Dogancan AlRegib, Ghassan OLIVES at the Center for Signal and Information Processing School of Electrical and Computer Engineering Georgia Institute of Technology AtlantaGA30332-0250 United States

Visual explanations are logical arguments based on visual features that justify the predictions made by neural networks. Current modes of visual explanations answer questions of the form ‘Why P?’. These Why questions operate under broad contexts thereby providing answers that are irrelevant in some cases. We propose to constrain these Why questions based on some context Q so that our explanations answer contrastive questions of the form ‘Why P, rather than Q?’. In this paper, we formalize the structure of contrastive visual explanations for neural networks. We define contrast based on neural networks and propose a methodology to extract defined contrasts. We then use the extracted contrasts as a plug-in on top of existing ‘Why P?’ techniques, specifically Grad-CAM. We demonstrate their value in analyzing both networks and data in applications of large-scale recognition, fine-grained recognition, subsurface seismic analysis, and image quality assessment. Copyright © 2020, The Authors. All rights reserved.

关键词： Deep learning

IMPLICIT SALIENCY in DEEP NEURAL NETWORKS

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Sun, Yutong Prabhushankar, Mohit AlRegib, Ghassan OLIVES Center for Signal and Information Processing School of Electrical and Computer Engineering Georgia Institute of Technology AtlantaGA30332-0250 United States

In this paper, we show that existing recognition and localization deep architectures, that have not been exposed to eye tracking data or any saliency datasets, are capable of predicting the human visual saliency. We term this as implicit saliency in deep neural networks. We calculate this implicit saliency using expectancy-mismatch hypothesis in an unsupervised fashion. Our experiments show that extracting saliency in this fashion provides comparable performance when measured against the state-of-art supervised algorithms. Additionally, the robustness outperforms those algorithms when we add large noise to the input images. Also, we show that semantic features contribute more than low-level features for human visual saliency detection. Based on these properties and performances, our proposed method greatly lowers the threshold for saliency detection in terms of required data and bridges the gap between human visual saliency and model saliency. Copyright © 2020, The Authors. All rights reserved.

关键词： Visualization

Machine Learning based Multiscale Reduced Kernel PCA for Nonlinear Process Monitoring

学校读者我要写书评

暂无评论

Machine Learning based Multiscale Reduced Kernel PCA for Non...

IEEE SSD International Multi-Conference on Systems, signals and Devices

作者： Khaled Dhibi Radhia Fezai Kais Bouzrara Majdi Mansouri Abdelmalek Kouadri Mohamed-Faouzi Harkat Laboratory of Automatic Signal and Image Processing School of Engineers of Monastir University of Monastir Tunisia Electrical and Computer Engineering Program Texas A&M University at Qatar Doha Qatar

ISBN: (数字)9781728110806

ISBN: (纸本)9781728110813

Fault detection (FD) is fundamental for monitoring several chemical processes. Thus, this paper introduces a novel structure multiscale reduced kernel principal component analysis (MS-RKPCA). The proposed FD method aims to address the problem of great computation time and significant storage memory space by using a data reduction structure based on the Euclidean distance metric. Additionally, to further enhance the RKPCA method, a multiscale representation of data will be used. The enhanced MS-RKPCA method uses the wavelet coefficients of the reduced data at each scale to enhance the fault detection results. The detection performance of the proposed MS-RKPCA method is evaluated using the Tennessee Eastman Process (TEP). The effectiveness of the enhanced method is evaluated in terms of the missed detection rates (MDR), false alarms rates (FAR) and computation time (CT). The results demonstrate that the developed technique is more effective for fault detection mostly in terms of computation time and memory storage space.

