检索结果-内蒙古大学图书馆

Deep Minimax Classifiers for Imbalanced Datasets With a Small Number of Minority Samples

IEEE Journal on Selected Topics in Signal Processing 2025年第3期19卷 491-506页

作者： Choi, Hansung Seo, Daewon Electrical Engineering and Computer Science DGIST Daegu42988 Korea Republic of

The concept of a minimax classifier is well-established in statistical decision theory, but its implementation via neural networks remains challenging, particularly in scenarios with imbalanced training data having a limited number of samples for minority classes. To address this issue, we propose a novel minimax learning algorithm designed to minimize the risk of worst-performing classes. Our algorithm iterates through two steps: a minimization step that trains the model based on a selected target prior, and a maximization step that updates the target prior towards the adversarial prior for the trained model. In the minimization, we introduce a targeted logit-adjustment loss function that efficiently identifies optimal decision boundaries under the target prior. Moreover, based on a new prior-dependent generalization bound that we obtained, we theoretically prove that our loss function has a better generalization capability than existing loss functions. During the maximization, we refine the target prior by shifting it towards the adversarial prior, depending on the worst-performing classes rather than on per-class risk estimates. Our maximization method is particularly robust in the regime of a small number of samples. Additionally, to adapt to overparameterized neural networks, we partition the entire training dataset into two subsets: one for model training during the minimization step and the other for updating the target prior during the maximization step. Our proposed algorithm has a provable convergence property, and empirical results indicate that our algorithm performs better than or is comparable to existing methods. © 2007-2012 IEEE.

关键词： Adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

Debiasing vision-language models for vision tasks:a survey

引用

Frontiers of computer science 2025年第1期19卷 175-177页

作者： Beier ZHU Hanwang ZHANG School of Computer Science and Engineering Nanyang Technological UniversitySingapore 639798Singapore

1 Introduction In recent years,foundation Vision-Language Models(VLMs),such as CLIP[1],which empower zero-shot transfer to a wide variety of domains without fine-tuning,have led to a significant shift in machine learning *** the impressive capabilities,it is concerning that the VLMs are prone to inheriting biases from the uncurated datasets scraped from the Internet[2–5].We examine these biases from three perspectives.(1)Label bias,certain classes(words)appear more frequently in the pre-training data.(2)Spurious correlation,non-target features,e.g.,image background,that are correlated with labels,resulting in poor group robustness.(3)Social bias,which is a special form of spurious correlation,focuses on societal *** image-text pairs might contain human prejudice,e.g.,gender,ethnicity,and age,that are correlated with *** biases are subsequently propagated to downstream tasks,leading to biased predictions.

关键词： tasks image tuning

来源：评论

学校读者我要写书评

暂无评论

Synthesis of batik motifs using a diffusion - generative adversarial network

引用

Multimedia Tools and Applications 2025年第7期84卷 3407-3438页

作者： Octadion, One Yudistira, Novanto Kurnianingtyas, Diva Informatics Engineering Faculty of Computer Science Universitas Brawijaya Malang Indonesia

This paper presents a novel approach for generating intricate Batik motifs using a modified Diffusion-Generative Adversarial Network (Diffusion-GAN) augmented with StyleGAN2-Ada. Motivated by the rich cultural heritage of Indonesian Batik, our research addresses the challenge of synthesizing high-quality, diverse patterns that capture the artistry and complexity of traditional designs. Traditional generative models often struggle with stability and fidelity in artistic synthesis. We integrate StyleGAN2-Ada and Diffusion techniques to overcome these limitations, optimizing model architecture and employing a curated Batik dataset. Evaluation metrics including Frechet Inception Distance (FID), Kernel Inception Distance (KID), precision, recall, and non-redundancy assess the quality and diversity of generated motifs. Our results demonstrate significant advancements in the realism and authenticity of synthesized Batik patterns, leveraging Diffusion-GAN to enhance detail and artistic fidelity. By exploring the intersection of deep learning and cultural art, this research contributes to the preservation and evolution of Batik as a cultural artifact. References to prior works underscore the evolution of Artificial Intelligence (AI) in cultural heritage preservation, highlighting the potential of modern AI techniques to innovate within traditional art forms like Batik. In summary, this study showcases the efficacy of the proposed Diffusion-GAN methodology in synthesizing original and convincing Batik motifs, while identifying opportunities for further refinement and complexity in motif synthesis. The dataset and code can be accessed here: https://***/octadion/diffusion-stylegan2-ada-pytorch © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2025.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

Deep learning approaches to address cold start and long tail challenges in recommendation systems: a systematic review

引用

Multimedia Tools and Applications 2025年第5期84卷 2293-2325页

作者： Jangid, Manisha Kumar, Rakesh Department of Computer Science & Engineering Central University of Haryana Haryana India

