检索结果-内蒙古大学图书馆

Learning Feature Exploration and Selection With Handcrafted Features for Few-Shot Learning

IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS 2025年第4期55卷 2599-2610页

作者： Zhang, Yi Huang, Sheng huangfu, luwen Zeng, Daniel Dajun Chongqing Univ Sch Big Data & Software Engn Chongqing 400044 Peoples R China Chongqing Univ Minist Educ Key Lab Dependable Serv Comp Cyber Phys Soc Chongqing 400044 Peoples R China Chongqing Univ Sch Big Data & Software Engn Chongqing 400044 Peoples R China San Diego State Univ Fowler Coll Business San Diego CA 92182 USA San Diego State Univ Ctr Human Dynam Mobile Age San Diego CA 92182 USA Chinese Acad Sci Inst Automat Beijing 100190 Peoples R China

Interest in few-shot learning (FSL) has grown recently, but the value of feature learning, which bridges the gap between base and novel classes, remains largely understudied. The limited availability of labeled samples for each class poses a major challenge. To tackle this, we propose a simple yet effective approach called deep discriminative handcrafted feature regression (DDHFR) to explore intrinsic information and select improved discriminative features in few-shot data by mining knowledge from classical handcrafted features. To explore intrinsic information, we design several deep handcrafted feature regression (DHFR) modules and plugged them separately into different layers of the backbone to use feature engineering knowledge for feature learning optimization at different granularities. To achieve discriminative feature selection, we incorporate an auxiliary classifier (AC) into each DHFR module to enhance the acquisition of discriminative information. Furthermore, we employed self-distillation to boost ability of ACs ot be classified. Experimental results in three backbones on three datasets show that DDHFR can generally improve the performance of existing FSL methods. On average, it improves the recognition accuracy by 1.16% in two common few-shot settings.

关键词： Few-shot learning (FSL) handcrafted features histogram of oriented gradient (HOG) knowledge distillation local binary pattern (LBP)

来源：评论

学校读者我要写书评

暂无评论

Rethinking the sample relations for few-shot classification

引用

IMAGE AND VISION COMPUTING 2025年 159卷

作者： Yin, Guowei Huang, Sheng huangfu, luwen Zhang, Yi Zhang, Xiaohong Chongqing Univ Sch Big Data & Software Engn 55 Daxuecheng South Rd Chongqing 401331 Peoples R China Chongqing Univ Key Lab Dependable Serv Comp Cyber Phys Soc Minist Educ 174 Shazhengjie Chongqing 400044 Peoples R China San Diego State Univ Fowler Coll Business San Diego CA 92182 USA San Diego State Univ Ctr Human Dynam Mobile Age HDMA San Diego CA 92182 USA San Diego State Univ AI4Business Lab San Diego CA 92182 USA

Feature quality is paramount for classification performance, particularly in few-shot scenarios. Contrastive learning, a widely adopted technique for enhancing feature quality, leverages sample relations to extract intrinsic features that capture semantic information and has achieved remarkable success in Few-Shot Learning (FSL). Nevertheless, current few-shot contrastive learning approaches often overlook the semantic similarity discrepancies at different granularities when employing the same modeling approach for different sample relations, which limits the potential of few-shot contrastive learning. In this paper, we introduce a straightforward yet effective contrastive learning approach, Multi-Grained Relation Contrastive Learning (MGRCL), as a pre-training feature learning model to boost few-shot learning by meticulously modeling sample relations at different granularities. MGRCL categorizes sample relations into three types: intra-sample relation of the same sample under different transformations, intra-class relation of homogeneous samples, and inter-class relation of inhomogeneous samples. In MGRCL, we design Transformation Consistency Learning (TCL) to ensure the rigorous semantic consistency of a sample under different transformations by aligning predictions of input pairs. Furthermore, to preserve discriminative information, we employ Class Contrastive Learning (CCL) to ensure that a sample is always closer to its homogeneous samples than its inhomogeneous ones, as homogeneous samples share similar semantic content while inhomogeneous samples have different semantic content. Our method is assessed across four popular FSL benchmarks, showing that such a simple pre-training feature learning method surpasses a majority of leading FSL methods. Moreover, our method can be incorporated into other FSL methods as the pre-trained model and help them obtain significant performance gains.

