检索结果-内蒙古大学图书馆

International Conference on Computer Vision (ICCV)

作者： Sifan Long Zhen Zhao Junkun Yuan Zichang Tan Jiangjiang Liu Luping Zhou Shengsheng Wang Jingdong Wang College of Computer Science and Technology Jilin University Jilin China Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University Jilin China Baidu VIS University of Sydney Zhejiang University

Prompt learning has become one of the most efficient paradigms for adapting large pre-trained vision-language models to downstream tasks. Current state-of-the-art methods, like CoOp and ProDA, tend to adopt soft prompts to learn an appropriate prompt for each specific task. Recent CoCoOp further boosts the base-to-new generalization performance via an image-conditional prompt. However, it directly fuses identical image semantics to prompts of different labels and significantly weakens the discrimination among different classes as shown in our experiments. Motivated by this observation, we first propose a class-aware text prompt (CTP) to enrich generated prompts with label-related image information. Unlike CoCoOp, CTP can effectively involve image semantics and avoid introducing extra ambiguities into different prompts. On the other hand, instead of reserving the complete image representations, we propose text-guided feature tuning (TFT) to make the image branch attend to class-related representation. A contrastive loss is employed to align such augmented text and image representations on downstream tasks. In this way, the image-to-text CTP and text-to-image TFT can be mutually promoted to enhance the adaptation of VLMs for downstream tasks. Extensive experiments demonstrate that our method outperforms the existing methods by a significant margin. Especially, compared to CoCoOp, we achieve an average improvement of 4.03% on new classes and 3.19% on harmonic-mean over eleven classification benchmarks.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Multi-Task Self-Supervised Learning for Medical Image Segmentation

Multi-Task Self-Supervised Learning for Medical Image Segmen...

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Bo Wang Hang Zhao Xiongfei Li Mingjie Tian Bo Huang Feiyang Yang Intelligent Information Processing Laboratory Yanbian University Yanji China College of Computer Science and Technology Jilin University Changchun China Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University Changchun China School of Artificial Intelligence Jilin University Changchun China

Although medical image segmentation has achieved remarkable results with supervised learning, obtaining labeled data remains challenging and costly. To counteract this, we present the MTSPSeg, a multi-task self-supervised learning framework. We establish the dynamic gradient learning rate (DGLR) strategy and the dynamic multi-task loss weight adjustment method (DWL) to relieve potential early convergence and diverse learning difficulties of self-supervised learning. Our method deeply integrates the prior knowledge inherent in medical images, facilitates collaborative learning among tasks, and delves further into the latent information within the images. Experimental results demonstrate that regardless of the proportion of labeled data, our method outperforms baseline models on the KiTS19 and LiTS datasets, achieving state-of-the-art performance on the KiTS19 dataset.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Closed Loop Networks for Open-Set Semi-Supervised Learning

SSRN

引用

SSRN 2023年

作者： Ouyang, Jihong Meng, Qingyi Li, Ximing Zhang, Zhengjie Li, Changchun Wang, Wenting College of Computer Science and Technology Jilin University China Key Laboratory of Symbolic Computation and Knowledge Engineering Ministry of Education Jilin University China Department of Computer Science University of Texas Dallas United States

Open-Set Semi-Supervised Learning (OS-SSL) refers to the task of learning classifiers with labeled and unlabeled instances, but the unlabeled data may contain the instances associated with unseen labels, dubbed as Out-Of-Distribution (OOD) instances. The main idea of OS-SSL is to detect and filter out the OOD instances to avoid the negative effects for classifier training. The current methods treat the detection of OOD instances and the classification of seen labels as two specific tasks, and achieve OS-SSL by using a shared backbone with two task-specific heads. However, we found that such way may result in a conflict problem between the two specific heads by observing the gradients of the shared backbone parameters. To solve this problem, we develop a novel OS-SSL method, namely Closed LOop NEtworks (CLONE), which employs two respective networks to detect OOD instances and classify seen labels, but makes a feedback loop between them. We can train CLONE in an end-to-end manner efficiently. We empirically compare CLONE with the existing OS-SSL methods on several benchmark datasets. The experimental results indicate that CLONE can outperform existing baseline methods in most settings. © 2023, The Authors. All rights reserved.

