检索结果-内蒙古大学图书馆

Measuring the contribution of multiple model representations in detecting adversarial instances

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Steinberg, Daniel Munro, Paul Intelligent Systems Program University of Pittsburgh School of Computing and Information University of Pittsburgh

Deep learning models have been used for a wide variety of tasks. They are prevalent in computer vision, natural language processing, speech recognition, and other areas. While these models have worked well under many scenarios, it has been shown that they are vulnerable to adversarial attacks. This has led to a proliferation of research into ways that such attacks could be identified and/or defended against. Our goal is to explore the contribution that can be attributed to using multiple underlying models for the purpose of adversarial instance detection. Our paper describes two approaches that incorporate representations from multiple models for detecting adversarial examples. We devise controlled experiments for measuring the detection impact of incrementally utilizing additional models. For many of the scenarios we consider, the results show that performance increases with the number of underlying models used for extracting representations. Code is available at https://***/dstein64/multi-advdetect. Copyright © 2021, The Authors. All rights reserved.

关键词： Speech recognition

Visualizing representations of adversarially perturbed inputs

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Steinberg, Daniel Munro, Paul Intelligent Systems Program University of Pittsburgh School of Computing and Information University of Pittsburgh

It has been shown that deep learning models are vulnerable to adversarial attacks. We seek to further understand the consequence of such attacks on the intermediate activations of neural networks. We present an evaluation metric, POP-N, which scores the effectiveness of projecting data to N dimensions under the context of visualizing representations of adversarially perturbed inputs. We conduct experiments on CIFAR-10 to compare the POP-2 score of several dimensionality reduction algorithms across various adversarial attacks. Finally, we utilize the 2D data corresponding to high POP-2 scores to generate example visualizations. Copyright © 2021, The Authors. All rights reserved.

关键词： Deep learning

Improvement of Spatial Ambiguity in Multi-Channel Speech Separation Using Channel Attention

学校读者我要写书评

暂无评论

Improvement of Spatial Ambiguity in Multi-Channel Speech Sep...

2021 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA ASC 2021

作者： Hong, Qian-Bei Wu, Chung-Hsien Nguyen, Thanh Binh Wang, Hsin-Min National Cheng Kung University and Academia Sinica Graduate Program of Multimedia Systems and Intelligent Computing Tainan Taiwan National Cheng Kung University Department of Computer Science and Information Engineering Tainan Taiwan

ISBN: (纸本)9789881476890

Multi-channel speech separation has been successfully applied in a complex real-world environment such as the far-field condition. The common solution to deal with the far-field condition is using a multi-channel signal captured by a structured microphone array and leveraging the inner difference between channels to enhance the speech separation performance. The spatial feature has been widely used in recent speech separation research. This feature appears to be insufficient when the location information becomes ambiguous. This is known as the spatial ambiguity problem. In order to deal with the spatial ambiguity problem, this study proposes an attention mechanism for the Temporal-Spatial Neural Filter (TSNF), in which the channel attention on merged features and the feature map of 1D convolution block in the temporal convolution network is proposed. The proposed method is evaluated on the multi-channel reverberant dataset which is built based on the WSJ0-2mix dataset. The dataset is simulated in the real-environment room by using the Room Impulse Response generator. In the experimental results, the proposed methods produced the SI -SNR improvement of about 1.2dB in close speakers' case, while a small decrease of 0.1dB in other cases. © 2021 APSIPA.

关键词： Convolution

Boosting Patch Correctness Checking via Combining Code-Edit Embeddings as Additional What-to-Change Information

学校读者我要写书评

暂无评论

Boosting Patch Correctness Checking via Combining Code-Edit ...

IEEE WIC ACM International Conference on Web Intelligence (WI)

作者： Chontipan Plengvittaya Zee Hen Tang Mi-Yen Yeh Chih-Ya Shen Social Networks and Human Centered Computing Taiwan International Graduate Program Taiwan Institute of Information Science Academia Sinica Taiwan Institute of Information Systems and Applications National Tsing Hua University Taiwan Data Science Degree Program National Taiwan University and Academia Sinica Taiwan

Patch Correctness Checking is the task of detecting automated-program-repair-generated patches that can pass the test suite but are still buggy. In this study, we investigated a new feature to improve the classification performance for automated patch correctness checking. Our approach involved leveraging code-change information at a finer granular level by embedding the code edits and combining them with existing code-change vectors. Through extensive experiments, the results revealed gains up to 3.4% and 3.2% on the Fl-score and AUC.

