检索结果-内蒙古大学图书馆

38th Conference on Neural Information Processing Systems, NeurIPS 2024

作者： Dhahri, Rayen Immer, Alexander Charpentier, Betrand Günnemann, Stephan Fortuin, Vincent Department of Computer Science Technical University of Munich Germany Munich Center for Machine Learning Germany Department of Computer Science ETH Zürich Switzerland Max Planck Institute for Intelligent Systems Tübingen Germany Pruna AI Munich Germany Helmholtz AI Munich Germany

Neural network sparsification is a promising avenue to save computational time and memory costs, especially in an age where many successful AI models are becoming too large to naïvely deploy on consumer hardware. While much work has focused on different weight pruning criteria, the overall sparsifiability of the network, i.e., its capacity to be pruned without quality loss, has often been overlooked. We present Sparsifiability via the Marginal likelihood (SpaM), a pruning framework that highlights the effectiveness of using the Bayesian marginal likelihood in conjunction with sparsity-inducing priors for making neural networks more sparsifiable. Our approach implements an automatic Occam's razor that selects the most sparsifiable model that still explains the data well, both for structured and unstructured sparsification. In addition, we demonstrate that the pre-computed posterior precision from the Laplace approximation can be re-used to define a cheap pruning criterion, which outperforms many existing (more expensive) approaches. We demonstrate the effectiveness of our framework, especially at high sparsity levels, across a range of different neural network architectures and datasets. © 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

HGAMLP: Heterogeneous Graph Attention MLP with De-Redundancy Mechanism 40

HGAMLP: Heterogeneous Graph Attention MLP with De-Redundancy...

引用

40th IEEE International Conference on Data Engineering, ICDE 2024

作者： Liang, Yuxuan Zhang, Wentao Sheng, Zeang Yang, Ling Jiang, Jiawei Tong, Yunhai Cui, Bin School of Intelligence Science and Technology Peking University China Center for Machine Learning Research Peking University China School of Computer Science Wuhan University China Peking University China China

ISBN: (纸本)9798350317152

Heterogeneous graphs contain rich semantic information that can be exploited by heterogeneous graph neural networks (HGNNs). However, scaling HGNNs to large graphs is challenging due to the high computational cost. Existing non-parametric HGNNs use general subgraphs construction method and mean aggregator before training to reduce the complexity. Despite their success, they ignore two key characteristics of heterogeneous graphs, leading to low predictive performance. First, they adopt fixed local and global knowledge extractor for the feature aggregation and the semantic fusion. Besides, they bury the graph structure information of the higher-order meta-paths and fail to explore deeper graph structure information. In this paper, we address these two limitations and propose a new non-parametric HGNN framework called Heterogeneous Graph Attention Multi-Layer Perceptron (HGAMLP). Our framework employs the local multi-knowledge extractor to enhance the node representation, and leverages the de-redundancy mechanism to extract the pure graph structure information from higher-order meta-paths. Besides, it adopts a node-adaptive weight adjustment mechanism as an efficiency training model to fuse global knowledge and local knowledge. We evaluate our framework on ten commonly used heterogeneous graph datasets and show that it outperforms the state-of-the-art baselines in both accuracy and speed. Notably, our framework achieves the best performance on the large public heterogeneous graph dataset (i.e., Ogbn-mag) of Open Graph Benchmark. https://***/docs/leader-nodeprop. © 2024 IEEE.

关键词： Graph neural networks

来源：评论

学校读者我要写书评

暂无评论

Understanding adversarial attacks on observations in deep reinforcement learning

引用

science China(Information sciences) 2024年第5期67卷 69-83页

作者： You QIAOBEN Chengyang YING Xinning ZHOU Hang SU Jun ZHU Bo ZHANG Department of Computer Science and Technology Beijing National Research Center for Information Science and Technology Tsinghua-Bosch Joint Center for Machine Learning Institute for Artificial Intelligence Tsinghua University Peng Cheng Laboratory

Deep reinforcement learning models are vulnerable to adversarial attacks that can decrease the cumulative expected reward of a victim by manipulating its observations. Despite the efficiency of previous optimization-based methods for generating adversarial noise in supervised learning, such methods might not achieve the lowest cumulative reward since they do not generally explore the environmental ***, a framework is provided to better understand the existing methods by reformulating the problem of adversarial attacks on reinforcement learning in the function space. The reformulation approach adopted herein generates an optimal adversary in the function space of targeted attacks, repelling them via a generic two-stage framework. In the first stage, a deceptive policy is trained by hacking the environment and discovering a set of trajectories routing to the lowest reward or the worst-case performance. Next, the adversary misleads the victim to imitate the deceptive policy by perturbing the observations. Compared to existing approaches, it is theoretically shown that our adversary is strong under an appropriate noise level. Extensive experiments demonstrate the superiority of the proposed method in terms of efficiency and effectiveness, achieving state-of-the-art performance in both Atari and MuJoCo environments.