关键词： Few-shot classification Sample relation Transformation consistency learning Class contrastive learning

来源：评论

学校读者我要写书评

暂无评论

Learning complementary visual information for few-shot food recognition by Regional Erasure and Reactivation

引用

EXPERT SYSTEMS WITH APPLICATIONS 2025年 268卷

作者： Zhang, Yi Li, Heng huangfu, luwen Balazs, Lili Huang, Sheng Chongqing Normal Univ Coll Comp & Informat Sci Chongqing 401331 Peoples R China ChongQing Univ Sch Big Data & Software Engn Chongqing 400044 Peoples R China San Diego State Univ San Diego CA 92182 USA Univ Pecs H-7624 Pecs Hungary

Food plays a vital role inhuman life, covering a wide range of health, cultural and medical aspects. With the increasing demand for applications such as food recommendation and dietary assessment, few-shot food recognition (FSFR) has become an emerging topic in computer vision. FSFR aims to train a model to recognize novel food classes with only a handful amount of training data. Existing methods for FSFR either ignore the ingredient information in food images or rely on ingredient labels that are not always available or visible. In this paper, we propose a simple and efficient method named Regional Erase-Reactivation (RER), which uses an adversarial erasing strategy to automatically discover the ingredients in the image and learn a more comprehensive representation. RER consists of two classifiers that mine different regions of the target separately. A classifier is first used to recognize and erase the local discriminative regions on the intermediate feature maps. Then, the erased features are fed into another classifier to discover complementary information about the target. Moreover, we use soft labels instead of ingredient labels to supervise the second classifier, which reduces the model bias caused by ingredients that are not visible in the sample. We conduct experiments on two benchmark datasets, including Food-101 and VIREO Food-172, and show that RER performs better than the state-of-the-art FSFR methods.

关键词： Few-shot learning Food image recognition Adversarial erasing

来源：评论

学校读者我要写书评

暂无评论

Rethinking the Sample Relations for Few-Shot Classification

arXiv

引用

arXiv 2025年

作者： Yin, Guowei Huang, Sheng huangfu, luwen Zhang, Yi Zhang, Xiaohong School of Big Data and Software Engineering Chongqing University No.55 Daxuecheng South Rd. Shapingba Chongqing401331 China Ministry of Education Key Laboratory of Dependable Service Computing in Cyber Physical Society Chongqing University No.174 Shazhengjie Shapingba Chongqing400044 China Fowler College of Business San Diego State University San DiegoCA92182 United States San Diego State University San DiegoCA92182 United States AI4Business Lab San Diego State University San DiegoCA92182 United States

Feature quality is paramount for classification performance, particularly in few-shot scenarios. Contrastive learning, a widely adopted technique for enhancing feature quality, leverages sample relations to extract intrinsic features that capture semantic information and has achieved remarkable success in Few-Shot Learning (FSL). Nevertheless, current few-shot contrastive learning approaches often overlook the semantic similarity discrepancies at different granularities when employing the same modeling approach for different sample relations, which limits the potential of few-shot contrastive learning. In this paper, we introduce a straightforward yet effective contrastive learning approach, Multi-Grained Relation Contrastive Learning (MGRCL), as a pretraining feature learning model to boost few-shot learning by meticulously modeling sample relations at different granularities. MGRCL categorizes sample relations into three types: intra-sample relation of the same sample under different transformations, intra-class relation of homogenous samples, and inter-class relation of inhomogeneous samples. In MGRCL, we design Transformation Consistency Learning (TCL) to ensure the rigorous semantic consistency of a sample under different transformations by aligning predictions of input pairs. Furthermore, to preserve discriminative information, we employ Class Contrastive Learning (CCL) to ensure that a sample is always closer to its homogenous samples than its inhomogeneous ones, as homogenous samples share similar semantic content while inhomogeneous samples have different semantic content. Our method is assessed across four popular FSL benchmarks, showing that such a simple pre-training feature learning method surpasses a majority of leading FSL methods. Moreover, our method can be incorporated into other FSL methods as the pre-trained model and help them obtain significant performance gains. © 2025, CC BY-NC-SA.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Class-Prototype Discriminative Network for Generalized Zero-Shot Learning

引用

IEEE SIGNAL PROCESSING LETTERS 2020年第0期27卷 301-305页

作者： Huang, Sheng Lin, Jingkai huangfu, luwen Chongqing Univ Minist Educ Key Lab Dependable Serv Comp Cyber Phys Soc Chongqing 400044 Peoples R China Chongqing Univ Sch Big Data & Software Engn Chongqing 400044 Peoples R China San Diego State Univ Fowler Coll Business San Diego CA 92182 USA

We present a novel end-to-end deep metric learning model named Class-Prototype Discriminative Network (CPDN) for Generalized Zero-Shot Learning (GZSL). It consists of a generative network for producing the visual prototype of each class by feeding its semantic representation, and a metric network for measuring the similarities between the sample and the generated class-prototypes to accomplish the classification. In CPDN, a query sample intends to posses a higher similarity with its homogenous class-prototypes while the lower similarities with the inhomogenous ones, and the class-prototypes also intend to be distinguished with each other through the metric network. Moreover, a discriminative version of Relation Network (RN) named Discriminative Relation Network (DRN) is presented by incorporating the aforementioned idea into the conventional RN model for further achieving the complementation CDPN and RN in metric learning. Extensive experimental results on standard benchmarks demonstrate that our proposed approaches consistently outperform RN, and achieve the competitive performances compared with the state-of-the-arts in GZSL.

