检索结果-内蒙古大学图书馆

TENET: Beyond Pseudo-labeling for Semi-supervised Few-shot Learning

MACHINE INTELLIGENCE RESEARCH 2025年 1-13页

作者： Ma, Chengcheng Dong, Weiming Xu, Changsheng Univ Chinese Acad Sci UCAS Sch Artificial Intelligence Beijing 100049 Peoples R China Chinese Acad Sci CASIA Inst Automat Natl Lab Pattern Recognit NLPR Beijing 100190 Peoples R China

Few-shot learning attempts to identify novel categories by exploiting limited labeled training data, while the performances of existing methods still have much room for improvement. Thanks to a very low cost, many recent methods resort to additional unlabeled training data to boost performance, known as semi-supervised few-shot learning (SSFSL). The general idea of SSFSL methods is to first generate pseudo labels for all unlabeled data and then augment the labeled training set with selected pseudo-labeled data. However, almost all previous SSFSL methods only take supervision signal from pseudo-labeling, ignoring that the distribution of training data can also be utilized as an effective unsupervised regularization. In this paper, we propose a simple yet effective SSFSL method named feature reconstruction based regression method (TENET), which takes low-rank feature reconstruction as the unsupervised objective function and pseudo labels as the supervised constraint. We provide several theoretical insights on why TENET can mitigate overfitting on low-quality training data, and why it can enhance the robustness against inaccurate pseudo labels. Extensive experiments on four popular datasets validate the effectiveness of TENET.

关键词： Semi-supervised few-shot learning few-shot learning pseudo-labeling linear regression low-rank reconstruction

来源：评论

学校读者我要写书评

暂无评论

Do singular values contain adequate information for face recognition?

引用

pattern recognitION 2003年第3期36卷 649-655页

作者： Tian, Y Tan, TN Wang, YH Fang, YC Chinese Acad Sci Inst Automat Natl Lab Pattern Recognit NLPR Beijing Peoples R China

Singular values (SVs) have been used for face recognition by many researchers. In this paper, we show that the SVs contain little useful information for face recognition and most important information is encoded in the two orthogonal matrices of the SVD. Experimental results are given to support this observation. To overcome this problem, a new method for face recognition based on the above finding is proposed. The face image is projected on to the orthogonal basis of SVD and then the vectors of coefficients are used as the face image features. By using probability density of this image feature obtained by a simplified EM algorithm, the Bayesian classifier is adopted to recognize the unknown faces. The proposed algorithm obtains acceptable experimental results on the ORL face database. (C) 2002 pattern recognition Society. Published by Elsevier Science Ltd. All rights reserved.

关键词： face recognition orthogonal decomposition SVD Bayesian decision

来源：评论

学校读者我要写书评

暂无评论

Error-correcting output codes based ensemble feature extraction

引用

pattern recognitION 2013年第4期46卷 1091-1100页

作者： Zhong, Guoqiang Liu, Cheng-Lin Chinese Acad Sci Inst Automat Natl Lab Pattern Recognit NLPR Beijing 100190 Peoples R China

This paper proposes a novel feature extraction method based on ensemble learning. Using the error-correcting output codes (ECOC) to design binary classifiers (dichotomizers) for separating subsets of classes, the outputs of the dichotomizers are linear or nonlinear features that provide powerful separability in a new space. In this space, the vector quantization based meta classifier can be viewed as an ECOC decoder, where each learned prototype of a class can be seen as a codeword of the class in the new representation space. We conducted extensive experiments on 16 multi-class data sets from the UCI machine learning repository. The results demonstrate the superiority of the proposed method over both existing ECOC approaches and classic feature extraction approaches. In particular, the decoding strategy using a meta classifier is shown to be more computationally efficient than the linear loss-weighted decoding in state-of-the-art ECOC methods. (C) 2012 Elsevier Ltd. All rights reserved.

关键词： Feature extraction Ensemble learning Error-correcting output codes (ECOC) Meta learner

来源：评论

学校读者我要写书评

暂无评论

Online and offline handwritten Chinese character recognition: Benchmarking on new databases

引用

pattern recognitION 2013年第1期46卷 155-162页

作者： Liu, Cheng-Lin Yin, Fei Wang, Da-Han Wang, Qiu-Feng Chinese Acad Sci Inst Automat Natl Lab Pattern Recognit NLPR Beijing 100190 Peoples R China