关键词： deep learning reinforcement learning adversarial robustness adversarial attack

来源：评论

学校读者我要写书评

暂无评论

Deep Bayesian Image Set Classification Approach for Defense against Adversarial Attacks

Deep Bayesian Image Set Classification Approach for Defense ...

引用

2023 International Conference on Digital Image Computing: Techniques and Applications, DICTA 2023

作者： Mirnateghi, Nima Shah, Syed Afaq Ali Bennamoun, Mohammed Edith Cowan University Center for Ai and Machine Learning Australia The University of Western Australia School of Computer Science and Software Engineering Australia

ISBN: (纸本)9798350382204

Deep learning has become an integral part of various pattern recognition and computer vision systems in recent years due to its outstanding achievements in object recognition, facial recognition, and scene understanding. However, deep neural networks (DNNs) are susceptible to being fooled with nearly high confidence by an adversary. In practice, the vulnerability of deep learning systems against carefully perturbed images, known as adversarial examples, poses a dire security threat in physical-world applications. To address this phenomenon, we present, what to our knowledge, is the first ever image-set-based adversarial defense approach. Image set classification has shown exceptional performance for object and face recognition, owing to its intrinsic property of handling appearance variability. We propose a robust deep Bayesian image set classification as a defense framework against a broad range of adversarial attacks. We extensively experiment the performance of the proposed technique with several voting strategies. We further analyse the effects of image size, perturbation magnitude, along with the ratio of perturbed images in each image set. We also evaluate our technique with the recent state-of-the-art defense methods and single-shot recognition task. The empirical results demonstrate superior performance on the CIFAR-10, MNIST, ETH-80, and Tiny ImageNet datasets. Our code is available at https://***/ai-voyage/***. © 2023 IEEE.

关键词： Object recognition

来源：评论

学校读者我要写书评

暂无评论

End-to-End Hyperspectral Image Classification Using Hybrid GCNN with SWIN Transformer 1

End-to-End Hyperspectral Image Classification Using Hybrid G...

引用

1st IEEE International Conference on Ambient Intelligence, Knowledge Informatics and Industrial Electronics, AIKIIE 2023

作者： Ghotekar, Rahul Shaw, Kailash Rout, Minakhi Kiit University Department of Computer Science and Engineering Odisha India Department of Artificial Intelligence and Machine Learning Maharashtra Pune Lavale India

ISBN: (纸本)9798350316469

Hyperspectral image (HSI) classification holds immense significance in remote sensing applications like land cover mapping and environmental monitoring. This study introduces Graph Convolutional Neural Networks (GCNN) with the SWIN Transformer model (G_SwTr) to enhance HSI classification accuracy. GCNNs are utilized to incorporate spatial information through graph representations, capturing spectral-spatial dependencies. Graph convolutions, effectively harnesses spectral and spatial attributes, further the integration of the SWIN Transformer, recognized for its proficiency in capturing long-range dependencies via shifted windows and hierarchical processing, elevates feature extraction and contextual understanding. The Proposed G_SwTr model is evaluated across three HSI datasets (Pavia-University, Indian-Pines, and Salinas) and is compared against traditional machine learning and Deep learning methods. The results achieved outperform state-of-art methods, with an outstanding Overall Accuracy (O.A) of 99.7, 99.9, and 99.5, an average accuracy (A.A) of 99.5, 99.6, and 98.2, and a kappa coefficient (K) of 99.5, 99.6, and 98.3 for all three datasets. © 2023 IEEE.

关键词： Deep learning Graph Convolution Neural Network Hybrid Model Hyperspectral Image Classification SWIN Transformer

来源：评论

学校读者我要写书评

暂无评论

SightAid: empowering the visually impaired in the Kingdom of Saudi Arabia (KSA) with deep learning-based intelligent wearable vision system

引用

Neural Computing and Applications 2024年第19期36卷 11075-11095页

作者： Talaat, Fatma M. Farsi, Mohammed Badawy, Mahmoud Elhosseini, Mostafa Machine Learning Department Faculty of Artificial Intelligence Kafrelsheikh University Kafrelsheikh Egypt Faculty of Computer Science & amp Engineering New Mansoura University Gamasa35712 Egypt College of Computer Science and Engineering Taibah University Yanbu46421 Saudi Arabia Computers and Control Systems Engineering Department Faculty of Engineering Mansoura University Mansoura35516 Egypt Computer Science and Information Department Applied College Taibah University Madinah46537 Saudi Arabia