关键词： Zero-shot learning deep learning cross-modal metric learning relation network

来源：评论

学校读者我要写书评

暂无评论

Bootstrapping Polar-Opposite Emotion Dimensions from Online Reviews 11

Bootstrapping Polar-Opposite Emotion Dimensions from Online ...

引用

11th International Conference on Language Resources and Evaluation (LREC)

作者： huangfu, luwen Surdeanu, Mihai Univ Arizona Tucson AZ 85719 USA

ISBN: (纸本)9791095546009

We propose a novel bootstrapping approach for the acquisition of lexicons from unannotated, informal online texts (in our case, Yelp reviews) for polar-opposite emotion dimension values from the Ortony/Clore/Collins model of emotions (e.g., desirable/undesirable). Our approach mitigates the intrinsic problem of limited supervision in bootstrapping with an effective strategy that softly labels unlabeled terms, which are then used to better estimate the quality of extraction patterns. Further, we propose multiple solutions to control for semantic drift by taking advantage of the polarity of the categories to be learned (e.g., praiseworthy vs. blameworthy). Experimental results demonstrate that our algorithm achieves considerably better performance than several baselines.

关键词： bootstrapping semantic drift limited supervision

来源：评论

学校读者我要写书评

暂无评论

MSTIL: Multi-cue Shape-aware Transferable Imbalance Learning for effective graphic API recommendation

引用

JOURNAL OF SYSTEMS AND SOFTWARE 2023年 200卷

作者： Qin, Rong Wang, Zeyu Huang, Sheng huangfu, luwen Chongqing Univ Sch Big Data & Software Engn Chongqing 400044 Peoples R China San Diego State Univ Fowler Coll Business San Diego CA 92182 USA San Diego State Univ Ctr Human Dynam Mobile Age San Diego CA 92182 USA

Application Programming Interface (API) recommendation based on graphs is a valuable task in the fields of data visualization and software engineering. However, this task was previously undefined until a recently published paper coining the task as Plot2API and utilizing a deep learning-based method named SPGNN. Compared to general image classification methods, this dedicated approach uses semantic parsing to exploit deep features and yields better performance. However, its performance declines sharply in unbalanced datasets, thus limiting its generalizability. To address this issue, we propose a method named Multi-cue Shape and software engineering-aware Transferable Imbalance Learning (MSTIL), consisting of three major components: Cross-Language Shape-Aware Plot Transfer Learning (CLSAPTL), Cross-Language API Semantic Similarity-based Data Augmentation (CLASSDA), and Imbalance Plot2API Learning (IPL). Motivated by the hierarchical classification of the graphs, CLSAPTL guides the model to learn the graphs' class hierarchy and thereby enabling the model to learn more transferable visual features. Given that a graph can be associated with multiple APIs and motivated by the fact that many APIs that exert similar functions in different languages have semantically similar names, CLASSDA leverages the samples of APIs with semantically similar names to assist in feature learning. Finally, inspired by the essence of softmax cross entropy loss, IPL alleviates the imbalances between positive and negative samples during training. We conduct our experiments on two public datasets. Extensive experimental results shows that MSTIL improves the performance of classic CNNs along with the state-of-the-art method, demonstrating its effectiveness. Specifically, MSTIL has an average relative mAP improvement of 12.94% across the models on all datasets.(c) 2023 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.o

关键词： API recommendation Data augmentation Transfer learning Data visualization Semantic similarity

来源：评论

学校读者我要写书评

暂无评论

CURVE4COVID: Comprehensive Understanding via Representative Variable Exploration for COVID-19 Completed Research 27

CURVE4COVID: Comprehensive Understanding via Representative ...

引用

27th Annual Americas Conference on Information Systems (AMCIS)

作者： Serrano, Patrick huangfu, luwen San Diego State Univ San Diego CA 92182 USA

ISBN: (纸本)9781733632584

Coronavirus disease 2019 (COVID-19) is the most significant pandemic in the 21st century and affects the lives of millions of people today. By the end of 2020, more than 83 million cases were confirmed, and more than 1.6 million deaths were reported globally. Understanding how COVID-19 spreads across diverse communities is key to public health surveillance and management for such a devastating and highly infectious disease. Current studies mainly analyze either non-human factors or human factors, focusing on a limited number of variables that can influence COVID-19 transmission. However, in a real-world context, these factors interact with each other and collectively shape the infection rate. Therefore, a comprehensive study based on both non-human and human factors on a large scale is required to fully understand disease transmission. Here, we propose a research framework named Comprehensive Understanding via Representative Variable Exploration for COVID-19 (CURVE4COVID). With the accessibility of various data online, including COVID-19-related Google Trends (e.g., human search behavior) and government-managed data (e.g., weather, air pollution, economic indicators), we conduct a large-scale and multivariable analysis of the critical factors for COVID-19 transmission, which can shed light on the complexity of infectious disease management. The results demonstrate that combining non-human and human factors provides better predictive power for infection rates than non-human factors or human factors alone. This study's findings can provide new insights into disease transmission and help policymakers enhance preventative measures and healthcare management, thus having a far-reaching impact on society.

