检索结果-内蒙古大学图书馆

2024 IEEE International Conference on Computer Vision and Machine Intelligence, CVMI 2024

作者： Sivabalamurugan, M. Swapna, T.R. Amrita Vishwa Vidyapeetham Amrita School of Computing Department of Computer Science and Engineering Coimbatore India

ISBN: (纸本)9798350376876

The rise of deepfake technology presents a critical challenge to the integrity of digital media, prompting the need for advanced forgery detection techniques. This paper proposes a novel approach for detecting deepfake image forgery by integrating local surface geometry analysis into the detection process. Leveraging the inherent alterations in local surface geometry caused by deepfake generation, our method extracts and merges this information with the original image, enhancing the discriminative features essential for accurate identification of manipulated content. Using classification networks such as Resnet, Densenet, and Efficient, our method achieves a remarkable accuracy rate of above 98 % in correctly identifying deepfake images for the classification models. This high accuracy underscores the effectiveness of our approach in detecting sophisticated deepfake content. By combining local surface geometry analysis with deep learning-based classification, our method represents a significant advancement in the field of deepfake detection, offering a promising solution to combat the proliferation of manipulated media and preserve the authenticity of digital imagery. © 2024 IEEE.

关键词： Image enhancement

来源：评论

学校读者我要写书评

暂无评论

Code Confabulator: Harnessing LLMs to Compile Code for Visualization 15

Code Confabulator: Harnessing LLMs to Compile Code for Visua...

引用

15th International Conference on computing Communication and Networking Technologies, ICCCNT 2024

作者： Rayvanth, Nannapaneni Pranavi, Pasupuleti Suryaa, E. Challa, Venkatahemant Kumar Reddy Belwal, Meena Amrita School of Computing Amrita Vishwa Vidyapeetham Department of Computer Science and Engineering Bengaluru India

ISBN: (纸本)9798350370249

Code visualization tools play a core role for visualization of the code and helping novices in understanding, analyzing, and optimization of their programs. Tools such as Algorithm Visualizer and PythonTutor are developed to make algorithm visualization possible across different programming languages like Python, Java, C, and C++. Despite their advantages, these tools suffer from unnecessary data collection, increased time complexity, and limitation in visualizing a few data structures. In our research, we want to introduce a more humanlike view of understanding and solve the shortcomings mentioned above by proposing a framework, which will make use of LLMs for extracting data from the code in Python, Java, C, or C++ to be visualized. In our framework, we have used LLMs instead of the traditional compiler for feasibility and adaptability. Our objective is the appropriateness evaluation of LLMs for this task, starting from freeform prompts. The LLM is asked to generate outputs in a given format;hence, these outputs are taken and converted into visualizations with a rule-based approach. The task will be done to enhance quality visualization and reduce time complexity while allowing the user to view the code at a human level view. We have also publicly released the framework so that people are able to provide further contributions and improvements. © 2024 IEEE.

关键词： Code visualization Large language Models

来源：评论

学校读者我要写书评

暂无评论

Tri-Model Synergy: Redefining Depression Detection through the Convergence of ANN, CNN, and SVM 5

Tri-Model Synergy: Redefining Depression Detection through t...

引用

5th International Conference on Smart Electronics and Communication, ICOSEC 2024

作者： George, Angelina Baniya, Achal Jose, Alphonsa Kumar, K. Dinesh Amrita School of Computing Amrita Vishwa Vidyapeetham Department of Computer Science & Engineering Bengaluru India

ISBN: (纸本)9798331504403

Depression being a universal mental health challenge that affects millions globally, necessitating reliable detection methods in order to aid in timely intervention. Like physical health, mental health is not to be ignored. This study presents a novel hybrid model that combines Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Support Vector Machines (SVM) to analyze and detect depression in individuals. The hybrid model achieved a training accuracy of 95.6% and a test accuracy of 91.2%, demonstrating its robustness and reliability. The total training time was recorded at 3.01 seconds, with ANN contributing 1.11 seconds and CNN contributing 1.9 seconds. Furthermore, the model's performance metrics included an F1-Score of 0.91 and 0.91 for detecting the classes of absence and presence of depression respectively, highlighting its effectiveness in correctly identifying depressive states. This hybrid approach provides a comprehensive analysis by leveraging the strengths of each individual model, paving the way for more reliable and scalable mental health assessment tools. © 2024 IEEE.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Ice hockey game using Deep Reinforcement Learning 9