关键词： Fault detection Computational modeling Euclidean distance Wavelet analysis Kernel Principal component analysis Wavelet coefficients

Novelty detection through model-based characterization of neural networks

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Kwon, Gukyeong Prabhushankar, Mohit Temel, Dogancan AlRegib, Ghassan OLIVES at the Center for Signal and Information Processing School of Electrical and Computer Engineering Georgia Institute of Technology AtlantaGA30332-0250 United States

In this paper, we propose a model-based characterization of neural networks to detect novel input types and conditions. Novelty detection is crucial to identify abnormal inputs that can significantly degrade the performance of machine learning algorithms. Majority of existing studies have focused on activation-based representations to detect abnormal inputs, which limits the characterization of abnormality from a data perspective. However, a model perspective can also be informative in terms of the novelties and abnormalities. To articulate the significance of the model perspective in novelty detection, we utilize backpropagated gradients. We conduct a comprehensive analysis to compare the representation capability of gradients with that of activation and show that the gradients outperform the activation in novel class and condition detection. We validate our approach using four image recognition datasets including MNIST, Fashion-MNIST, CIFAR-10, and CURE-TSR. We achieve a significant improvement on all four datasets with an average AUROC of 0.953, 0.918, 0.582, and 0.746, respectively. Copyright © 2020, The Authors. All rights reserved.

关键词： Chemical activation

Cross-Layer Interference Mitigation for Cognitive Radio MIMO Systems

学校读者我要写书评

暂无评论

Cross-Layer Interference Mitigation for Cognitive Radio MIMO...

IEEE International Conference on Communications (ICC)

作者： Zengmao Chen Cheng-Xiang Wang Xuemin Hong John Thompson Sergiy A. Vorobyov Dongfeng Yuan Joint Research Institute of Signal and Image Processing Heriot-Watt University Edinburgh UK Joint Research Institute of Signal and Image Processing University of Edinburgh Edinburgh UK Department of Electrical and Computer Engineering University of Alberta Edmonton AB Canada School of Information Science and Engineering Shandong University Jinan Shandong China

In this paper, we investigate the interference mitigation from a cross-layer perspective for a cognitive radio (CR) multiple-input multiple-output (MIMO) network coexisting with a primary time-division-duplexing (TDD) system. The channel allocation in the media access control (MAC) layer and a subspace-based precoding scheme in the physical layer of the CR network are jointly considered to minimise the interference to the primary user and maximise the CR throughput. Two distributed cross-layer algorithms, namely, joint iterative channel allocation and precoding (JICAP) and non-iterative channel allocation and precoding (NICAP), are proposed for the cases with and without channel information among CR nodes, respectively. Moreover, a channel estimation scheme is also proposed to enable the NICAP. The effectiveness of the proposed algorithms over non-cross-layer counterpart is demonstrated via simulations.

关键词： Channel allocation Downlink Interference channels Channel estimation Covariance matrix Physical layer

image Captioning with Sparse Recurrent Neural Network

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Tan, Jia Huei Chan, Chee Seng Chuah, Joon Huang Center of Image and Signal Processing Faculty of Computer Science and Technology University of Malaya Kuala Lumpur50603 Malaysia Department of Electrical Engineering Faculty of Engineering University of Malaya Kuala Lumpur50603 Malaysia

Recurrent Neural Network (RNN) has been widely used to tackle a wide variety of language generation problems and are capable of attaining state-of-the-art (SOTA) performance. However despite its impressive results, the large number of parameters in the RNN model makes deployment in mobile and embedded devices infeasible. Driven by this problem, many works have proposed a number of pruning methods to reduce the sizes of the RNN model. In this work, we propose an end-to-end pruning method for image captioning models equipped with visual attention. Our proposed method is able to achieve sparsity levels up to 97.5% without significant performance loss relative to the baseline (∼ 2% loss at 40× compression after fine-tuning). Our method is also simple to use and tune, facilitating faster development times for neural network practitioners. We perform extensive experiments on the popular MS-COCO dataset in order to empirically validate the efficacy of our proposed method. Copyright © 2019, The Authors. All rights reserved.

