检索结果-内蒙古大学图书馆

Responsible Music Genre Classification Using Interpretable Model-Agnostic Visual Explainers

SN computer science 2025年第1期6卷 1-26页

作者： Murindanyi, Sudi Hamza, Kyamanywa Kagumire, Sulaiman Marvin, Ggaliwango Department of Computer Science Makerere University Kampala Uganda Department of Electrical Engineering Korea University 145 Anam-ro Seongbuk-gu Seoul South Korea

The responsible artificial intelligence (AI) paradigm requires machine learning (ML) and AI engineers to ensure transparent and interpretable intelligent models across domains. This requirement becomes even more complex and crucial when dealing with feature learning using sound data, as deriving model explainers in this context is a more detailed and sophisticated process. In this work, we demonstrate a responsible approach to AI modeling and leverage three explainable artificial intelligence (XAI) tools to derive and establish explainers for the music genre classification model-agonistically. Explain like am 5 (ELi5), Shapley Additive exPlanations (SHAP), and Local Interpretable Model-agnostic Explanations (LIME) were particularly used to interpret and simplify the model’s decision-making processes regardless of their discrete mathematical foundations. This transparency provides a deeper understanding of sound feature patterns and characteristics that inform genre classes for the models through graphical and qualitative feature contributions to explain the model’s justification. We developed convolutions neural networks (CNNs) with and without cross-validation and a vision transformer (ViT) approach utilizing MobileNets. Ultimately, the CNN with cross-validation demonstrated superior performance, achieving 80% accuracy on the test set and 84% accuracy on the validation set. This work advances the border of music intelligence research and promotes the broader cause of responsible AI to ensure that complex models remain comprehensible and accountable. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2024.

关键词： Machine learning Model interpretability Music genre Responsible AI

来源：评论

学校读者我要写书评

暂无评论

Supporting Formal Methods for Machine Learning Verification in Urban Air Mobility

Supporting Formal Methods for Machine Learning Verification ...

引用

2024 IEEE International Conferences of AI, science, engineering, and Technology, AIxSET 2024

作者： Davidoff, Alexandra Couder, Juan Ochoa, Omar Embry-Riddle Aeronautical University Department of Electrical Engineering and Computer Science Daytona BeachFL United States

ISBN: (纸本)9798350390995

The development of Urban Air Mobility (UAM) infrastructure has garnered significant attention in recent years. Specifically, researchers have explored using Machine Learning (ML) technology to improve the efficiency and scalability of the UAM ecosystem. However, incorporating ML into safety-critical applications requires increased Verification and Validation (V&V) efforts to ensure safety to human life, an endeavor that is heavily scrutinized but not yet solved in current research. This paper proposes utilizing formal methods to verify ML systems integrated within UAM as a mitigation to safety concerns. We present brief surveys on ML trends within UAM and formal methods trends within UAM. We then combine these disciplines to argue for the usage of formal methods for ML verification in UAM while discussing modern trends within formal verification of ML systems. By utilizing formal methods alongside traditional verification techniques to verify ML systems, the safety and reliability of UAM can be increased. Therefore, this paper calls for further research into formal verification of ML within UAM. © 2024 IEEE.

关键词： Adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

Modeling, Simulation and Control of Irrigation Canal Networks 12

Modeling, Simulation and Control of Irrigation Canal Network...

引用

2024 12th International Conference on Systems and Control, ICSC 2024

作者： Georgakis, Panagiotis Bechlioulis, Charalampos P. University of Patras Department of Electrical and Computer Engineering Patra26504 Greece

