检索结果-内蒙古大学图书馆

Hybrid Heuristic Optimization of Deep Learning Models for Robust Plant Disease Detection 7

学校读者我要写书评

暂无评论

Hybrid Heuristic Optimization of Deep Learning Models for Ro...

7th IEEE International Conference on Emerging Smart Computing and informatics, ESCI 2025

作者： Ananthajothi, K. Kshirsagar, Pravin R. Tak, Tan Kuan Kumar, Kailash Department of Computer Science and Engineering Rajalakshmi Engineering College Chennai India Department of Electronics & Telecommunication Engineering J D College of Engineering & Management Nagpur India Electrical Power Engineering Programme Singapore Institute of Technology Singapore College of Computing and informatics Saudi Electronic University Riyadh Saudi Arabia

ISBN: (纸本)9798331515683

Plant diseases directly affect farm output by lowering grain, fruit, and vegetable quality, threatening global food security. Thus, farmers inspect plant leaves with their eyes. Plant leaf monitoring is unreliable and error-prone. A multitude of deep learning algorithms have been developed to detect plant leaf diseases;however, most rely on low-resolution photos utilizing convolutional neural nets (CNNs). This research seeks to create an improved nine-layer the CNN network model for the accurate classification of leaf-related illnesses. A contrast enhancement approach is used to pre-process plant leaf pictures. Following binary thresholding, the pre-processed pictures are separated into leaf images and abnormality segmented using the “Enhanced U-Net (EU-Net)” approach. A Multilevel Feature Fusion Network Convolutional Neural Network (MFFN-CNN) is employed to categorize illnesses of leaves based on segmented images. The "Hybrid Leader Cat Swarm Optimization(HLCSO)" method improves U-Net parameter optimization. The effectiveness of the leaf diagnostic model is demonstrated experimentally employing a variety of factors through the use of numerous baseline approaches. © 2025 IEEE.

关键词： Cat Swarm Optimization Convolutional Neural Networks Multilevel Feature Fusion Network Plant Disease pre-processed images thresholding U-Net

Dynamics of hot random hyperbolic graphs

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Papadopoulos, Fragkiskos Zambirinis, Sofoclis Department of Electrical Engineering Computer Engineering and Informatics Cyprus University of Technology Limassol3036 Cyprus

We derive the most basic dynamical properties of random hyperbolic graphs (the distributions of contact and intercontact durations) in the hot regime (network temperature T > 1). We show that in the thermodynamic limit the contact distribution decays as a power law with exponent 2+T > 3 for durations t > T, while for t thmoment. Otherwise, the contact distribution depends on the expected degree distribution and we show that if the latter is a power law with exponent γ ∈ (2, T+1], then the former decays as a power law with exponent γ+1 > 3. On the other hand, the intercontact distribution exhibits power-law decays with exponent 2-T ∈ (0, 1) for T ∈ (1,2), while for T > 2 it displays linear decays with a slope that depends on the observation interval. This result holds irrespective of the expected degree distribution as long as it has a finite Tthmoment if T ∈ (1,2), or a finite second moment if T > 2. Otherwise, the intercontact distribution depends on the expected degree distribution and if the latter is a power law with exponent γ ∈ (2, 3), then the former decays as a power law with exponent 3-γ ∈ (0,1). Thus, hot random hyperbolic graphs can give rise to contact and intercontact distributions that both decay as power laws. These power laws however are unrealistic for the case of the intercontact distribution, as their exponent is always less than one. These results suggest that hot random hyperbolic graphs are not adequate null models for real temporal networks, in stark contrast to cold random hyperbolic graphs (T Copyright © 2021, The Authors. All rights reserved.

关键词： Graphic methods

Late Breaking Results: Weight Decay is All You Need for Neural Network Sparsification 23

学校读者我要写书评

暂无评论

Late Breaking Results: Weight Decay is All You Need for Neur...