In the Kingdom of Saudi Arabia, visual impairment poses significant challenges for approximately 17.5% of school-aged children, mainly due to refractive errors. These challenges extend to everyday navigation, environmental interaction, and overall life quality. Motivated by the desire to empower visually impaired individuals, who face navigational limitations, difficulties in object recognition, and inadequate assistance from traditional technologies, we propose SightAid. This innovative wearable vision system utilizes a deep learning-based framework, addressing the gaps left by current assistive solutions. Traditional methods, such as canes and GPS devices, often fail to meet the nuanced and dynamic needs of the visually impaired, especially in accurately identifying objects, understanding complex environments, and providing essential real-time feedback for independent navigation. SightAid comprises a seven-phase framework involving data collection, preprocessing, and training of a sophisticated deep neural network with multiple convolutional and fully connected layers. This system is integrated into smart glasses with augmented reality displays, enabling real-time object detection and recognition. Interaction with users is facilitated through audio or haptic feedback, informing them about the location and type of objects detected. A continuous learning mechanism, incorporating user feedback and new data, ensures the system's ongoing refinement and adaptability. For performance assessment, we utilized the MNIST dataset, and an Indoor Objects Detection dataset tailored for the visually impaired, featuring images of everyday objects crucial for safe indoor navigation. SightAid demonstrates remarkable performance with accuracy up to 0.9874, recall values between 0.98 and 0.99, F1-scores ranging from 0.98 to 0.99, and AUC-ROC values reaching as high as 0.9999. These metrics significantly surpass those of traditional methods, highlighting SightAid's potential to substan

关键词： Object recognition

来源：评论

学校读者我要写书评

暂无评论

Performance Analysis of Acoustic Scene Classification Using ANN and CNN Techniques

Performance Analysis of Acoustic Scene Classification Using ...

引用

2023 International Conference on Integrated Intelligence and Communication Systems, ICIICS 2023

作者： Rakshitha Karthik, K. Shetty, Aishwarya D Sowmya, P. Shettigar, Rashmitha Department of Artificial Intelligence & Machine Learning Nitte India School of Computer Science and Engineering Vellore India Department of Computer Science and Engineering Nitte India /CS Dhaisar Mumbai India

ISBN: (纸本)9798350315455

Artificial neural networks have achieved notable improvements in various learning tasks and have also been used to classify environmental sounds. The technique for allowing devices to recognize and interpret the surroundings from a recording is called Acoustic Scene Classification (ASC). Computational Auditory/Scene analysis is analyzing scenes using computational techniques. This research paper's primary goal is to categorize audio recordings based on the scenarios or environments in which they were captured. In most applications, deep learning is one of the most recent trends. In this work, we have concentrated on two deep learning techniques employed to classify audio scenes: Convolution Neural Networks (CNN) and Artificial Neural Networks (ANN) to develop the model. ReLU and LeakyReLU activation functions are used to train and validate the model. For evaluating the model performance, K-fold cross-validation was used with K=10 and obtained an accuracy of 75.8% using an Artificial Neural Network and a proposed model with 90.6% accuracy using a Convolutional Neural Network. Comparing the results with other baseline models that achieved a performance of 73.3% for MCLNN with Mel-spectrogram, 78.2% for Pyramid-Combined CNN, 78.3% for Envnet-v2, and 81.9% for D-CNN. © 2023 IEEE.

关键词： Acoustic Scene Classification ANN CNN Environmental Sound Recognition

来源：评论

学校读者我要写书评

暂无评论

User Experience with Grammarly's Generative AI: Ethical Implications for Improving Writing Skills 10

User Experience with Grammarly's Generative AI: Ethical Impl...

引用

10th International Conference on Education and Technology, ICET 2024

作者： Zunaidah, Asih Wiharja, Chandra Kurniawan Wicaksono, Danang Wahyu Bina Nusantara University Digital Language Learning Center Faculty of Humanities Communication Science Department Jakarta Indonesia School of Computer Science Bina Nusantara University Computer Science Department Jakarta Indonesia

ISBN: (纸本)9798331522483

The focus of this research is on college students' writing skill improvement with the help of Generative AI offered by Grammarly and their perception after using it, including its ethical concerns. While there is no doubt that in recent years AI tools have gained traction because of the ability to enhance one's written communication, questions have been raised regarding the excessive dependency on such tools and the student's academic integrity. The modeling of User Experience Questionnaire (UEQ) metric was used to complement the qualitative evaluation of opinions of students as users of Grammarly's Generative AI, with the investigation addressing specific dimensions: attractiveness, efficiency, and dependability. It has been established that students are attracted to AI and engaged when using it, though the concern for ethics regarding user control is prominent. This finding shows that ethical use is necessary when integrating AI technology into education to enhance learning without replacing critical analysis and self-learning skills;this extends the ongoing dialogue regarding the use of AI technologies in an education setting by giving precise guidance for educators and policy makers on how to appropriately use AI tools in academic role. © 2024 IEEE.

