检索结果-内蒙古大学图书馆

Interpretable Deep Learning for Solar Flare Prediction

学校读者我要写书评

暂无评论

Interpretable Deep Learning for Solar Flare Prediction

International Conference on Tools for Artificial Intelligence (ICTAI)

作者： Vinay Ram Gazula Katherine G. Herbert Yasser Abduallah Jason T. L. Wang Department of Data Science New Jersey Institute of Technology Newark NJ USA School of Computing Montclair State University Montclair NJ USA Department of Computer Science New Jersey Institute of Technology Newark NJ USA

ISBN: (数字)9798331527235

ISBN: (纸本)9798331527242

We propose to incorporate three interpretable methods, namely SHAP (SHapley Additive exPlanations), PDP (partial dependence plots) and Anchors, into a deep learning-based model, called SolarFlareNet, for operational flare forecasting. SolarFlareNet takes as input a sample of SHARP (Space-weather HMI Active Region Patches) magnetic parameters and predicts as output whether a solar flare would occur within the next 24 hours. We analyze flare events that occurred from May 2010 to December 2022 using the Geostationary Operational Environmental Satellite's X-ray flare catalogs and construct a database of flares with identified active regions in the catalogs. This database, together with the SHARP magnetic parameters, is used to train and test the SolarFlareNet model. Our experimental results describe the use of the three proposed methods (SHAP, PDP, and Anchors) to interpret the SolarFlareNet model and demonstrate the effectiveness of the methods.

关键词： Deep learning Additives Accuracy databases Explainable AI Predictive models Forecasting

Inverse Multiobjective Optimization by Generative Model Prompting

学校读者我要写书评

暂无评论

Inverse Multiobjective Optimization by Generative Model Prom...

Artificial Intelligence (CAI), IEEE Conference on

作者： Jiao Liu Abhishek Gupta Yew-Soon Ong Puay Siew Tan College of Computing & Data Science Nanyang Technological University (NTU) Singapore School of Mechanical Sciences Indian Institute of Technology Goa India A*STAR Singapore Institute of Manufacturing Technology (SIMTech) Singapore

ISBN: (数字)9798350354096

ISBN: (纸本)9798350354102

The integration of multiobjective optimizers with inverse models—that map points on the Pareto front to corresponding nondominated solutions—has drawn attention. These inverse models serve a dual purpose, not only facilitating the generation of candidate solutions during the optimization process, but also offering insights for multiobjective decision-making upon completion of optimization. However, today’s inverse models mainly serve to capture one-to-one mapping relations, restricting them to learn only from nondominated solution samples. As a result, the information embedded in dominated samples is not fully utilized. In this paper, we introduce a novel approach of building conditional inverse generative models (invGMs) from optimization data, making the most of both nondominated and dominated solution samples during training. Different from standard inverse models, decision-makers can query such invGMs with prompts expressed in the form of any desired objective function values, leading them to produce a corresponding solution. Through iterative prompting, invGMs are shown to accelerate the creation of diverse sets of high-quality solutions even during the course of multiobjective optimization runs. Empirical studies on three industrial optimization problems highlight the proposed method’s faster convergence rate and improved inverse modeling accuracy.

关键词： Training Accuracy Inverse problems Decision making Linear programming data models Iterative methods

A Comprehensive Survey on EEG-Based Emotion Recognition: A Graph-Based Perspective

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Liu, Chenyu Zhou, Xinliang Wu, Yihao Ding, Yi Zhai, Liming Wang, Kun Jia, Ziyu Liu, Yang College of Computing and Data Science Nanyang Technological University Singapore School of Computer Science Central China Normal University China University of Science and Technology of China China Institute of Automation Chinese Academy of Sciences China

Compared to other modalities, electroencephalogram (EEG) based emotion recognition can intuitively respond to emotional patterns in the human brain and, therefore, has become one of the most focused tasks in affective computing. The nature of emotions is a physiological and psychological state change in response to brain region connectivity, making emotion recognition focus more on the dependency between brain regions instead of specific brain regions. A significant trend is the application of graphs to encapsulate such dependency as dynamic functional connections between nodes across temporal and spatial dimensions. Concurrently, the neuroscientific underpinnings behind this dependency endow the application of graphs in this field with a distinctive significance. However, there is neither a comprehensive review nor a tutorial for constructing emotion-relevant graphs in EEG-based emotion recognition. In this paper, we present a comprehensive survey of these studies, delivering a systematic review of graph-related methods in this field from a methodological perspective. We propose a unified framework for graph applications in this field and categorize these methods on this basis. Finally, based on previous studies, we also present several open challenges and future directions in this field. Copyright © 2024, The Authors. All rights reserved.