关键词： Cloning

来源：评论

学校读者我要写书评

暂无评论

Todo: Task Offloading Decision Optimizer for the Efficient Provision of Offloading Schemes

SSRN

引用

SSRN 2023年

作者： Chen, Shilin Wang, Xingwang Sun, Yafeng College of Computer Science and Technology Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University Changchun130012 China College of Computer Science and Technology Jilin University Changchun130012 China

As the volume of data stored on local devices increases, users turn to edge devices to help with processing tasks. Developing offloading schemes is challenging due to the varying configurations of edge devices and user preferences. While traditional methods provide schemes for offloading in various scenarios, they face unavoidable challenges, including the requirement to manage device workloads in real-time, significant computational costs, and the difficulty of balancing multi-objectives in offloading schemes. To solve these problems, we propose the Task Offloading Decision Optimizer, which offers efficient multi-objective offloading schemes considering real-time device workload and user preference. The proposed offloading scheme contains three goals: reducing task execution time, decreasing device energy consumption, and lowering rental costs. It comprises two important parts: Scheme Maker and Scheme Assistor. Scheme Maker utilizes deep reinforcement learning, optimizes the internal architecture, and enhances the performance of the operation. It optimizes buffer storage to generate dependable multi-objective offloading schemes considering real-time environmental conditions. Scheme Assistor utilizes the data in the Scheme Maker buffer to enhance efficiency by reducing computational costs. Extensive experiments have proved that the proposed framework efficiently provides offloading schemes considering the real-time conditions of the devices and the users, and it offers offloading schemes that enhance task completion rate by 50%. Compared to the baseline, the task execution time is reduced by 12.1%, and the device energy consumption is reduced by 11.4%. © 2023, The Authors. All rights reserved.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

IoT Device Identification via A Bio-Inspired Feature Selection Approach

IoT Device Identification via A Bio-Inspired Feature Selecti...

引用

IEEE International Conference on Communications (ICC)

作者： Boxiong Wang Hui Kang Geng Sun Jiahui Li College of Software Jilin University Changchun China College of Computer Science and Technology Jilin University Changchun China Key Laboratory of Symbolic Computation and Knowledge Engineering Ministry of Education Jilin University Changchun China

The rapid development of the Internet-of-Things (IoT) also brings security and other problems. Device identification is a crucial tool for IoT security issues, which can detect and prevent cyber-attacks. Feature selection is an effective data preprocessing technique in IoT device identification, which can improve the performance of classification and reduce computational complexity. In this paper, we propose a novel wrapper feature selection approach based on the improved binary honey badger algorithm (IBHBA) to select features in IoT traffic. Four improved factors are employed in IBHBA to expand the search scope, balance the exploration and exploitation phases, and enhance the search capability. Moreover, a binary mechanism is adopted to make the algorithm more suitable for feature selection in IoT device identification. The experimental results on several real IoT traffic datasets denote that IBHBA outperforms some classical and latest comparison algorithms in the feature selection of IoT device identification.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Talking Head Generation for Media Interaction System with Feature Disentanglement

Talking Head Generation for Media Interaction System with Fe...

引用

International Conference on Parallel and Distributed Systems (ICPADS)

作者： Lei Zhang Qingyang Chen Zhilei Liu College of Intelligence and Computing School of New Media and Communication Tianjin University Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University Changchun P.R. China State Key Laboratory for Novel Software Technology Nanjing University P.R. China School of New Media and Communication Tianjin University Tianjin China College of Intelligence and Computing Tianjin University Tianjin China

The task of talking head generation for the media interaction system is to take images and audio clips of the target face as input, and generate a realistic video of the target synchronized with the audio. Most of the existing works directly take all the information in the image as input, which causes the problem of feature information redundancy. At the same time, in the stage of feature fusion, the image and audio information is directly spliced, and the relationship between the two modal information is ignored. To solve the problems, we initially implement the disentanglement of image by introducing a loss function to separate the image into identity features and content-related features. Besides, we introduce a multi-head selfattention mechanism to learn the relationship between the two modal information of image and audio, and implement the full fusion of multi-modal information. In addition, we validate the effectiveness of our model through extensive quantitative and qualitative analysis of two datasets. Extensive experiments show the superiority of the proposed model in all aspects.

关键词： Analytical models Redundancy Media Synchronization Task analysis Faces

来源：评论

学校读者我要写书评

暂无评论

Efficient Heuristics for Learning Scalable Bayesian Network Classifier from Labeled and Unlabeled Data

SSRN

引用

SSRN 2023年

作者： Wang, Limin Wang, Junjie Guo, Lu Li, Qilong Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University Changchun130012 China College of Software Jilin University Changchun130012 China College of Instrumentation and Electrical Engineering Jilin University Changchun130012 China

Naive Bayes (NB) is one of the top ten machine learning algorithms whereas its attribute independence assumption rarely holds in practice. A feasible and efficient approach to improving NB is relaxing the assumption by adding augmented edges to the restricted topology of NB. In this paper we prove theoretically that the generalized topology may be a suboptimal solution to model multivariate probability distributions if its fitness to data cannot be measured. Thus we propose to apply log-likelihood function as the scoring function, then introduce an efficient heuristic search strategy to explore high-dependence relationships, and for each iteration the learned topology will be improved to fit data better. The proposed algorithm, called log-likelihood Bayesian classifier (LLBC), can respectively learn two submodels from labeled training set and individual unlabeled testing instance, and then make them work jointly for classification in the framework of ensemble learning. Our extensive experimental evaluations on 36 UCI benchmark datasets reveal that, LLBC demonstrates excellent classification performance and provides a competitive approach to learn from labeled and unlabeled data. © 2023, The Authors. All rights reserved.