Ice hockey game using Deep Reinforcement Learning

引用

9th IEEE International Conference for Convergence in Technology, I2CT 2024

作者： Mounika, Y. Prasanna, Y. Lakshmi Tarakaram, Y. Amudha, J. Aiswariya Milan, K. Amrita School of Computing Amrita Vishwa Vidyapeetham Department of Computer Science And Engineering Bengaluru India

ISBN: (纸本)9798350394474

Modern deep Reinforcement Learning (RL) techniques are quite good at choosing the best possible rules to maximise rewards. By using rich visual information for policy selection, this method combined with Deep Learning methodologies shows promise for difficult tasks. In this work, we have used Double DQN and also three variants of Duelling DQN to implement the ice hockey environment. The Duelling DQN outperformed the Double DQN and gave better results. © 2024 IEEE.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Malware Image Classification with Enhanced Feature Extraction Using VGG16 and Spatial Pyramid Pooling 8

Malware Image Classification with Enhanced Feature Extractio...

引用

8th IEEE International Conference on Information and Communication Technology, CICT 2024

作者： Ryan, Collin Diviya, M. Amrita School of Computing Department of Computer Science and Engineering Amrita Vishwa Vidyapeetham Chennai India

ISBN: (纸本)9798331531539

In the present era, the rise in number and complexity of malware poses a major threat leading to the rise in the importance of smarter Malware detection systems to understand and detect modern malwares. Modern systems are often very slow and inefficient with outdated machine learning models and feature selection techniques. These inefficiencies are overcome by the use of a multiclass detection system with CNN application to automatically classify malware. The system represents malware binaries in image format and uses VGG16 based network architecture with Spatial Pyramid Pooling (SPP) and ReLU activation to efficiently retrieve multiscale features from images of variable dimensionality. The model also includes dropout regularization, categorical cross entropy and Adam optimizer for better performance, adaptability and generalization. This approach works with a well balanced dataset with extensive training and achieved an accuracy of 95.94% making it a reliable and scalable solution for real time threat detection in a dynamic environment. © 2024 IEEE.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Hand-Written Bramhi Script Recognition using Deep Learning

Hand-Written Bramhi Script Recognition using Deep Learning

引用

2024 Asian Conference on Intelligent Technologies, ACOIT 2024

作者： Swaroop, Bandi Vishnu Prasad, Angala Hari Reddy, Busa Thanuj Sathwik Vardhan, Puvvadi Harsha Singh, Rimjhim Padam Palaniswamy, Suja Department of Computer Science and Engineering Amrita School of Computing Amrita Vishwa Vidyapeetham Bengaluru India

ISBN: (纸本)9798350374933

This work involves the classification of handwritten script using deep learning with a special interest in the comparison of different pre- trained models. As one of the oldest writing systems in the world, derived from the Dravidian and southeast Semitic scripts that were developed 4,500 years ago in India and Central America, Brahmi script also has variability due to the hand-written accented forms. To address these obstacles, different pre-trained models viz., Vision Transformer (ViT), MobileNetV2, VGG16, ResNet50, DenseNet, and EfficientNet with the CNN are implemented and compared. The result of the implementation infers that all of these pre- trained models yield a higher accuracy than the custom CNN traditional CNN approach, with a peak improvement in performance shown by MobileNetV2. This lays the ground for future utilization of deep learning architectures which have the capability of preserving and understanding handwritten symbolic scripts. © 2024 IEEE.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

Exploring Sarcasm Detection: Leveraging Neural Network Models with BERT Embeddings 4

Exploring Sarcasm Detection: Leveraging Neural Network Model...

引用

4th International Conference on Intelligent Technologies, CONIT 2024

作者： Rayvanth, N. Shreya Shree, S. Challa, VenkataHemant Kumar Reddy Sharma, Vishwash Venugopalan, Manju Amrita School of Computing Department of Computer Science and Engineering Amrita Vishwa Vidyapeetham Bengaluru India

ISBN: (纸本)9798350349900

Sarcasm is a form of verbal mockery, which poses a unique challenge for common people to understand, this also leads to difficulties in Natural Language Processing and sentiment analysis. This study deals with Sarcasm detection from textual data and a few machine learning and ensemble models were compared with the proposed Deep Learning architecture. BERT was used for the extraction of features and vectorization of textual data. The study utilizes curated datasets consisting of non-sarcastic and sarcastic samples sourced from Kaggle and Reddit platforms. The models will be evaluated using standard metrics. The best ML model accuracy achieved was 0.88 and an F1 score of 0.89 with a combination of Bert Embeddings and the SVM model. The proposed methodology using BERT embeddings a simple neural architecture achieved an accuracy of 0.92 on the test dataset with a F1-score of 0.91. It has a far lesser complexity than other well-known Deep learning architectures in use in other works. With the live streaming of The proposed Neural Network model achieved an accuracy of 0.94 and an F measure of 0.90 on the data streamed from Reddit. © 2024 IEEE.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Adversarial Modality Alignment Network for Cross-Modal Molecule Retrieval