Recommendation systems (RS) have become prevalent across different domains including music, e-commerce, e-learning, entertainment, and social media to address the issue of information overload. While traditional RS approaches have achieved significant success in delivering recommendations, they still face issues including sparse data, diversity, cold start, and long tail problem. The emergence of deep learning as a prominent and extensively studied topic has shown significant potential in addressing these challenges in RS. Deep learning captures intricate patterns of interaction and precisely reflects user preferences, allowing for encoding complex abstractions in data representation and enhancing information processing capabilities. This paper provides an extensive survey of the existing literature on recommendation systems. We will begin by providing a foundational understanding of the core concepts and terminology of recommendation systems and significance of deep learning. Secondly, we talk about the original studies being conducted on deep learning methods and solutions to address "Cold start and long tail" challenges in recommendation. Thirdly, we examine the potential future directions of research pertaining to deep learning-based recommender systems (DLRS). Our review provides valuable insights for both researchers and practitioners in using deep learning to address challenges in recommendation and in developing effective and efficient recommendation systems. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Implementation of ANN and response surface method for dual-fuel CI engine optimization and prediction using water infusion and biofuel

引用

Multimedia Tools and Applications 2025年 1-43页

作者： Pramanik, Sabyasachi Department of Computer Science & Engineering Haldia Institute of Technology Haldia India

In this study, tests were done to see what would happen if hydrogen (H2) and lemon grass oil (LO) were used for a lone-cylinder compression ignition engine as a partial diesel replacement. After starting the trial with pure diesel, the engine was modified to run on two fuels by adding H2 through the inlet pipe and LO as direct injection (DI) gasoline. In this study, the energy fraction of H2 ranged from 5 to 10%. A 3% volume mixture of water (W) and diesel-LO mixtures was used to study how NOX emissions were affected. Using a common back- transmission method, an Artificial Neural Network model was created to forecast the correlation between engine output reactions and input variables (load, LO, and H2). The goal of response surface methodology is for improve the motor input settings for low pollution and high heat efficiency. The trials were designed using response surface methodology (RSM), and the optimization procedure was conducted using RSM. An artificial neural network (ANN) was used to predict the test engine’s emission patterns and performance. The results showed that ANN and RSM were both very good and accurate modeling strategies. When 10% H2 + LO25 was used, the BTE rose to 31%, up from 20% for clean LO and 28% for 12% H2 + LO25 + 2%W means. Compared to clean LO operation, the cylinder pressure raised in 15% H2 mode by up to 63 psi. The measured value for the HRR, under H2 premixed combustion framework was 81 k Joule/CA deg in 10% H2, that is significantly greater than the value for every LO processes. The highest NOX release for 10% H2 + LO 25% was discovered at 690 ppm peak load, while it dropped to 645 ppm in gasoline that had been emulsified in water. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2025.

关键词： Gasoline

来源：评论

学校读者我要写书评

暂无评论

An ensemble based approach for violence detection in videos using deep transfer learning

引用

Multimedia Tools and Applications 2025年第12期84卷 11001-11025页

作者： Kaur, Gurmeet Singh, Sarbjeet Department of Computer Science and Engineering UIET Panjab University Chandigarh India

The detection of violence in videos has become an extremely valuable application in real-life situations, which aim to maintain and protect people’s safety. Despite the complexities inherent in videos and the abrupt nature of violent actions, the field has seen several approaches, yet achieving consistent performance remains elusive, especially with advanced real-life datasets. Presenting a solution, the paper proposes a Bagging ensemble based approach comprising three pretrained models integrated with stacked Long Short-Term Memory (LSTM) to enhance individual model performance. This ensemble approach is rigorously analyzed on two publicly accessible datasets, RLVS and RWF-2000, providing remarkable accuracy (96.6%, 92.7%) and F1-scores (96.6%, 93.0%). Additionally, a cross-dataset analysis demonstrates the model’s ability to generalize across diverse datasets. Furthermore, a study of ablation highlighting the efficacy and optimal selection of components in augmenting the proposed ensemble’s efficiency. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Long short-term memory

来源：评论

学校读者我要写书评

暂无评论

Evaluation of phone posterior probabilities for pathology detection in speech data using deep learning models

引用

International Journal of Speech Technology 2025年第1期28卷 99-116页

作者： Farazi, Sahar Shekofteh, Yasser Faculty of Computer Science and Engineering Shahid Beheshti University Tehran Iran

Voice pathology detection (VPD) aims to accurately identify voice impairments by analyzing speech signals. This study proposes models based on deep learning (DL) for binary classification to distinguish between healthy and pathological voices, with a unique focus on the integration of phone posterior probabilities (PPPs as phonetic-based features) alongside Mel-frequency cepstral coefficients (MFCCs as acoustic-based features) for input models. By incorporating PPPs as supplementary information, we investigate the model’s performance across spontaneous, sustained vowel, and read speech datasets, addressing the gap in comparing these speech types for VPD. Our results highlight that PPPs significantly enhance classification accuracy, particularly for read speech and spontaneous speech data types. Using the AVFAD database, we show that the proposed CNN-based model achieves its highest performance on spontaneous speech, with an accuracy of approximately 87% on test data and 93% on validation data. This study emphasizes the impact of PPPs as phonetic-based features in VPD tasks and clarifies which types of speech benefit most from their inclusion, paving the way for more refined models in this field. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2025.