Recently, the Institute of Automation of Chinese Academy of Sciences (CASIA) released the unconstrained online and offline Chinese handwriting databases CASIA-OLHWDB and CASIA-HWDB, which contain isolated character samples and handwritten texts produced by 1020 writers. This paper presents our benchmarking results using state-of-the-art methods on the isolated character datasets OLHWDB1.0 and HWDB1.0 (called DB1.0 in general), OLHWDB1.1 and HWDB1.1 (called DB1.1 in general). The DB1.1 covers 3755 Chinese character classes as in the level-1 set of GB2312-80. The evaluated methods include 1D and pseudo 2D normalization methods, gradient direction feature extraction from binary images and from gray-scale images, online stroke direction feature extraction from pen-down trajectory and from pen lifts, classification using the modified quadratic discriminant function (MQDF), discriminative feature extraction (DFE), and discriminative learning quadratic discriminant function (DLQDF). Our experiments reported the highest test accuracies 89.55% and 93.22% on the HWDB1.1 (offline) and OLHWDB1.1 (online), respectively, when using the MQDF classifier trained with DB1.1. When training with both the DB1.0 and DB1.1, the test accuracies on HWDB1.1 and OLHWDB are improved to 90.71% and 93.95%, respectively. Using DFE and DLQDF, the best results on HWDB1.1 and OLHWDB1.1 are 92.08% and 94.85%, respectively. Our results are comparable to the best results of the ICDAR2011 Chinese Handwriting recognition Competition though we used less training samples. (C) 2012 Elsevier Ltd. All rights reserved.

关键词： Handwritten Chinese character recognition Online Offline Databases Benchmarking

来源：评论

学校读者我要写书评

暂无评论

Feature Coding in Image Classification: A Comprehensive Study

引用

IEEE TRANSACTIONS ON pattern ANALYSIS AND MACHINE INTELLIGENCE 2014年第3期36卷 493-506页

作者： Huang, Yongzhen Wu, Zifeng Wang, Liang Tan, Tieniu Chinese Acad Sci CASIA Inst Automat Natl Lab Pattern Recognit NLPR CRIPAC 95 ZhongGuanCun East St Beijing 100190 Peoples R China

Image classification is a hot topic in computer vision and pattern recognition. Feature coding, as a key component of image classification, has been widely studied over the past several years, and a number of coding algorithms have been proposed. However, there is no comprehensive study concerning the connections between different coding methods, especially how they have evolved. In this paper, we first make a survey on various feature coding methods, including their motivations and mathematical representations, and then exploit their relations, based on which a taxonomy is proposed to reveal their evolution. Further, we summarize the main characteristics of current algorithms, each of which is shared by several coding strategies. Finally, we choose several representatives from different kinds of coding approaches and empirically evaluate them with respect to the size of the codebook and the number of training samples on several widely used databases (15-Scenes, Caltech-256, PASCAL VOC07, and SUN397). Experimental findings firmly justify our theoretical analysis, which is expected to benefit both practical applications and future research.

关键词： Image classification feature coding bag-of-features

来源：评论

学校读者我要写书评

暂无评论

Real-time ink simulation using a grid-particle method

引用

COMPUTERS & GRAPHICS-UK 2012年第8期36卷 1025-1035页

作者： Xu, Shibiao Mei, Xing Dong, Weiming Zhang, Zhiyi Zhang, Xiaopeng Chinese Acad Sci Inst Automat NLPR LIAMA Natl Lab Pattern Recognit Beijing 100190 Peoples R China

This paper presents an effective method to simulate the ink diffusion process in real time that yields realistic visual effects. Our algorithm updates the dynamic ink volume using a hybrid grid-particle method: the fluid velocity field is calculated with a low-resolution grid structure, whereas the highly detailed ink effects are controlled and visualized with the particles. To facilitate user interaction and extend this method, we propose a particle-guided method that allows artists to design the overall states using the coarse-resolution particles and to preview the motion quickly. To treat coupling with solids and other fluids, we update the grid-particle representation with no-penetration boundary conditions and implicit interaction conditions. To treat moving "ink-emitting" objects, we introduce an extra drag-force model to enhance the particle motion effects;this force might not be physically accurate, but it proves effective for producing animations. We also propose an improved ink rendering method that uses particle sprites and motion blurring techniques. The simulation and the rendering processes are efficiently implemented on graphics hardware at interactive frame rates. Compared to traditional fluid simulation methods that treat water and ink as two mixable fluids, our method is simple but effective: it captures various ink effects, such as pinned boundaries (Nelson, 2005 [1]) and filament patterns (Shiny et al., 2010 [2]), while still running in real time, it allows easy control of the animation, it includes basic solid-fluid interactions, and it can address multiple ink sources without complex interface tracking. Our method is attractive for animation production and art design. (c) 2012 Elsevier Ltd. All rights reserved.