关键词： Ethical technology

来源：评论

学校读者我要写书评

暂无评论

Deep learning based RAGAE-SVM for Chronic kidney disease diagnosis on internet of health things platform

引用

Multimedia Tools and Applications 2024年 1-39页

作者： Kandukuri, Prabhakar Abdul, Ashu Kumar, Kuchipudi Prasanth Sreenivas, Velagapudi Ramesh, G. Gundu, Venkateswarlu Department of Artificial Intelligence and Machine Learning Chaitanya Bharathi Institute of Technology Telangana Hyderabad500075 India Department of Computer Science and Engineering SRM University Andhra PradeshNeerukonda Mangalagiri 522503 India Department of Computer Science & amp Engineering Koneru Lakshmaiah Education Foundation Andhra PradeshVaddeswaram 522302 India Department of Computer Science and Engineering SRK Institute of Technology Andhra Pradesh Vijayawada521108 India Department of Computer Science and Engineering Gokaraju Rangaraju Institute of Engineering and Technology TelanganaBachupally Hyderabad500090 India Department of Computer Science and Engineering Koneru Lakshmaiah Education Foundation Andhra PradeshVaddeswaram 522302 India

Chronic kidney disease (CKD) is a prominent disease that causes loss of functionality in the kidney. Doctors can now more easily gather patient health status data due to the growth of the Internet of Health Things (IoHT). The IoHT data contains a huge number of redundant data, making it challenging to predict CKD disease quickly and accurately. In healthcare applications like feature-based classification, a variety of disease diagnosis systems were used to address this problem. Current disease detection algorithms suffer from imbalanced dataset processing, low-accuracy feature learning, and high computational power requirements. Thus, deep learning-based clinical decision support systems have been developed to solve these complexities. To remove outliers from medical data, data collected with IoHT devices is first pre-processed using an enhanced K-means clustering technique. The Synthetic Minority over Sampling Technique is used to balance data because the IoHT dataset is highly imbalanced. The classifier detects anomalies in less time because of the implemented processing step. The balanced CKD dataset is then presented for use with a novel classifier called Residual Attention-Gated Autoencoder with Support Vector machine. In order to improve accurate detection, the adopted classifier can learn and extract features. The proposed method results in an increased accuracy of 99.43% within 70.1 s of computation time. The mean intersection of the union and kappa coefficient of the proposed method is 98.15% and 98.1%, respectively. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Classification (of information)

来源：评论

学校读者我要写书评

暂无评论

SEED-VII: A Multimodal Dataset of Six Basic Emotions with Continuous Labels for Emotion Recognition

引用

IEEE Transactions on Affective Computing 2024年第2期16卷 969-985页

作者： Jiang, Wei-Bang Liu, Xuan-Hao Zheng, Wei-Long Lu, Bao-Liang Shanghai Jiao Tong University Center for Brain-Like Computing and Machine Intelligence Department of Computer Science and Engineering Key Laboratory of Shanghai Education Commission for Intelligent Interaction and Cognitive Engineering Brain Science and Technology Research Center Shanghai200240 China

Recognizing emotions from physiological signals is a topic that has garnered widespread interest, and research continues to develop novel techniques for perceiving emotions. However, the emergence of deep learning has highlighted the need for comprehensive and high-quality emotional datasets that enable the accurate decoding of human emotions. To systematically explore human emotions, we develop a multimodal dataset consisting of six basic (happiness, sadness, fear, disgust, surprise, and anger) emotions and the neutral emotion, named SEED-VII. This multimodal dataset includes electroencephalography (EEG) and eye movement signals. The seven emotions in SEED-VII are elicited by 80 different videos and fully investigated with continuous labels that indicate the intensity levels of the corresponding emotions. Additionally, we propose a novel Multimodal Adaptive Emotion Transformer (MAET), that can flexibly process both unimodal and multimodal inputs. Adversarial training is utilized in the MAET to mitigate subject discrepancies, which enhances domain generalization. Our extensive experiments, encompassing both subject-dependent and cross-subject conditions, demonstrate the superior performance of the MAET in terms of handling various inputs. Continuous labels are used to filter the data with high emotional intensity, and this strategy is proven to be effective for attaining improved emotion recognition performance. Furthermore, complementary properties between the EEG signals and eye movements and stable neural patterns of the seven emotions are observed. © 2010-2012 IEEE.

关键词： Eye movements

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：