检索结果-内蒙古大学图书馆

IEEE International Symposium on Industrial Electronics (ISIE)

作者： Cheng-Fu Liou Yi-Cheng Yu Research Labs (MZAIR Labs) Meta Zen Artificial Intelligence Tainan Taiwan Graduate Degree Program of College of Electrical and Computer Engineering National Yang Ming Chiao Tung University Hsinchu Taiwan

ISBN: (数字)9798350394085

ISBN: (纸本)9798350394092

Comparing the structures of current neural networks and the biological brain, it can be observed that many mechanisms correspond to each other according to their functions. From the perspective of the biological brain, different components have different long-term or short-term memory effects, such as the hippopotamus. However, current neural networks do not pay much attention to this aspect. In this work, we incorporated many hypotheses or theories inspired by the territory of neurobiology to retain the learned knowledge. To start with, we follow the inspiration of the synaptic homeostasis hypothesis (SHY) [1] and add an additional training stage to the training process of the model, which can ensure that only the most important information remains intact and the insignificant synapse can be pruned. In other words, we divide the overall training process into two learning stages: synaptogenesis and synaptic sparsifying. We experimentally demonstrate that our novel learning strategy significantly outperforms the traditional solution in training a sequence of tasks at different times on three public datasets, which supports that the proposed method is more efficient for resource-limited edge devices.

关键词： Training Industrial electronics Neurons Homeostasis Task analysis Biological neural networks Synapses

来源：评论

学校读者我要写书评

暂无评论

DAFA: DISTANCE-AWARE FAIR ADVERSARIAL TRAINING

arXiv

引用

arXiv 2024年

作者： Lee, Hyungyu Lee, Saehyung Jang, Hyemi Park, Junsung Bae, Ho Yoon, Sungroh Electrical and Computer Engineering Seoul National University Korea Republic of Department of Cyber Security Ewha Womans University Korea Republic of Interdisciplinary Program in Artificial Intelligence Seoul National University Korea Republic of

The disparity in accuracy between classes in standard training is amplified during adversarial training, a phenomenon termed the robust fairness problem. Existing methodologies aimed to enhance robust fairness by sacrificing the model's performance on easier classes in order to improve its performance on harder ones. However, we observe that under adversarial attacks, the majority of the model's predictions for samples from the worst class are biased towards classes similar to the worst class, rather than towards the easy classes. Through theoretical and empirical analysis, we demonstrate that robust fairness deteriorates as the distance between classes decreases. Motivated by these insights, we introduce the Distance-Aware Fair Adversarial training (DAFA) methodology, which addresses robust fairness by taking into account the similarities between classes. Specifically, our method assigns distinct loss weights and adversarial margins to each class and adjusts them to encourage a trade-off in robustness among similar classes. Experimental results across various datasets demonstrate that our method not only maintains average robust accuracy but also significantly improves the worst robust accuracy, indicating a marked improvement in robust fairness compared to existing methods. Copyright © 2024, The Authors. All rights reserved.

关键词： Economic and social effects

来源：评论

学校读者我要写书评

暂无评论

Accumulating Curriculum Knowledge Based on Neural Network Isolated Algorithm

Accumulating Curriculum Knowledge Based on Neural Network Is...

引用

IEEE International Symposium on Industrial Electronics (ISIE)

作者： Cheng-Fu Liou Yi-Cheng Yu Meta Zen Artificial Intelligence Research Labs (MZAIR Labs) Tainan Taiwan Graduate Degree Program of College of Electrical and Computer Engineering National Yang Ming Chiao Tung University Hsinchu Taiwan

ISBN: (数字)9798350394085

ISBN: (纸本)9798350394092

Feeding the training data to neural networks in a specific sequence, from the simple data to the difficult data, utilizing curriculum learning can enhance performance improvements over the typical learning strategy based on shuffling training samples randomly, without any additional computational costs. This training approach has been successfully applied in all fields of machine learning. However, the current curriculum learning will gradually introduce difficult samples until the model eventually performs joint training. In addition, our experimental results find and demonstrate how the order of training tasks arranged according to the difficulty of different tasks influences the reuse percentage of model capacity, allowing us to observe the impact of curriculum learning on model performance from different perspectives. Finally, we also separately conduct three ablations analyses of our model to comprehensively enhance the understanding of the impacts of the proposed approach and further demonstrate its scalability to a standard large-scale dataset, i.e., the ImageNet dataset.