关键词： COVID-19 coronavirus SARS-CoV-2 disease transmission weather Google Trends air pollutants economic indicators Infodemiology

来源：评论

学校读者我要写书评

暂无评论

Shape primitive histogram: low-level face representation for face recognition

引用

IET BIOMETRICS 2014年第4期3卷 325-334页

作者： Huang, Sheng Yang, Dan Zhang, Haopeng huangfu, luwen Zhang, Xiaohong Chongqing Univ Coll Comp Sci Chongqing 400044 Peoples R China Chongqing Univ Sch Software Engn Chongqing 400044 Peoples R China Beihang Univ Sch Astronaut Beijing 100191 Peoples R China Chinese Acad Sci Inst Automat State Key Lab Management & Control Complex Syst Beijing 100190 Peoples R China Minist Educ Key Lab Dependable Serv Comp Cyber Phys Soc Chongqing 400044 Peoples R China

Human face contains abundant shape features. This fact motivates a lot of shape feature-based face detection and three-dimensional (3D) face recognition approaches. However, as far as we know, there is no prior low-level face representation which is purely based on shape feature proposed for conventional 2D (image-based) face recognition. In this study, the authors present a novel low-level shape-based face representation named 'shape primitives histogram' (SPH) for face recognition. In this approach, the face images are separated into a number of tiny shape fragments and they reduce these shape fragments to several uniform atomic shape patterns called 'shape primitives'. Then the face representation is obtained by implementing a histogram statistic of shape primitives in a local image region. To take scale information into consideration, they also produce multi-scale SPHs (MSPHs) by concatenating the SPHs extracted from different scales. Moreover, they experimentally study the influences of each stage of SPH computation on performance, concluding that a small cell with 1/2 overlap and a fine size block with 1/2 overlap are important for good results. Four popular face databases, namely ORL, AR, YaleB and LFW-a, are employed to evaluate SPH and MSPH. Surprisingly, such seemingly naive shape-based face representations outperform the state-of-the-art low-level face representations.

关键词： face recognition feature extraction image representation shape recognition visual databases AR LFW ORL SPH YaleB atomic shape patterns face databases face detection face images face representations human face image region low-level face representation shape feature shape features shape primitive histogram tiny shape fragments

来源：评论

学校读者我要写书评

暂无评论

DeepApp: characterizing dynamic user interests for mobile application recommendation

引用

WORLD WIDE WEB-INTERNET AND WEB INFORMATION SYSTEMS 2023年第5期26卷 2623-2645页

作者： Liang, Yunji Liu, Lei huangfu, luwen Wang, Zhu Guo, Bin Northwestern Polytech Univ Xian 710072 Shaanxi Peoples R China San Diego State Univ Fowler Coll Business San Diego CA 92182 USA San Diego State Univ Ctr Human Dynam Mobile Age HDMA San Diego CA 92182 USA

It is extremely difficult to find one app in app stores that exactly meets the needs of users with the boom in mobile applications nowadays. Although numerous app recommendation services are available, they mainly employ static data (e.g., information of installed apps) and rarely consider the dynamics of user interests. In this paper, we assume that user interest consists of two components: short-term temporal interests and long-term preferences, and we propose one general framework, namely DeepApp, to enhance app recommendation performance. In DeepApp, on the one hand, we use one linear model to characterize the stable user-app associations;on the other hand, we employ the Long Short-Term Memory (LSTM) model to capture the evolution of interests based on the usage patterns of mobile apps. Finally, the Wide &Deep model is applied to fuse the effects of these two types of interests, long-term preferences and short-term temporal interests, by learning the latent interaction between linear and nonlinear features. DeepApp was evaluated on a large-scale dataset, with 4,775,293 users and 238,206 mobile apps. DeepApp achieves a significant performance gain compared with baselines (more than 6% in terms of NDCG@6 over probability matrix factorization (PMF), neural collaborative filtering (NCF), and neural tensor factorization (NTF)). This demonstrates that the integration of dynamic user interests is beneficial for mobile app recommendations.

关键词： Mobile application recommendation Dynamic interests Wide& Deep

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：