关键词：

A Framework for Moving from Traditional to Smart University 10

学校读者我要写书评

暂无评论

A Framework for Moving from Traditional to Smart University

17th INDIACom;2023 10th International Conference on computing for Sustainable Global Development, INDIACom 2023

作者： Sen, Adnan Ahmed Abi Aljohani, Yaser Mohammed Hakami, Abdulelah Alhussain Hassan Bahbouh, Nour Mahmoud Alkhodre, Ahmad Alghamdi, Mohammed S. Islamic University Computing and Information Technology Al-Madinah Saudi Arabia YasRef Program Yanbu Saudi Arabia Saudi Electronic University Al-Madinah Saudi Arabia Granada University Department of Information and Communication Sciences Granada Spain Islamic University Faculty of Computer and Information Systems Al-Madinah Saudi Arabia Islamic University Faculty of the Holy Quran and Islamic Studies Al-Madinah Saudi Arabia

ISBN: (纸本)9789380544472

The revolution of technologies and communications pushed for smarter applications and services. A smart university has become a target for educational institutions in many countries in the last few years. The smart-university provides many advantages in all aspects: academic, administrative, and research. Transformation to a smart university will save time, effort, and cost, in addition to increasing the quality of services, users' satisfaction, and sustainability. This research provided reviewing study for the most important standards, services, applications, and tools that must be present in smart universities. Then, the research proposed a framework to transform universities into smart ones. In recent years, the Kingdom of Saudi Arabia (KSA) is considered one of the leading countries in the field of digital transformation. So, in the next phase, we will try to apply the proposed framework in one of the KSA universities. © 2023 Bharati Vidyapeeth, New Delhi.

关键词： Internet of things

Adaptive Contextual Caching for Mobile Edge Large Language Model Service

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Liu, Guangyuan Liu, Yinqiu Wang, Jiacheng Du, Hongyang Niyato, Dusit Kang, Jiawen Xiong, Zehui College of Computing and Data Science The Energy Research Institute @ NTU Interdisciplinary Graduate Program Nanyang Technological University Singapore College of Computing and Data Science Nanyang Technological University Singapore Department of Electrical and Electronic Engineering University of Hong Kong Hong Kong School of Automation Guangdong University of Technology China Pillar of Information Systems Technology and Design Singapore University of Technology and Design Singapore

Mobile edge Large Language Model (LLM) deployments face inherent constraints, such as limited computational resources and network bandwidth. Although Retrieval-Augmented Generation (RAG) mitigates some challenges by integrating external knowledge bases, inefficient cache management can still result in high retrieval latency and frequent cache updates. To address these issues, we propose an Adaptive Contextual Caching (ACC) framework that anticipates user needs by proactively caching semantically relevant data for mobile-edge LLMs. ACC utilizes a deep reinforcement learning (DRL) module to refine cache replacement policies, balancing user context, document similarity, and the overhead associated with cache misses. Experimental results demonstrate that ACC increases cache hit rates to over 80% after only 11 training episodes, outperforming FIFO, LRU, and semantic-only caching while reducing retrieval latency by up to 40%. In particular, ACC also reduces local caching overhead (i.e., the cost of updating the cache when a miss occurs) by as much as 55%, enabling scalable, low-latency LLM services in resource-constrained edge environments. © 2025, CC BY.

关键词： Deep reinforcement learning

HAEE: Question Classification Using Hierarchical Intra-Attention Enhancement Encoder

学校读者我要写书评

暂无评论

HAEE: Question Classification Using Hierarchical Intra-Atten...