关键词： Training Industrial electronics Scalability Knowledge based systems Training data Machine learning Multitasking

来源：评论

学校读者我要写书评

暂无评论

On mitigating stability-plasticity dilemma in CLIP-guided image morphing via geodesic distillation loss

arXiv

引用

arXiv 2024年

作者： Oh, Yeongtak Lee, Saehyung Hwang, Uiwon Yoon, Sungroh Department of Electrical and Computer Engineering Seoul National University Korea Republic of Interdisciplinary Program in Artificial Intelligence Seoul National University Korea Republic of Division of Digital Healthcare Yonsei University Korea Republic of

Large-scale language-vision pre-training models, such as CLIP, have achieved remarkable text-guided image morphing results by leveraging several unconditional generative models. However, existing CLIP-guided image morphing methods encounter difficulties when morphing photorealistic images. Specifically, existing guidance fails to provide detailed explanations of the morphing regions within the image, leading to misguidance. In this paper, we observed that such misguidance could be effectively mitigated by simply using a proper regularization loss. Our approach comprises two key components: 1) a geodesic cosine similarity loss that minimizes inter-modality features (i.e., image and text) on a projected subspace of CLIP space, and 2) a latent regularization loss that minimizes intra-modality features (i.e., image and image) on the image manifold. By replacing the naïve directional CLIP loss in a drop-in replacement manner, our method achieves superior morphing results on both images and videos for various benchmarks, including CLIP-inversion. © 2024, CC BY.

关键词： Distillation

来源：评论

学校读者我要写书评

暂无评论

SF(DA)2: SOURCE-FREE DOMAIN ADAPTATION THROUGH THE LENS OF DATA AUGMENTATION

arXiv

引用

arXiv 2024年

作者： Hwang, Uiwon Lee, Jonghyun Shin, Juhyeon Yoon, Sungroh Division of Digital Healthcare Yonsei University Korea Republic of Department of Electrical and Computer Engineering Seoul National University Korea Republic of Interdisciplinary Program in Artificial Intelligence Seoul National University Korea Republic of

In the face of the deep learning model’s vulnerability to domain shift, source-free domain adaptation (SFDA) methods have been proposed to adapt models to new, unseen target domains without requiring access to source domain data. Although the potential benefits of applying data augmentation to SFDA are attractive, several challenges arise such as the dependence on prior knowledge of class-preserving transformations and the increase in memory and computational requirements. In this paper, we propose Source-free Domain Adaptation Through the Lens of Data Augmentation (SF(DA)2), a novel approach that leverages the benefits of data augmentation without suffering from these challenges. We construct an augmentation graph in the feature space of the pretrained model using the neighbor relationships between target features and propose spectral neighborhood clustering to identify partitions in the prediction space. Furthermore, we propose implicit feature augmentation and feature disentanglement as regularization loss functions that effectively utilize class semantic information within the feature space. These regularizers simulate the inclusion of an unlimited number of augmented target features into the augmentation graph while minimizing computational and memory demands. Our method shows superior adaptation performance in SFDA scenarios, including 2D image and 3D point cloud datasets and a highly imbalanced dataset. Copyright © 2024, The Authors. All rights reserved.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

Fireworks Image Classification with Deep Learning 26

Fireworks Image Classification with Deep Learning

引用

26th International Conference on Technologies and Applications of artificial intelligence, TAAI 2021

作者： Chang, Chih-Lung Tseng, Hsin-Ming Chu, Hsueh-Ting Asia University Department of Computer Science and Information Engineering Taichung Taiwan Asia University Ph.D. Program in Artificial Intelligence Taichung Taiwan

ISBN: (纸本)9781665408257

Before the advent of deep learning, traditional image recognition used algorithms to find features and then classify images using classical machine learning algorithms. But it is difficult to define features for variable types of images. Instead, Representation learning is a way to allow a system to discover the representations of feature for image processing. Deep-learning algorithms attempt to learn multiple levels of representation and play the key role of modern Representation learning. When deep learning neural networks recognize images, the shallow layers usually extract lower-level features, then start with intermediate-level features, and finally the full images. This study uses CNN (Convolutional Neural Network) to classify fireworks images, and the trained modules are used to evaluate the accuracy and suitability. © 2021 IEEE.

关键词： Image classification

来源：评论

学校读者我要写书评

暂无评论

MERNet: Multimodal Emotion Recognition in Conversation Network via Cross-Modal Knowledge Distillation and Contrastive Learning

MERNet: Multimodal Emotion Recognition in Conversation Netwo...

引用

International Conference on Big Data and Smart Computing (BIGCOMP)

作者： Sooyeon Hwang Gahgene Gweon Eunseong Kwon Interdisciplinary Program in Artificial Intelligence Seoul National University South Korea Graduate School of Convergence Science and Technology Seoul National University South Korea Dept. of Computer Science and Engineering POSTECH South Korea

ISBN: (数字)9798331529024

ISBN: (纸本)9798331529031

Multimodal Emotion Recognition in Conversation (ERC) is a task of predicting the emotion of each utterance in a conversation by utilizing both verbal and non-verbal modalities. However, existing approaches often struggle to bridge cross-modal gaps, resulting in misaligned features and frequent misclassification of minority emotions into semantically similar majority emotions. To address these challenges, we propose MERNet, a framework that employs cross-modal knowledge distillation and contrastive learning to align multimodal features and effectively distinguish subtle emotions in conversations. Our framework consists of two stages: 1) guiding non-verbal modalities with the text modality to transfer knowledge and align their features, and 2) applying contrastive learning with emotion labels as anchors to distinguish subtle differences between similar emotions and address the class imbalance problem. Experiments conducted on two benchmark datasets, IEMOCAP and MELD, demonstrate that our MERNet outperforms existing state-of-the-art models.