关键词： Classification (of information)

来源：评论

学校读者我要写书评

暂无评论

Semi-supervised multi-label learning with balanced binary angular margin loss 24

Semi-supervised multi-label learning with balanced binary an...

引用

Proceedings of the 38th International Conference on Neural Information Processing Systems

作者： Ximing Li Silong Liang Changchun Li Pengfei Wang Fangming Gu College of Computer Science and Technology Jilin University China and Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University China Computer Network Information Center Chinese Academy of Sciences China and University of Chinese Academy of Sciences Chinese Academy of Sciences China

ISBN: (纸本)9798331314385

Semi-supervised multi-label learning (SSMLL) refers to inducing classifiers using a small number of samples with multiple labels and many unlabeled samples. The prevalent solution of SSMLL involves forming pseudo-labels for unlabeled samples and inducing classifiers using both labeled and pseudo-labeled samples in a self-training manner. Unfortunately, with the commonly used binary type of loss and negative sampling, we have empirically found that learning with labeled and pseudo-labeled samples can result in the variance bias problem between the feature distributions of positive and negative samples for each label. To alleviate this problem, we aim to balance the variance bias between positive and negative samples from the perspective of the feature angle distribution for each label. Specifically, we extend the traditional binary angular margin loss to a balanced extension with feature angle distribution transformations under the Gaussian assumption, where the distributions are iteratively updated during classifier training. We also suggest an efficient prototype-based negative sampling method to maintain high-quality negative samples for each label. With this insight, we propose a novel SSMLL method, namely Semi-Supervised Multi-Label Learning with Balanced Binary Angular Margin loss (S2ML2-BBAM). To evaluate the effectiveness of S2ML2-BBAM, we compare it with existing competitors on benchmark datasets. The experimental results validate that S2ML2-BBAM can achieve very competitive performance.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Why Misinformation is Created? Detecting them by Integrating Intent Features

arXiv

引用

arXiv 2024年

作者： Wang, Bing Fu, Bo Li, Ximing Pei, Songwen Li, Changchun Wang, Shengsheng College of Computer Science and Technology Jilin University Changchun Jilin China Liaoning Normal University Liaoning Dalian China University of Shanghai for Science and Technology Shanghai China Key Laboratory of Symbolic Computation and Knowledge Engineering The Ministry of Education Jilin University China

Various social media platforms, e.g., Twitter and Reddit, allow people to disseminate a plethora of information more efficiently and conveniently. However, they are inevitably full of misinformation, causing damage to diverse aspects of our daily lives. To reduce the negative impact, timely identification of misinformation, namely Misinformation Detection (MD), has become an active research topic receiving widespread attention. As a complex phenomenon, the veracity of an article is influenced by various aspects. In this paper, we are inspired by the opposition of intents between misinformation and real information. Accordingly, we propose to reason the intent of articles and form the corresponding intent features to promote the veracity discrimination of article features. To achieve this, we build a hierarchy of a set of intents for both misinformation and real information by referring to the existing psychological theories, and we apply it to reason the intent of articles by progressively generating binary answers with an encoder-decoder structure. We form the corresponding intent features and integrate it with the token features to achieve more discriminative article features for MD. Upon these ideas, we suggest a novel MD method, namely Detecting Misinformation by Integrating Intent featuRes (Dm-inter). To evaluate the performance of Dm-inter, we conduct extensive experiments on benchmark MD datasets. The experimental results validate that Dm-inter can outperform the existing baseline MD methods. Copyright © 2024, The Authors. All rights reserved.

关键词： Data mining

来源：评论

学校读者我要写书评

暂无评论

Discrepancy-Guided Reconstruction Learning for Image Forgery Detection

arXiv

引用

arXiv 2023年

作者： Shi, Zenan Chen, Haipeng Chen, Long Zhang, Dong College of Computer Science and Technology Jilin University China Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University China Department of CSE The Hong Kong University of Science and Technology Hong Kong

In this paper, we propose a novel image forgery detection paradigm for boosting the model learning capacity on both forgery-sensitive and genuine compact visual patterns. Compared to the existing methods that only focus on the discrepant-specific patterns (e.g., noises, textures, and frequencies), our method has a greater generalization. Specifically, we first propose a Discrepancy-Guided Encoder (DisGE) to extract forgery-sensitive visual patterns. DisGE consists of two branches, where the mainstream backbone branch is used to extract general semantic features, and the accessorial discrepant external attention branch is used to extract explicit forgery cues. Besides, a Double-Head Reconstruction (DouHR) module is proposed to enhance genuine compact visual patterns in different granular spaces. Under DouHR, we further introduce a Discrepancy-Aggregation Detector (DisAD) to aggregate these genuine compact visual patterns, such that the forgery detection capability on unknown patterns can be improved. Extensive experimental results on four challenging datasets validate the effectiveness of our proposed method against state-of-the-art competitors. Copyright © 2023, The Authors. All rights reserved.

关键词： Textures

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：