Proceedings of the 60th Annual ACM/IEEE Design Automation Conference

作者： Xizi Chen Rui Pan Xiaomeng Wang Fengshi Tian Chi-Ying Tsui College of Informatics Huazhong Agricultural University China Department of Electronic and Computer Engineering The Hong Kong University of Science and Technology Hong Kong China

ISBN: (纸本)9798350323481

The heuristic iterative pruning strategy has been widely used for neural network sparsification. However, it is challenging to identify the right connections to remove at each pruning iteration with only a one-shot evaluation of weight magnitude, especially at the early pruning stage. The erroneously removed connections, unfortunately, can hardly be recovered. In this work, we propose a weight decay strategy as a substitute for pruning, which let the "insignificant" weights moderately decay instead of being directly clamped to zero. At the end of the training, the vast majority of redundant weights will naturally become close to zero, making it easier to identify which connections could be removed safely. Experimental results show that the proposed weight decay method can achieve an ultra-high sparsity of 99%. Compared to the current pruning strategy, the model size is further reduced by 34%, improving the compression rate from 69× to 106× at the same accuracy.

关键词： DNN

Transformer-Based Embeddings for Greek Language Categorization

学校读者我要写书评

暂无评论

Transformer-Based Embeddings for Greek Language Categorizati...

International Conference on computer and Information Science (ACIS)

作者： Charalampos M. Liapis Konstantinos Kyritsis Isidoros Perikos Michael Paraskevas Mathematics Dept. University of Patras Computer Technology Institute & Press “Diophantus” Patras Greece Electrical and Computer Engineering Dept. University of Peloponnese Patras Greece Computer Engineering and Informatics Department Electrical and Computer Engineering Dept. University of Patras University of Peloponnese Computer Technology Institute and Press “Diophantus” Patras Greece Electrical and Computer Engineering Dept. University of Peloponnese Computer Technology Institute and Press “Diophantus” Patras Greece

ISBN: (数字)9798350376647

ISBN: (纸本)9798350376654

The Greek School Network (GSN) provides support to students, teachers, and school units in secondary education across Greece. Handling numerous user queries manually can be challenging, necessitating the development of an automated system for accurate categorization of these queries. This paper presents a comparative study of various transformer-based models for multi-class text categorization of Greek language queries submitted to the GSN helpdesk. We introduce a new experimental balanced dataset and extract vector representations from eleven transformer-based models. These representations are evaluated using ten classic machine learning classifiers. Our findings highlight the superior performance of the Multilingual E5 Text Embeddings model, particularly when paired with the extreme gradient-boosting classifier. This combination demonstrates a clear advantage in accurately categorizing user queries, paving the way for more efficient automated helpdesk systems.

关键词： Information science Accuracy Computational modeling Text categorization Education Machine learning Transformers Vectors

A Pipeline of Augmentation and Sequence Embedding for Classification of Imbalanced Network Traffic

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Shokri, Matin Hasibi, Ramin Department of Electrical and Computer Engineering K. N. Toosi University of Technology Iran Department of Informatics University of Bergen Bergen Norway

Network Traffic Classification (NTC) is one of the most important tasks in network management. The imbalanced nature of classes on the internet presents a critical challenge in classification tasks. For example, some classes of applications are much more prevalent than others, such as HTTP. As a result, machine learning classification models do not perform well on those classes with fewer data. To address this problem, we propose a pipeline to balance the dataset and classify it using a robust and accurate embedding technique. First, we generate artificial data using Long Short-Term Memory (LSTM) networks and Kernel Density Estimation (KDE). Next, we propose replacing one-hot encoding for categorical features with a novel embedding framework based on the"Flow as a Sentence" perspective, which we name FS-Embedding. This framework treats the source and destination ports, along with the packet’s direction, as one word in a flow, then trains an embedding vector space based on these new features through the learning classification task. Finally, we compare our pipeline with the training of a Convolutional Recurrent Neural Network (CRNN) and Transformers, both with imbalanced and sampled datasets, as well as with the one-hot encoding approach. We demonstrate that the proposed augmentation pipeline, combined with FS-Embedding, increases convergence speed and leads to a significant reduction in the number of model parameters, all while maintaining the same performance in terms of accuracy. © 2025, CC BY.