关键词： Knowledge engineering Bridges Emotion recognition Visualization Computational modeling Contrastive learning Benchmark testing Linguistics Big Data

来源：评论

学校读者我要写书评

暂无评论

MERNet: Multimodal Emotion Recognition in Conversation Network via Cross-Modal Knowledge Distillation and Contrastive Learning

MERNet: Multimodal Emotion Recognition in Conversation Netwo...

引用

2025 IEEE International Conference on Big Data and Smart Computing, BigComp 2025

作者： Hwang, Sooyeon Gweon, Gahgene Kwon, Eunseong Seoul National University Interdisciplinary Program in Artificial Intelligence Korea Republic of Seoul National University Graduate School of Convergence Science and Technology Korea Republic of POSTECH Dept. of Computer Science and Engineering Korea Republic of

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Evaluating the Impact of Data Cleaning for Improving Road Damage Detection Through Vision Transformer

Evaluating the Impact of Data Cleaning for Improving Road Da...

引用

2025 IEEE International Conference on Consumer Electronics, ICCE 2025

作者： Desman, Andy Muchtar, Kahlil Oktiana, Maulisa Fitria, Maya Away, Yuwaldi Fardian, Fardian Riza, Hammam Dept. of Electrical and Computer Engineering Universitas Syiah Kuala Banda Aceh Indonesia Banda Aceh Indonesia Indonesia The Master Program in Artificial Intelligence The Department of Informatics Universitas Syiah Kuala Banda Aceh Indonesia

ISBN: (纸本)9798331521165

The WHO predicts that by 2030 road accidents will be the 5th leading cause of death. Globally, road accidents account for 1.25 million casualties each year, and road defects cause 34% of these casualties. The road survey process in many countries have several challenges, one of which is detection using cameras that do not have a recognition system. In this study, a model with YOLOS architecture based on Vision Transformer trained on the RDD2022 dataset successfully recognizes road damage well, as indicated by the number of objects detected, bounding box on accurate objects, and the ability to recognize objects with inconsistent shadow and light inference. This research uses assessment parameters such as Average Precision (AP) and Average Recall (AR) to determine the overall performance of the model. The model achieves the highest AP value at Intersection of Union (IoU) 0.5, 0.75, and 0.5-0.95, worth 62.1%, 37.1%, and 36.2% respectively, and the highest AR value in Large, Medium, and Small Areas, worth 42.1%, 60.3%, and 75.4% respectively. The supplementary material can be found through this link: https://***/watch?v=LzkI2e_IORE. © 2025 IEEE.

关键词： Traffic surveys

来源：评论

学校读者我要写书评

暂无评论

3D Facial Reconstruction Applied to Medical Cosmetology

3D Facial Reconstruction Applied to Medical Cosmetology

引用

IEEE Eurasia Conference on Biomedical engineering, Healthcare and Sustainability (ECBIOS)

作者： Sheng-Hsien Hsieh Ming-Yi Lin Ching-Han Chen Program of Artificial Intelligence and Mechatronics National Pingtung University of Science and Technology Pingtung Taiwan Department of Computer Science and Engineering Yuan Ze University Taoyuan Taiwan Department of Computer Science and Engineering National Central University Taoyuan Taiwan

We present a high-accuracy 3D facial reconstruction system with the following features: real-time 3D facial reconstruction using exposure synchronization multi-camera, feature alignment to quantify facial differences, a software system based on a naked eye 3D display, and medical records for a 3D face database. The proposed system is novel and practical, and the algorithm and the hardware/software architecture improve the current non-quantitative communication status in medical aesthetics. The system achieves preoperative expectations, real-time recording and feedback during surgery, and postoperative tracking analysis, thus making doctor-patient communication more efficient. The facial data database records personal and quantitative beauty data and its changes, providing accurate medical and aesthetic treatment analysis for individuals. In addition, the collected data of different genders, ages, injection sites, and dosages contribute to the development of more accurate medical materials. With artificial intelligence and cloud architecture, we provide reliable data for medical aesthetics customers, helping them understand their medical aesthetic needs. Doctors also provide stable medical care quality to consumers through cloud data, and medical aesthetic consultants do not need to exaggerate medical effects excessively. They can inform customers of specific differences before and after surgery through precise data and provide visual communication and customer expansion to improve consultation conversion rates.

关键词：

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：