ISBN: (纸本)9798331531812

Irrigation canal networks play a pivotal role in satisfying the water demands of farmlands around the world. In a time of global warming, automatic irrigation systems are crucial in order to minimize water losses caused by human error and complex dynamics. This work serves as a holistic approach towards the problem of modeling, simulating and controlling the fluid dynamics of irrigation canal networks. The PDEs describing the flow dynamics are briefly illustrated, along with the dynamics introduced by the canal hydraulic structures. The PDEs are simulated using a 2nd order Finite Volume Scheme. The scheme utilizes an HLL Riemann Solver for the spatial discretization and an SSPRK2 time integrator for the temporal discretization. The full flow dynamics are approximated using various control-oriented lumped-parameter LTI models derived from literature. The LTI models are compared against the full system dynamics in the time and frequency domains. Subsequently, a centralized MPC master-slave controller is utilized for stabilizing the water levels of the irrigation canal network under changes in the flow regime. The proposed controller is coupled with the devised fluid simulator. Finally, the response of the irrigation canal control system is exhibited in a simulated version of the Corning Canal in California, USA, which serves as Test Case 2 according to the benchmarks set by ASCE for developing canal control algorithms. © 2024 IEEE.

关键词： Irrigation canals

来源：评论

学校读者我要写书评

暂无评论

A Proposed Hybrid Deep Learning Model for Wind Power Forecasting

A Proposed Hybrid Deep Learning Model for Wind Power Forecas...

引用

2024 ASU International Conference in Emerging Technologies for Sustainability and Intelligent Systems, ICETSIS 2024

作者： Aly, Hamed H. Dalhousie University Electrical and Computer Engineering Department HalifaxNS Canada

ISBN: (纸本)9798350372229

Renewable energy forecasting is crucially important because of its fluctuation and stochastic characteristics. In this paper, a hybrid model for wind speed and power forecasting using neuro wavelet and long short-term memory (LSTM) is proposed. The architecture of the proposed forecasting model involves two steps;the first step is to employ a time-based neuro wavelet for the wind speed or power forecasting. The second step is to subtract the forecasted wind speed or power from the actual ones to calculate the error (residuals). This error is then fed as an input to the LSTM to determine the forecasted wind speed or power error. The forecasted wind speed will be equal to that from the first step and the forecasted wind error from the second step. The same procedures are repeated for the forecasted wind power. In this paper, a simulated model for wind power is used. The results demonstrate the effectiveness of the proposed model for wind speed and power forecasting. © 2024 IEEE.

关键词： DFIG LSTM Neuro Wavelet Wind Power

来源：评论

学校读者我要写书评

暂无评论

Exact Transmission through a Metamaterial Parabolic-Cylinder Radome 25

Exact Transmission through a Metamaterial Parabolic-Cylinder...

引用

25th International Conference on Electromagnetics in Advanced Applications, ICEAA 2024

作者： Uslenghi, Piergiorgio L.E. Department of Electrical and Computer Engineering University of Illinois Chicago United States

ISBN: (纸本)9798350360974

A radome whose interfaces with the surrounding medium are confocal parabolic cylinders with the same symmetry plane is considered. The medium on both convex and concave sides of the radome is linear, homogeneous, lossless and isotropic;it is characterized by a real positive electric permittivity si and a real positive magnetic permeability \i\ or, equivalently, by a refractive index n=V(si \i\) and an intrinsic impedance Zi=V(ni/si). The radome is made of a linear, homogeneous, lossless and isotropic metamaterial characterized by a real negative permittivity -S2 and a real negative permeability -\i2 such that E\\i\=e,2\i2- Because of causality [1], this entails that the media outside and inside the radome have real and opposite refractive indexes, and positive intrinsic impedances: Z2=V((i2/s2) is real positive and possibly different from Z\. The primary field is a plane wave symmetrically incident on the concave side of the radome, and the analysis is conducted in the phasor domain, with a time-dependent factor exp(+jd>t) omitted throughout. © 2024 IEEE.

关键词： Cylinders (shapes)

来源：评论

学校读者我要写书评

暂无评论

Implicit Optimization Bias of Next-token Prediction in Linear Models 38

Implicit Optimization Bias of Next-token Prediction in Linea...