关键词： Electroencephalography

PRISM: PRIVACY-PRESERVING INTER-SITE MRI HARMONIZATION VIA DISENTANGLED REPRESENTATION LEARNING

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Galada, Sarang Halder, Tanurima Deo, Kunal Krish, Ram P. Jadhav, Kshitij Trust Lab Indian Institute of Technology Bombay India Koita Centre for Digital Health Indian Institute of Technology Bombay India School of Computing and Data Science Sai University Chennai India

Multi-site MRI studies often suffer from site-specific variations arising from differences in methodology, hardware, and acquisition protocols, thereby compromising accuracy and reliability in clinical AI/ML tasks. We present PRISM (Privacy-preserving Inter-Site MRI Harmonization), a novel Deep Learning framework for harmonizing structural brain MRI across multiple sites while preserving data privacy. PRISM employs a dual-branch autoencoder with contrastive learning and variational inference to disentangle anatomical features from style and site-specific variations, enabling unpaired image translation without traveling subjects or multiple MRI modalities. Our modular design allows harmonization to any target site and seamless integration of new sites without the need for retraining or fine-tuning. Using multi-site structural MRI data, we demonstrate PRISM’s effectiveness in downstream tasks such as brain tissue segmentation and validate its harmonization performance through multiple experiments. Our framework addresses key challenges in medical AI/ML, including data privacy, distribution shifts, model generalizability and interpretability. Code is available at https://***/saranggalada/PRISM Copyright © 2024, The Authors. All rights reserved.

关键词： Contrastive Learning

A Novel Pipeline for Compressing Image Buffers in Remote Education Video Conferencing using Harris Corner Detection and Pixel Map Array

学校读者我要写书评

暂无评论

A Novel Pipeline for Compressing Image Buffers in Remote Edu...

International Conference on Electronics, Communication and computing Technologies (ICECCT)

作者： Gita Alekhya Paul Ansh Sharma Yashvardhan Jagnani Abhishek Saxena P Supraja Networking and Communications Department SRM Institute of Science and Technology Chennai India Computing Technologies Department SRM Institute of Science and Technology Chennai India Data Science and Business Systems Department SRM Institute of Science and Technology Chennai India

The COVID-19 pandemic has compelled educational institutions worldwide to shift to remote online education. Addressing the growing trend, an Oxford University Press report titled ‘Addressing the Deepening Digital Divide’ states that poor digital access is the most significant barrier to digital learning according to 68 percent of academicians. Students in many remote parts of India frequently have access to limited bandwidth internet, which is insufficient for the modern standards of network-hogging online video conference software solutions. This paper provides an algorithmic compression of image buffers to aid low-cost remote online video education. This compression can be done by translating the teacher's blackboard images to pixel arrays projected on a canvas on the student's dashboard while the instructor constantly communicates via real-time voice. The image is first converted to grayscale and dilated with a square kernel. Using Harris Corner Detector, probable board corners are identified and compared to a geometrical center of the points and the corners recovered by cornerSubPix. An adaptive threshold is employed, distinguishing the board's contents from the backdrop on the cropped picture based on the recovered points. The pixel-mapped array is then transmitted to the students through the webRTC real-time protocol, which includes support for two-way audio, allowing the teacher to deliver lectures. Using Canvas API on the application front-end, the array is projected onto the student's device as a dot matrix display. This paper has achieved an effective rate in the video transmission format, aiding online remote education on low-bandwidth network devices.

关键词：

ChronoFact: Timeline-based Temporal Fact Verification

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Barik, Anab Maulana Hsu, Wynne Lee, Mong Li School of Computing National University of Singapore Singapore Institute of Data Science National University of Singapore Singapore Centre for Trusted Internet and Community National University of Singapore Singapore

Temporal claims, often riddled with inaccuracies, are a significant challenge in the digital misinformation landscape. Fact-checking systems that can accurately verify such claims are crucial for combating misinformation. Current systems struggle with the complexities of evaluating the accuracy of these claims, especially when they include multiple, overlapping, or recurring events. This work introduces a novel timeline-based fact verification framework that identify events from both claim and evidences and organize them into their respective chronological timelines. The framework systematically examines the relationships between the events in both claim and evidence to predict the veracity of each claim event and their chronological accuracy. This allows us to accurately determine the overall veracity of the claim. We also introduce a new dataset of complex temporal claims involving timeline-based reasoning for the training and evaluation of our proposed framework. Experimental results demonstrate the effectiveness of our approach in handling the intricacies of temporal claim verification. © 2024, CC BY.

关键词： Spatio-temporal data

FedCal: achieving local and global calibration in federated learning via aggregated parameterized scaler 24

学校读者我要写书评

暂无评论

FedCal: achieving local and global calibration in federated ...