关键词： Ink diffusion Particle-guided Coupling Drag force Motion blur Multiple ink Animation

来源：评论

学校读者我要写书评

暂无评论

Latent Topic Model Based on Gaussian-LDA for Audio Retrieval

Latent Topic Model Based on Gaussian-LDA for Audio Retrieval

引用

Chinese Conference on pattern recognition

作者： Hu, Pengfei Liu, Wenju Jiang, Wei Yang, Zhanlei Chinese Acad Sci Inst Automat Natl Lab Pattern Recognit NLPR Beijing 100190 Peoples R China

ISBN: (纸本)9783642335051

In this paper,we introduce a new topic model named Gaussian-LDA, which is more suitable to model continuous data. Topic Model based on latent Dirichlet allocation (LDA) is widely used for the statistical analysis of document collections and other discrete data. The LDA model assumes that the words of each document arise from a mixture of topics, each of which is a multinomial distribution over the vocabulary. To apply the original LDA to process continuous data, discretization based vector quantization must be done beforehand, which usually results in information loss. In the proposed model, we consider continuous emission probability, Gaussian instead of multinomial distribution. This new topic model demonstrates higher performance than standard LDA in the experiments of audio retrieval.

关键词： Topic model LDA Gaussian distribution Audio retrieval

来源：评论

学校读者我要写书评

暂无评论

Bayesian Inference based Temporal Modeling for Naturalistic Affective Expression Classification

Bayesian Inference based Temporal Modeling for Naturalistic ...

引用

5th Biannual Conference of the Humaine-Association on Affective Computing and Intelligent Interaction (ACII)

作者： Chao, Linlin Tao, Jianhua Yang, Minghao Li, Ya Chinese Acad Sci Inst Automat Natl Lab Pattern Recognit NLPR Beijing Peoples R China

ISBN: (纸本)9780769550480

in real life, the affective state of human beings changes gradually and smoothly. There is a high probability that the affective state of a certain moment depends on the states of a previous period. In this study, we propose to explicitly model the temporal relationship using a Bayesian fusion based two-stage classification approach. This approach could involve knowledge about the dynamics of affective states during a period, so that the inferred affective states are predicted by considering a certain amount of context. Evaluations on the Audio Sub-Challenge of the 2011 Audio/Visual Emotion Challenge show our approach obtains competitive results to those of Audio Sub-Challenge winners. The temporal context modeling method proposed in this paper is also helpful for other sequential pattern recognition problems.

关键词： emotion recognition affective dimensions Bayesian fusion temporal modeling Markov Chain

来源：评论

学校读者我要写书评

暂无评论

Partial discriminative training of neural networks for classification of overlapping classes

引用

3rd IAPR Workshop on Artificial Neural Networks in pattern recognition

作者： Liu, Cheng-Lin Chinese Acad Sci Inst Automat Natl Lab Pattern Recognit NLPR Beijing 100190 Peoples R China

ISBN: (纸本)9783540699385

In applications such as character recognition, some classes are heavily overlapped but are not necessarily to be separated. For classification of such overlapping classes, either discriminating between them or merging them into a metaclass does not satisfy. Merging the overlapping classes into a metaclass implies that with in- metaclass substitution is considered as correct classification. For such classification problems, I propose a partial discriminative training (PDT) scheme for neural networks, in which, a training pattern of an overlapping class is used as a positive sample of its labeled class, and neither positive nor negative sample for its allied classes (classes overlapping with the labeled class). In experiments of handwritten letter recognition using neural networks and support vector machines, the PDT scheme mostly outperforms crosstraining (a scheme for multi-labeled classification), ordinary discriminative training and metaclass classification.

关键词： Merging

来源：评论

学校读者我要写书评

暂无评论

Choosing Better Seeds for Entity Set Expansion by Leveraging Wikipedia Semantic Knowledge

Choosing Better Seeds for Entity Set Expansion by Leveraging...

引用

Chinese Conference on pattern recognition

作者： Qi, Zhenyu Liu, Kang Zhao, Jun Chinese Acad Sci Inst Automat Natl Lab Pattern Recognit NLPR Beijing 100190 Peoples R China

ISBN: (纸本)9783642335051

Entity Set Expansion, which refers to expanding a human-input seed set to a more complete set which belongs to the same semantic category, is an important task for open information extraction. Because human-input seeds may be ambiguous, sparse etc., the quality of seeds has a great influence on expansion performance, which has been proved by many previous researches. To improve seeds quality, this paper proposes a novel method which can choose better seeds from original input ones. In our method, we leverage Wikipedia semantic knowledge to measure semantic relatedness and ambiguity of each seed. Moreover, to avoid the sparseness of the seed, we use web corpus to measure its population. Lastly, we use a linear model to combine these factors to determine the final selection. Experimental results show that new seed sets chosen by our method can improve expansion performance by up to average 13.4% over random selected seed sets.

关键词： information extraction seed set refinement semantic knowledge

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：