引用

38th Conference on Neural Information Processing Systems, NeurIPS 2024

作者： Thrampoulidis, Christos Department of Electrical and Computer Engineering University of British Columbia Vancouver Canada

We initiate an investigation into the optimization properties of next-token prediction (NTP), the dominant training paradigm for modern language models. Specifically, we study the structural properties of the solutions selected by gradient-based optimizers among the many possible minimizers of the NTP objective. By framing NTP as cross-entropy minimization across distinct contexts, each tied with a sparse conditional probability distribution across a finite vocabulary of tokens, we introduce "NTP-separability conditions" that enable reaching the data-entropy lower bound. With this setup, and focusing on linear models with fixed context embeddings, we characterize the optimization bias of gradient descent (GD): Within the data subspace defined by the sparsity patterns of distinct contexts, GD selects parameters that equate the logits' differences of in-support tokens to their log-odds. In the orthogonal subspace, the GD parameters diverge in norm and select the direction that maximizes a margin specific to NTP. These findings extend previous research on implicit bias in one-hot classification to the NTP setting, highlighting key differences and prompting further research into the optimization and generalization properties of NTP, irrespective of the specific architecture used to generate the context embeddings. © 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

Considering Power Losses and Voltage Drop in Average-Value Modeling of Voltage Source Inverters in Simulations of AC Machine Drives

Considering Power Losses and Voltage Drop in Average-Value M...

引用

2024 IEEE International Conference on Modeling, Simulation and Intelligent Computing, MoSICom 2024

作者： Kaushik, Abhay Tahim, Ekamjot Singh Rahman, Shadman Saqlain Jatskevich, Juri University of British Columbia Department of Electrical and Computer Engineering Vancouver Canada

ISBN: (纸本)9798331533311

Voltage-source inverters (VSIs) are used extensively in AC machine drive applications. Detailed switching models of VSIs offer accurate but computationally expensive simulations. Average-value models (AVMs) of VSIs are often used for fast and efficient simulations in electromagnetic transient (EMT) programs. However, most conventional AVMs do not consider power losses. This paper extends the previous work on representing the losses in parametric AVMs of VSIs by including the overall losses that are frequency- and current-dependent, as well as voltage drop due to the ON -state resistance of switches. Several new lossy AVMs (LAVMs) and their equivalent circuits are presented. The losses may be implemented on the DC side, AC side, or distributed between the DC and AC sides. The proposed LAVMs are demonstrated on an induction machine drive for which the losses are determined experimentally. The proposed LAVMs exhibit exceptional accuracy in capturing the effects of losses and voltage drop compared to experimental results. © 2024 IEEE.

关键词： Electromagnetic transients

来源：评论

学校读者我要写书评

暂无评论

Securing Electronic Health Records from Insider Threats in Smart City Healthcare Cloud Using Machine Learning Approach 5

Securing Electronic Health Records from Insider Threats in S...

引用

5th International Conference on Intelligent Communication Technologies and Virtual Mobile Networks, ICICV 2024

作者： Ramasami, S. Uma Maheswari, P. Bannari Amman Institute of Technology Department of Computer Science and Engineering Sathyamangalam Erode India Anna University Department of Computer Science and Engineering Chennai India

ISBN: (纸本)9798350385649

In recent years, healthcare in smart cities is considered as significant to create a more resilient and well-informed healthcare ecosystem. The integration of cloud computing in healthcare industry has facilitated the storage and sharing of electronic health records (EHRs), enabling seamless access to medical information. However, this transition to cloud-based systems has introduced new security challenges, particularly the risk of insider threats. Insiders with authorized access to cloud-based EHR systems, such as healthcare professionals and cloud service providers, may exploit vulnerabilities or misuse sensitive patient data, leading to privacy breaches and unauthorized disclosure. This research work presents an innovative approach to securing EHRs from insider threats in the cloud using Gaussian Mixture Model and Classification algorithms. These techniques analyze user behaviour patterns and detect anomalous activities that may indicate insider threats. Here UK hospital electronic patient health record data set is used to conduct the experiments. The proposed approach comprises two key stages. In the initial stage, the unsupervised Gaussian Mixture Model is utilized to find the abnormal patterns in the dataset and label each record in the dataset as normal or anomaly. The second stage involves different supervised classification algorithms namely SVM, KNN, DT, NB and RF are used to classify the new instance. The results demonstrate that the Random Forest (RF) achieves high accuracy in detecting insider threats with an accuracy of 99.97%. The findings of this research contribute to the field of healthcare data security by offering an intelligent and proactive solution for mitigating insider threats to cloud-based EHRs in Smart city environments. © 2024 IEEE.