关键词： AVFAD dataset Convolutional neural network Deep learning Phone posterior probabilities Spontaneous speech Voice pathology detection

来源：评论

学校读者我要写书评

暂无评论

Phrase-level emotion intensity detection of text using lexicon-based unit circle and pipelined neural networks approaches

引用

Neural Computing and Applications 2025年第5期37卷 3151-3166页

作者： Patil, Bhuvaneshwari Ajit Vadlakunta, Vishnu Vardhan Guruvu, Poojaa Pranathi Mathew, Meenu Prakash, Jay Computer Science and Engineering National Institute of Technology Calicut Kozhikode India

Emotions have a significant impact on how people make decisions. Due to its potential applications in various fields, emotion intensity detection has attracted a lot of attention recently. Several methods have been proposed in the past using natural language processing to recognize emotions from the text. These methods include the keyword-based approach, the lexicon-based approach and the machine learning approaches. Nevertheless, there were a few limitations with the lexicon-based unit circle approach, as it focuses on detecting emotion intensity at the word-level. This research proposes an integrated model by considering the advantages of the lexicon-based unit circle and pipelined neural networks approaches. The model is trained on a large corpus of text data. The proposed approach aims to determine emotion intensity, analyzing text at the phrase-level, thereby achieving higher performance in terms of accuracy and F1-score compared to other models. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.

关键词： Neural networks

来源：评论

学校读者我要写书评

暂无评论

Dual-Channel Multiplex Graph Neural Networks for Recommendation

引用

IEEE Transactions on Knowledge and Data engineering 2025年第6期37卷 3327-3341页

作者： Li, Xiang Fu, Chaofan Zhao, Zhongying Zheng, Guangjie Huang, Chao Yu, Yanwei Dong, Junyu Ocean University of China Faculty of Information Science and Engineering Qingdao266005 China Shandong University of Science and Technology College of Computer Science and Engineering Qingdao266590 China Shanghai Jiao Tong University Department of Computer Science and Engineering Shanghai200240 China The University of Hong Kong Department of Computer Science Hong Kong

Effective recommender systems play a crucial role in accurately capturing user and item attributes that mirror individual preferences. Some existing recommendation techniques have started to shift their focus towards modeling various types of interactive relations between users and items in real-world recommendation scenarios, such as clicks, marking favorites, and purchases on online shopping platforms. Nevertheless, these approaches still grapple with two significant challenges: (1) Insufficient modeling and exploitation of the impact of various behavior patterns formed by multiplex relations between users and items on representation learning, and (2) ignoring the effect of different relations within behavior patterns on the target relation in recommender system scenarios. In this work, we introduce a novel recommendation framework, Dual-Channel Multiplex Graph Neural Network (DCMGNN), which addresses the aforementioned challenges. It incorporates an explicit behavior pattern representation learner to capture the behavior patterns composed of multiplex user-item interactive relations, and includes a relation chain representation learner and a relation chain-aware encoder to discover the impact of various auxiliary relations on the target relation, the dependencies between different relations, and mine the appropriate order of relations in a behavior pattern. Extensive experiments on three real-world datasets demonstrate that our DCMGNN surpasses various state-of-the-art recommendation methods. It outperforms the best baselines by 10.06% and 12.15% on average across all datasets in terms of Recall@10 and NDCG@10 respectively. The source code of our paper is available at https://***/lx970414/TKDE-DCMGNN. © 1989-2012 IEEE.

关键词： Graph neural networks

来源：评论

学校读者我要写书评

暂无评论

MTUNet + + : explainable few-shot medical image classification with generative adversarial network

引用

Multimedia Tools and Applications 2025年第11期84卷 9219-9239页

作者： Titoriya, Ankit Kumar Singh, Maheshwari Prasad Singh, Amit Kumar Department of Computer Science and Engineering National Institute of Technology Patna India

Medical imaging, a cornerstone of disease diagnosis and treatment planning, faces the hurdles of subjective interpretation and reliance on specialized expertise. Deep learning algorithms show improvements in automating medical image analysis, reducing radiologists' burden, and potentially enhancing patient outcomes. However, these algorithms require substantial quantities of high-quality labelled data for effective training and refinement. This paper proposes an innovative approach that harnesses few-shot learning (FSL) and generative adversarial networks (GANs) to overcome conventional methods' limitations in medical image classification. FSL, capable of learning from limited labelled examples, holds promise for scenarios where labelled data is scarce. However, the lack of interpretability in existing FSL models impedes their clinical adoption. To tackle this, this paper proposes a explainable FSL network, "MTUNet + +," which integrates an attention mechanism to emphasize relevant regions in medical images. Furthermore, integrating a generative adversarial network, enhances the performance of MTUNet + + by generating synthetic medical images. Systematically eliminating misleading synthetic images improves the reliability and accuracy of medical image classification. Empirical evaluation on benchmark datasets underscores the effectiveness of the approach, achieving 85.19% and 69.28% accuracy for the HAM10000 and Kvasir datasets, respectively. This paper contributes to advancing AI-driven solutions in clinical practice, facilitating enhanced patient care and streamlined workflows within real-world healthcare settings. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Generative adversarial networks

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：