关键词： Network embeddings

Fast Mapping of Flexibility Regions at TSO-DSO Interfaces under Uncertainty

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Patig, Alice Stanojev, Ognjen Aristidou, Petros Kiprakis, Aristides Hug, Gabriela Eeh - Power Systems Laboratory Eth Zurich Switzerland Department of Electrical Engineering Computer Engineering and Informatics Cyprus University of Technology Cyprus Institute for Energy Systems School of Engineering University of Edinburgh United Kingdom

The gradual decommissioning of fossil fuel-driven power plants, that traditionally provide most operational flexibility in power systems, has led to more frequent grid stability issues. To compensate for the lack of flexible resources, Distributed Energy Resources (DERs) in distribution networks can be employed. To facilitate the use of DERs, the aggregated flexibility in a distribution grid is commonly represented on a PQ-plane displaying the feasible active and reactive power exchange with the upstream grid. This paper proposes a fast feasible operating region mapping mechanism that utilizes a linear power flow approximation in combination with linearized generator, current, and voltage constraints to construct a high-dimensional polyhedral feasible set of DER power injections. The obtained polytope is projected onto the PQ-plane using Fourier-Motzkin Elimination to represent the aggregate network flexibility. Additionally, uncertainty in DER generation is addressed using chance-constraints. Our analysis of a modified 33-bus IEEE test system demonstrates that the proposed method obtains more accurate approximations than former geometric methods and is ten times faster than the tested optimization-based method. Copyright © 2022, The Authors. All rights reserved.

关键词： Electric load flow

Transformative effects of ChatGPT on modern education: Emerging Era of AI Chatbots

学校读者我要写书评

暂无评论

Internet of Things and Cyber-Physical Systems 2024年第1期4卷 19-23页

作者： Gill, Sukhpal Singh Xu, Minxian Patros, Panos Wu, Huaming Kaur, Rupinder Kaur, Kamalpreet Fuller, Stephanie Singh, Manmeet Arora, Priyansh Parlikad, Ajith Kumar Stankovski, Vlado Abraham, Ajith Ghosh, Soumya K. Lutfiyya, Hanan Kanhere, Salil S. Bahsoon, Rami Rana, Omer Dustdar, Schahram Sakellariou, Rizos Uhlig, Steve Buyya, Rajkumar School of Electronic Engineering and Computer Science Queen Mary University of London London United Kingdom Shenzhen Institute of Advanced Technology Chinese Academy of Sciences Shenzhen China Raygun Performance Monitoring Wellington New Zealand Center for Applied Mathematics Tianjin University Tianjin China Department of Science Kings Education London United Kingdom Cymax Group Technologies BC Canada QM Academy Queen Mary University of London London United Kingdom Jackson School of Geosciences University of Texas at Austin AustinTX United States Centre for Climate Change Research Indian Institute of Tropical Meteorology Pune India Microsoft Hyderabad India Institute for Manufacturing Department of Engineering University of Cambridge Cambridge United Kingdom Faculty of Computer and Information Science University of Ljubljana Ljubljana Slovenia Machine Intelligence Research Labs AuburnWA United States Faculty of Computing and Data Science FLAME University Maharashtra Pune India Department of Computer Science and Engineering Indian Institute of Technology Kharagpur India Department of Computer Science University of Western Ontario London Canada Sydney Australia School of Computer Science University of Birmingham United Kingdom School of Computer Science and Informatics Cardiff University Cardiff United Kingdom Distributed Systems Group Vienna University of Technology Vienna Austria Department of Computer Science University of Manchester Oxford Road Manchester United Kingdom Laboratory School of Computing and Information Systems The University of Melbourne Australia