关键词： Recurrent neural networks

Reduced Neural Network Based Ensemble Approach for Fault Detection and Diagnosis of Wind Energy Converter Systems

学校读者我要写书评

暂无评论

SSRN

SSRN 2022年

作者： Mansouri, Majdi Dhibi, Khaled Bouzrara, Kais Nounou, Hazem Nounou, Mohamed Electrical and Computer Engineering Program Texas A&M University Qatar Qatar Laboratory of Automatic Signal and Image Processing National School of Engineers of Monastir University of Monastir 5019 Tunisia

Fault Detection and Diagnosis (FDD) is very important to achieve the best operation and ensure the safe and continuous operation of wind energy conversion (WEC) systems. Over the last decade, artificial neural networks surpass classical machine learning models in a variety of FDD tasks. On the other hand, ensemble learning (EL) techniques have gained significant attention from the scientific community. EL is a technique that creates and combines multiple machine learning models in order to produce one optimal predictive model which gives improved results. The goal of this paper is to develop and validate effective neural networks based ensemble approach. First, an ensemble classifier based on neural networks techniques and using bagging, boosting, and random subspace combination techniques is proposed. Second, an improved extension of the proposed neural networks-based ensemble technique is presented. The main idea behind this proposal is to determine variable relationships via a data-reduction framework using Hierarchical K-means (H-K-means) clustering method in order to overcome the problem of computation time and storage cost. Finally, the results obtained from the proposed neural networks-based ensemble techniques are compared with other ensemble and single machine learning methods to illustrate and validate the advantages of the proposed techniques. © 2022, The Authors. All rights reserved.

关键词： Neural networks

Fault detection of uncertain nonlinear process using interval-valued data-driven approach

学校读者我要写书评

暂无评论

Fault detection of uncertain nonlinear process using interva...

IEEE SSD International Multi-Conference on Systems, signals and Devices

作者： Khaled Dhibi Radhia Fezai Kais Bouzrara Majdi Mansouri Abdelmalek Kouadri Mohamed-Faouzi Harkat Laboratory of Automatic Signal and Image Processing National School of Engineers of Monastir University of Monastir Tunisia Electrical and Computer Engineering Program Texas A&M University at Qatar Doha Qatar

ISBN: (数字)9781728110806

ISBN: (纸本)9781728110813

Kernel PCA (KPCA) has been extensively applied in fault detection (FD) field. However, it is constantly not optimal for uncertain systems and is not designed to handle large-scale process monitoring. Thus, a nonlinear fault detection (FD) method based interval reduced KPCA (IRKPCA) is developed for fault detection. The proposed IRKPCA technique uses interval-valued Euclidean distance as a criterion to maintain only the more pertinent measurements. The FD abilities of the IRKPCA technique is assessed using the Tennessee Eastman Process (TEP). The effectiveness of the proposed technique is assessed in terms of computation time (CT), false alarm rate (FAR)and missed detection rate (MDR).

关键词： Uncertain systems Uncertainty Fault detection Tools Nonlinear systems Kernel Principal component analysis

Machine Learning based Reduced Kernel PCA for Nonlinear Process Monitoring

学校读者我要写书评

暂无评论

Machine Learning based Reduced Kernel PCA for Nonlinear Proc...

International Conference on Internet of Things, Embedded Systems and Communications (IINTEC)

作者： Khaled Dhibi Radhia Fezai Majdi Mansouri Abdelmalek Kouadri Kais Bouzrara Laboratory of Automatic Signal and Image Processing National School of Engineers of Monastir University of Monastir Tunisia Electrical and Computer Engineering Program Texas A&M University at Qatar Doha QATAR

ISBN: (数字)9781728151847

ISBN: (纸本)9781728151854

Process monitoring is an essential part of industrial systems. It requires higher product quality and safety operations. Therefore, a nonlinear data-driven approach based reduced KPCA (RKPCA) for statistical monitoring of industrial processes is developed. RKPCA is a novel machine learning tool which merges dimensionality reduction and supervised learning. The use of classical KPCA for modeling and monitoring purposes can impose a high computational load when a large number of measurements are recorded. The main idea of the proposed RKPCA approach is to reduce the number of observations (samples) in the data matrix using the Euclidean distance between samples as dissimilarity metric so that only one observation is kept in case of redundancy. The Tennessee Eastman Process (TEP) is used to evaluate the fault detection abilities of the proposed RKPCA technique. The performance of the proposed method is evaluated and compared to the classical KPCA in terms of false alarms rates (FAR), missed detection rates (MDR) and computation times (CT).

关键词：