26th International Conference on Technologies and Applications of Artificial Intelligence, TAAI 2021

作者： Wang, Jen-Wei Chen, Kai-Hsiang Huang, Jen-Wei National Cheng Kung University Department of Electrical Engineering Tainan Taiwan National Cheng Kung University and Academia Sinica Graduate Program for Multimedia Systems and Intelligent Computing Taiwan

ISBN: (纸本)9781665408257

With the development of E-commerce, an Automated Question-Answering system takes a crucial part in customer service. Question classification, which assigns labels to questions according to the answer types, is one of the tasks in question answering. Previous methods usually used handcraft features like named entity recognition, but it needs the predefined dictionary or tools. The machine learning approaches are recently applied to this task and achieve high accuracy. In this paper, we proposed HAEE, a Hierarchical intra-Attention Enhancement Encoder which composed of bidirectional GRUs and intra-attentions. In addition, we adopt the character input to address the issue of the OOV (Out-Of-Vocabulary) problem and create multiple intra-attentions to simulate the certain relationships between characters (Chinese) or words (English) to enhance the influence of tokens on the sentence. We evaluate the HAEE model in an actual corporate setting and several datasets. As shown in the experimental results, our HAEE model outperforms the existing state-of-the-art models on question classification tasks, especially for the Chinese corpus. © 2021 IEEE.

关键词： Signal encoding

Confidentialchain: Enhancing the Security of a Notarization System Based on Blockchain

学校读者我要写书评

暂无评论

Confidentialchain: Enhancing the Security of a Notarization ...

Blockchain computing and Applications (BCCA), International Conference on

作者： Magno Alves Cavalcante José Antônio Moreira Xexéo Anderson Fernandes Pereira Dos Santos Information Technology and Telecommunications Petróleo Brasileiro S.A. – PETROBRAS Rio de Janeiro Brazil Graduate Program in Systems and Computing Instituto Militar de Engenharia Rio de Janeiro Brazil Venturus Centro de Inovação Tecnológica Brazil

ISBN: (数字)9798350351538

ISBN: (纸本)9798350351545

Blockchain platforms offer traceability, integrity, immutability and availability by native construction, providing resources for auditors to attest the security and temporal traceability of data stored in the blocks. However, permissioned blockchain platforms do not provide confidentiality and nonrepudiation natively for sensitive user data. In this work, Confidentialchain is proposed as a hybrid cryptography schema to aggregate the characteristics of confidentiality and non-repudiation to a confidential data storage system, which is executed on a permissioned blockchain network, ensuring sufficient confidence for the subsequent secure use of sensitive user data that are stored on the blockchain network.

关键词： Buffer storage Aggregates Non-repudiation Blockchains

2D Basement Relief Inversion using Sparse Regularization.

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Barboza, Francisco Márcio da Cunha Romão Silva, E. Arthur Anthony de Carvalho, Bruno Motta Department of Computing and Technology Federal University of Rio Grande do Norte Rio Grande do Norte Caicó Brazil Graduate Program on Systems and Computing Federal University of Rio Grande do Norte Rio Grande do Norte Natal Brazil Department of Informatics and Applied Mathematics Federal University of Rio Grande do Norte Rio Grande do Norte Natal Brazil

Basement relief gravimetry is a key application in geophysics, particularly important for oil exploration and mineral prospecting. It involves solving an inverse geophysical problem, where the parameters of a geological model are inferred from observed data. In this context, the geological model consists of the depths of constant-density prisms representing the basement relief, and the data correspond to the gravitational anomalies caused by these prisms. Inverse geophysical problems are typically ill-posed, as defined by Hadamard, meaning that small perturbations in the data can result in large variations in the solutions. To address this instability, regularization techniques, such as those proposed by Tikhonov, are employed to stabilize the solutions. This study presents a comparative analysis of various regularization techniques applied to the gravimetric inversion problem, including Smoothness Constraints, Total Variation, the Discrete Cosine Transform (DCT), and the Discrete Wavelet Transform (DWT) using Daubechies D4 wavelets. Optimization methods are commonly used in inverse geophysical problems because of their ability to find optimal parameters that minimize the objective function—in this case, the depths of the prisms that best explain the observed gravitational anomalies. The Genetic Algorithm (GA) was selected as the optimization technique. GA is based on Darwinian evolutionary theory, specifically the principle of natural selection, where the fittest individuals in a population are selected to pass on their traits. In optimization, this translates to selecting solutions that most effectively minimize the objective function. The results, evaluated using fit metrics and cumulative error analysis, demonstrate the effectiveness of all the regularization techniques and the Genetic Algorithm. Among the methods tested, the Smoothness constraint was briefly the most effective for the first and second synthetic models. For the third model, which was based on re

关键词： Discrete wavelet transforms