IEEE Transactions on Artificial Intelligence

引用

IEEE Transactions on Artificial Intelligence 2024年第1期5卷 278-289页

作者： Zhao, Wenyu Zhou, Dong Cao, Buqing Zhang, Kai Chen, Jinjun Hunan University of Science and Technology School of Computer Science and Engineering Xiangtan425100 China Swinburne University of Technology Department of Computing Technologies MelbourneVIC3122 Australia Guangdong University of Foreign Studies School of Information Science and Technology Guangzhou510420 China

The cross-modal molecule retrieval (Text2Mol) task aims to bridge the semantic gap between molecules and natural language descriptions. A solution to this nontrivial problem relies on a graph convolutional network (GCN) and cross-modal attention with contrastive learning for reasonable results. However, there exist the following issues. First, the cross-modal attention mechanism is only in favor of text representations and cannot provide helpful information for molecule representations. Second, the GCN-based molecule encoder ignores edge features and the importance of various substructures of a molecule. Finally, the retrieval learning loss function is rather simplistic. This article further investigates the Text2Mol problem and proposes a novel adversarial modality alignment network (AMAN) based method to sufficiently learn both description and molecule information. Our method utilizes a SciBERT as a text encoder and a graph transformer network as a molecule encoder to generate multimodal representations. Then, an adversarial network is used to align these modalities interactively. Meanwhile, a triplet loss function is leveraged to perform retrieval learning and further enhance the modality alignment. Experiments on the ChEBI-20 dataset show the effectiveness of our AMAN method compared with baselines. © 2020 IEEE.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

Augmented Super Resolution GAN (ASRGAN) for Image Enhancement Through Reinforced Discriminator 5th

Augmented Super Resolution GAN (ASRGAN) for Image Enhanceme...

引用

5th International Conference on Deep Learning, Artificial Intelligence and Robotics, ICDLAIR 2023

作者： Bagavathi, C. Gadde, Uday Kiran Immadisetty, Ruthin Gupta Vasireddy, Ananth Krishna Department of Computer Science and Engineering Amrita School of Computing Amrita Vishwa Vidyapeetham Coimbatore India

ISBN: (纸本)9783031609343

Image super-resolution improves an image resolution and provides better detail. Super resolution achieved through Generative Adversarial Networks has been found to produce good resolution but has few challenges such as discrimination ability, stability and convergence of SRGAN. To address these challenges, we propose an Augmented Super resolution Generative Adversarial Network (ASRGAN) that incorporates reinforcement learning (RL) techniques. In ASRGAN, we introduce an RL agent into the discriminator component to enhance its discrimination capability and adapt its behavior to optimize the training process. The proposed ASRGAN signifies that the SRGAN has been expanded with reinforcement learning, which can have implications for the training stability, discrimination ability, convergence, and fine-grained control of the super resolution process. The generalization capabilities of ASRGAN across diverse image types, highlighting its adaptability has also been brought through this work. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

Interactive Code Parsing and Expression Tree Visualization

Interactive Code Parsing and Expression Tree Visualization

引用

2024 International Conference on Innovative computing, Intelligent Communication and Smart Electrical Systems, ICSES 2024

作者： Akathya, Ve.Ram Suresh, Yadukul Shivadutt, Sahen Kavitha, C.R. Amrita School of Computing Amrita Vishwa Vidyapeetham Department of Computer Science and Engineering Bengaluru India

ISBN: (纸本)9798331543617

Binary information trees are key visualization and evaluation tools, particularly during the passing of algorithms and synthesizing designs. This paper is an interactive tool for producing binary information trees over time. This tool will allow the user to input variables, selected, truncating and then show the same as a binary tree. Using Network X and Matplotlib for graphing, this presents the user with an interactive pie chart to enable operator control and variable assignment. Dynamic information evaluation and substitution is supported;this allows for an immediate view of change when variables and functions are altered. It is aimed at user interaction with a didactic value, and with this in mind, it contributes to easier understanding of the fundamental issues of distributed algorithms, clustering, and information evaluation. Though derived from such existing tools as VMEXT and PAVT, it offers a simple, and thus very practical approach, becoming suitable for educational environments where a visualization shows better understanding of the abstract computer concepts. © 2024 IEEE.

关键词： Trees (mathematics)

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：