Search

Despite achieving state-of-the-art performance, deep learning methods generally require a large amount of labeled data during training and may suffer from overfitting when the sample size is small. To ensure good generalizability of deep networks under small sample sizes, learning discriminative features is crucial. To this end, several loss functions have been proposed to encourage large intra-class compactness and inter-class separability. In this paper, we propose to enhance the discriminative power of features from a new perspective by introducing a novel neural network termed Relation-and-Margin learning Network (ReMarNet). Our method assembles two networks of different backbones so as to learn the features that can perform excellently in both of the aforementioned two classification mechanisms. Specifically, a relation network is used to learn the features that can support classification based on the similarity between a sample and a class prototype; at the meantime, a fully connected network with the cross entropy loss is used for classification via the decision boundary. Experiments on four image datasets demonstrate that our approach is effective in learning discriminative features from a small set of labeled samples and achieves competitive performance against state-of-the-art methods. Code is available at https://github.com/liyunyu08/ReMarNet. [ABSTRACT FROM AUTHOR]

The performance of distance-based classifiers heavily depends on the underlying distance metric, so it is valuable to learn a suitable metric from the data. To address the problem of multimodality, it is desirable to learn local metrics. In this short paper, we define a new intuitive distance with local metrics and influential regions, and subsequently propose a novel local metric learning algorithm called LMLIR for distance-based classification. Our key intuition is to partition the metric space into influential regions and a background region, and then regulate the effectiveness of each local metric to be within the related influential regions. We learn multiple local metrics and influential regions to reduce the empirical hinge loss, and regularize the parameters on the basis of a resultant learning bound. Encouraging experimental results are obtained from various public and popular data sets. [ABSTRACT FROM AUTHOR]

In this paper we propose simple, general tiered classifiers for relatively complex data. Empirical studies on real and simulated data show that three two-tier classifiers, which are respective extensions of linear discriminant analysis, linear logistic regression and support vector machines, can reduce noticeably the relatively high misclassification error of their original single-tier counterparts, without significantly increasing computational labour. [ABSTRACT FROM PUBLISHER]

Abstract: In statistical pattern classification, generative approaches, such as linear discriminant analysis (LDA), assume a data-generating process (DGP), whereas discriminative approaches, such as linear logistic regression (LLR), do not model the DGP. In general, a generative classifier performs better than its discriminative counterpart if the DGP is well-specified and worse than the latter if the DGP is clearly mis-specified. In view of this, this paper presents a joint discriminative–generative modelling (JoDiG) approach, by partitioning predictor variables X into two sub-vectors, namely , to which a generative approach is applied, and , to be treated by a discriminative approach. This partitioning of X is based on statistical tests of the assumed DGP: the variables that clearly fail the tests are grouped as and the rest as . Then the generative and discriminative approaches are combined in a probabilistic rather than a heuristic way. The principle of the JoDiG approach is quite generic, but for illustrative purposes numerical studies of the paper focus on a widely-used case, in which the DGP assumes a multivariate normal distribution for each class. In this case, the JoDiG approach uses LDA for and LLR for . Numerical experiments on real and simulated data demonstrate that the performance of this new approach to classification is similar to or better than that of its discriminative and generative counterparts, in particular when the size of the training-set is comparable to the dimension of the data. [Copyright &y& Elsevier]

Metric-based methods are one of the most common methods to solve the problem of few-shot image classification. However, traditional metric-based few-shot methods suffer from overfitting and local feature misalignment. The recently proposed feature reconstruction-based approach, which reconstructs query image features from the support set features of a given class and compares the distance between the original query features and the reconstructed query features as the classification criterion, effectively solves the feature misalignment problem. However, the issue of overfitting still has not been considered. To this end, we propose a self-reconstruction metric module for diversifying query features and a restrained cross-entropy loss for avoiding over-confident predictions. By introducing them, the proposed self-reconstruction network can effectively alleviate overfitting. Extensive experiments on five benchmark fine-grained datasets demonstrate that our proposed method achieves state-of-the-art performance on both 5-way 1-shot and 5-way 5-shot classification tasks. Code is available at https://github.com/liz-lut/SRM-main. • A novel fine-grained few-shot classification method is proposed to alleviate overfitting. • The proposed method uses self-reconstruction to increase the diversity of query features. • The proposed method introduces a restrained cross-entropy loss. • The proposed method generalizes well across five benchmark datasets. [ABSTRACT FROM AUTHOR]

Few-shot classification (FSC) is a highly challenging task, as only a small number of labeled samples are available when identifying new categories. Distance metric learning-based methods have emerged as a prominent approach to FSC, which typically use a distance function to measure the difference between query and support samples for identifying the class membership of the query sample. However, these methods simply treat each channel difference between query and support features equally when computing the class scores. Since different channels in the learned feature seek for different patterns, these distance metrics fail to consider that different channels are of different importance to FSC, and thus cannot accurately measure the similarity between samples. To address this issue, we propose a Q uery- C entric D istance M odulator (QCDM) to generate query-related weights for each channel difference adaptively. Specifically, since the distribution of difference between a query sample and all support samples in a particular channel can reflect the importance of this channel to the classification of the query sample, we take this difference vector as input and generate a query-specific channel weight through a meta-network. QCDM can guide FSC models to focus on discriminative channel differences and achieve better generalization. QCDM is a plug-and-play module that can be seamlessly integrated with existing distance metric learning-based FSC methods. Extensive experimental results indicate that our method can effectively improve the performance of distance metric learning-based FSC methods. The source code is available in https://github.com/Wu-Wenxiao/QCDM. • A plug-and-play query-centric distance modulator for distance metric learning-based FSC methods. • Weight generation through data inconsistency within each channel from different classes. • Decoupled from the process of feature extraction and integrated with any given backbone. [ABSTRACT FROM AUTHOR]

Deep neural networks have recently shown excellent performance on numerous image classification tasks. These networks often need to estimate a large number of parameters and require much training data. When the amount of training data is small, however, a network with high flexibility quickly overfits the training data, resulting in a large model variance and poor generalization. To address this problem, we propose a new, simple yet effective ensemble method called InterBoost for small-sample image classification. In the training phase, InterBoost first randomly generates two sets of complementary weights for training data, which are used for separately training two base networks of the same structure, and then the two sets of complementary weights are updated for refining the training of the networks through interaction between the two base networks previously trained. This interactive training process continues iteratively until a stop criterion is met. In the testing phase, the outputs of the two networks are combined to obtain one final score for classification. Experimental results on four small-sample datasets, UIUC-Sports, LabelMe, 15Scenes and Caltech101, demonstrate that the proposed ensemble method outperforms existing ones. Moreover, results from the Wilcoxon signed-rank tests show that our method is statistically significantly better than the methods compared. Detailed analysis is also provided for an in-depth understanding of the proposed method. [ABSTRACT FROM AUTHOR]