ChatGPT, an AI-based chatbot, offers coherent and useful replies based on analysis of large volumes of data. In this article, leading academics, scientists, distinguish researchers and engineers discuss the transformative effects of ChatGPT on modern education. This research discusses ChatGPT capabilities and its use in the education sector, identifies potential concerns and challenges. Our preliminary evaluation shows that ChatGPT perform differently in different subject areas including finance, coding, maths, and general public queries. While ChatGPT has the ability to help educators by creating instructional content, offering suggestions and acting as an online educator to learners by answering questions, transforming education through smartphones and IoT gadgets, and promoting group work, there are clear drawbacks in its use, such as the possibility of producing inaccurate or false data and circumventing duplicate content (plagiarism) detectors where originality is essential. The often reported "hallucinations" within GenerativeAI in general, and also relevant for ChatGPT, can render its use of limited benefit where accuracy is essential. What ChatGPT lacks is a stochastic measure to help provide sincere and sensitive communication with its users. Academic regulations and evaluation practices used in educational institutions need to be updated, should ChatGPT be used as a tool in education. To address the transformative effects of ChatGPT on the learning environment, educating teachers and students alike about its capabilities and limitations will be crucial. © 2023 The Authors

关键词： Machine learning

The Effect of Digital Marketing on Micro, Small and Medium Enterprise in Indonesia 1

学校读者我要写书评

暂无评论

International Conference on Applied informatics and Media Design, AIMD 2019

作者： Sasongko, Alvin Igo Widjaja, Gregorius Christian Theodore, Jason Afriliana, Nunik Matsuo, Tokuro Gaol, Ford Lumban School of Information System Bina Nusantara University Jakarta Indonesia Faculty of Engineering and Informatics Universitas Multimedia Nusantara Tangerang Indonesia Computer Science Department Bina Nusantara University Jakarta Indonesia Graduate School of Industrial Technology Advanced Institute of Industrial Technology Tokyo Japan

ISBN: (数字)9783031307690

ISBN: (纸本)9783031307683

Digital Marketing is a product marketing strategy using digital media and internet networks. Nowadays, digital marketing is a necessity for companies that thrive in the digital world. It greatly affects the growth of MSME (Micro, Small, Medium Enterprises) especially when it comes to increasing their sales performance by enhancing their brand awareness through various media or instruments such as a Facebook ad, Instagram advertising, Search Engine Optimization, and Search Engine Marketing. This research is aimed to measure how much impact of digital marketing utilization into Micro, Small, Medium enterprises. On the other hand, it also provides an understanding on what are the best digital marketing method to be utilized in the current situation. The method we used in this paper is quantitative. Data were collected through various questions constructed in an electronic questionnaire. This research provides a reference of digital marketing strategies for Micro, Small and Medium business owners. The result shows that digital marketing has a huge impact on the growth of Micro, Small and Medium Enterprise from increasing sales, increasing brand awareness, and gaining customers based on the data collected from both customer and Micro, Small, and Medium Enterprise players in Indonesia. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Digital storage

STREAMING PIANO TRANSCRIPTION BASED ON CONSISTENT ONSET AND OFFSET DECODING WITH SUSTAIN PEDAL DETECTION

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Wei, Weixing Zhao, Jiahao Wu, Yulun Yoshii, Kazuyoshi Graduate School of Informatics Kyoto University Japan School of Computer Science and Technology Fudan University China Graduate School of Engineering Kyoto University Japan

This paper describes a streaming audio-to-MIDI transcription method that can sequentially translate a piano recording into a sequence of note-on and note-off events. The sequence-to-sequence learning nature of this task may call for using a Transformer model, which has been used for offline transcription and could be extended for streaming transcription with a causal restriction of the attention mechanism. We assume that the decoder of this model suffers from the performance limitation. Although time-frequency features useful for onset detection are considerably different from those for offset detection, the single decoder is trained to output a mixed sequence of onset and offset events without guarantee of the correspondence between the onset and offset events of the same note. To overcome this limitation, we propose a streaming encoder-decoder model that uses a convolutional encoder aggregating local acoustic features, followed by an autoregressive transformer decoder detecting a variable number of onset events and another decoder detecting the offset events of the active pitches with validation of the sustain pedal at each time frame. Experiments using the MAESTRO dataset showed that the proposed streaming method performed comparably with or even better than the state-of-the-art offline methods while significantly reducing the computational cost. Copyright © 2025, The Authors. All rights reserved.

关键词： Decoding