Proceedings of the 41st International Conference on Machine Learning

作者： Hongyi Peng Han Yu Xiaoli Tang Xiaoxiao Li College of Computing and Data Science Nanyang Technological University Singapore Department of Electrical and Computer Engineering The University of British Columbia Vancouver BC Canada and Vector Institute Canada

Federated learning (FL) enables collaborative machine learning across distributed data owners. However, this approach poses a significant challenge for model calibration due to data heterogeneity. While prior work focused on improving accuracy for non-iid data, calibration remains under-explored. This study reveals existing FL aggregation approaches lead to sub-optimal calibration, and theoretical analysis shows despite constraining variance in clients' label distributions, global calibration error is still asymptotically lower bounded. To address this, we propose a novel Federated Calibration (FedCal) approach, emphasizing both local and global calibration. It leverages client-specific scalers for local calibration to effectively correct output misalignment without sacrificing prediction accuracy. These scalers are then aggregated via weight averaging to generate a global scaler, minimizing the global calibration error. Extensive experiments demonstrate that FedCal significantly outperforms the best-performing baseline, reducing global calibration error by 47.66% on average.

关键词：

MQM-APE: Toward High-Quality Error Annotation Predictors with Automatic Post-Editing in LLM Translation Evaluators

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Lu, Qingyu Ding, Liang Zhang, Kanjian Zhang, Jinxia Tao, Dacheng Southeast University China The University of Sydney Australia Southeast University Shenzhen Research Institute China College of Computing and Data Science Nanyang Technological University 639798 Singapore

Large Language Models (LLMs) have shown significant potential as judges for Machine Translation (MT) quality assessment, providing both scores and fine-grained feedback. Although approaches such as GEMBA-MQM (Kocmi and Federmann, 2023a) have shown state-of-the-art performance on reference-free evaluation, the predicted errors do not align well with those annotated by human, limiting their interpretability as feedback signals. To enhance the quality of error annotations predicted by LLM evaluators, we introduce a universal and training-free framework, MQM-APE, based on the idea of filtering out non-impactful errors by Automatically Post-Editing (APE) the original translation based on each error, leaving only those errors that contribute to quality improvement. Specifically, we prompt the LLM to act as 1 evaluator to provide error annotations, 2 post-editor to determine whether errors impact quality improvement and 3 pairwise quality verifier as the error filter. Experiments show that our approach consistently improves both the reliability and quality of error spans against GEMBA-MQM, across eight LLMs in both high- and low-resource languages. Orthogonal to trained approaches, MQM-APE complements translation-specific evaluators such as Tower, highlighting its broad applicability. Further analysis confirms the effectiveness of each module and offers valuable insights into evaluator design and LLMs selection. © 2024, CC0.

关键词： Errors

A Provably Secure and Lightweight Access Control Protocol for EI-Based Vehicle to Grid Environment

学校读者我要写书评

暂无评论

IEEE Internet of Things Journal 2023年第18期10卷 16650-16657页

作者： Shamshad, Salman Mahmood, Khalid Shamshad, Usman Hussain, Ibrar Hussain, Shafiq Das, Ashok Kumar The University of Lahore Department of Software Engineering Lahore54590 Pakistan National Yunlin University of Science and Technology Graduate School of Intelligent Data Science Douliu64002 Taiwan Riphah International University Riphah School of Computing and Innovation Lahore54000 Pakistan University of Sahiwal Department of Computer Science Sahiwal57000 Pakistan International Institute of Information Technology Hyderabad Center for Security Theory and Algorithmic Research Hyderabad500032 India Old Dominion University Virginia Modeling Analysis and Simulation Center SuffolkVA23435 United States

The energy Internet (EI) presents a novel paradigm for renewable energy distribution that utilizes communication and computing technologies to revolutionize the conventional intelligent transportation systems (ITSs) and power grid into a structure that provides assistance to open innovation. The EI offers sustainable and bidirectional transmission for the improvement and analysis of energy convention among electric vehicles (EVs) and service providers. To ensure efficient, reliable, and secure operation, EI must be safe from security attacks. Therefore, efficient and secure key negotiation is a significant issue for the EI-based Vehicle-to-Grid (V2G) architecture. Thus, to ensure the system's security, we have devised a robust scheme to facilitate a secure key agreement between the entities involved for the secure and efficient renewable energy distribution for the EI-based energy vehicles in V2G. The devised scheme's robustness is solicited through the widely accepted formal random or real (RoR) model. In addition, the informal security analysis is conducted on the devised scheme, which is evident that the designed scheme achieves all the required security features of EI-based ITS. Moreover, the performance evaluation results endorse that the designed scheme achieves the desired security and minimizes the communication, computation, and energy overhead by 29.33%, 27.94%, and 28.08% in comparison to the existing competitive schemes. © 2014 IEEE.

关键词： Vehicle-to-grid