关键词： Support vector machines

来源：评论

学校读者我要写书评

暂无评论

A Novel Time-Driven Twin CNN-Based NILM Framework Using Double Grayscale Images of Odd Harmonic Currents 24

A Novel Time-Driven Twin CNN-Based NILM Framework Using Doub...

引用

24th EEEIC International Conference on Environment and electrical engineering and 8th I and CPS Industrial and Commercial Power Systems Europe, EEEIC/I and CPS Europe 2024

作者： Papageorgiou, Petros G. Christoforidis, Georgios C. Bouhouras, Aggelos S. University of Western Macedonia Department of Electrical and Computer Engineering Kozani Greece

ISBN: (纸本)9798350355185

Non-intrusive load monitoring (NILM) is a key solution for the application of load identification to electrical measurements performed at the level of an aggregated input point. In recent years, a higher sampling rate for the measurements has been approved to improve the robustness of the extracted features, such as features from harmonic analysis. The latter are also valuable considering that the harmonic content of devices increases over time due to their smart operation based on power electronics. However, the higher sampling rate leads to computational complexity issues that prevent practical application of NILM in real time. Therefore, measures should be taken to reduce the computational complexity while keeping the robustness of the features at a high level. Convolutional neural networks have developed rapidly in recent years, especially in the field of image classification. It has also been shown that the extraction of grayscale images can mitigate the computational complexity. Therefore, a novel line graph pair of grayscale images of odd harmonic currents is proposed in this paper, which combines robustness and low computational complexity in terms of load classification. This novel feature pair is fed into a novel twin model based on convolutional neural networks, which is optimally tuned using particle swarm optimization. The results in terms of appliance classification and required execution time are promising and confirm the robustness and real-time capability of the proposed method. The micro-average multi-label FScore over the PLAID3 dataset is 92%, while the total average execution time is 518 ms with CPU calculations. © 2024 IEEE.

关键词： Strain measurement

来源：评论

学校读者我要写书评

暂无评论

Integrating Low Power Data Encoding in Intel Advanced Interface Bus

Integrating Low Power Data Encoding in Intel Advanced Interf...

引用

2024 Panhellenic Conference on Electronics and Telecommunications, PACET 2024

作者： Kotoulas, Emmanouil Pavlidis, Vasilis F. Aristotle University of Thessaloniki Department of Electrical and Computer Engineering Thessaloniki Greece

ISBN: (纸本)9798350318845

The ever-growing need for computational power has led to advanced packaging technologies for integrated circuits. One such technology is 2.5D integration, which reduces the distance between chiplets in a single package, resulting in a reduction of delay and power during data transfer. To connect chiplets, Intel created the Advanced Interface Bus (AIB), an open-source interface, as a universal solution for low power chiplet communication through a wide parallel channel with a low complexity circuit and an active redundancy scheme. However, with the increase in frequency of the interconnect, a considerable part of the total system power is used for data transfers. To reduce the power consumed by the interconnect, data encoding schemes are deployed. This paper explores the integration of our encoding scheme Adaptive Word Reordering (AWR) in Intel's AIB, evaluating the energy savings that can result by combining an emerging industrial interface with a modern data encoding technique. The results show that with the integration of AWR in AIB, the total power is reduced by 5.5% whereas the included Data Bus Invert (DBI) encoding scheme does not achieve noticeable improvement for the evaluated workloads. © 2024 IEEE.

关键词： Data transfer

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：