Teacher-student complementary sample contrastive distillation.

Knowledge distillation (KD) is a widely adopted model compression technique for improving the performance of compact student models, by utilizing the "dark knowledge" of a large teacher model. However, previous studies have not adequately investigated the effectiveness of supervision from the teacher model, and overconfident predictions in the student model may degrade its performance. In this work, we propose a novel framework, Teacher-Student Complementary Sample Contrastive Distillation (TSCSCD), that alleviate these challenges. TSCSCD consists of three key components: Contrastive Sample Hardness (CSH), Supervision Signal Correction (SSC), and Student Self-Learning (SSL). Specifically, CSH evaluates the teacher's supervision for each sample by comparing the predictions of two compact models, one distilled from the teacher and the other trained from scratch. SSC corrects weak supervision according to CSH, while SSL employs integrated learning among multi-classifiers to regularize overconfident predictions. Extensive experiments on four real-world datasets demonstrate that TSCSCD outperforms recent state-of-the-art knowledge distillation techniques.

Discriminative and Geometry-Preserving Adaptive Graph Embedding for dimensionality reduction.

Learning graph embeddings for high-dimensional data is an important technology for dimensionality reduction. The learning process is expected to preserve the discriminative and geometric information of high-dimensional data in a new low-dimensional subspace via either manual or automatic graph construction. Although both manual and automatic graph constructions can capture the geometry and discrimination of data to a certain degree, they working alone cannot fully explore the underlying data structure. To learn and preserve more discriminative and geometric information of the high-dimensional data in the low-dimensional subspace as much as possible, we develop a novel Discriminative and Geometry-Preserving Adaptive Graph Embedding (DGPAGE). It systematically integrates manual and adaptive graph constructions in one unified graph embedding framework, which is able to effectively inject the essential information of data involved in predefined graphs into the learning of an adaptive graph, in order to achieve both adaptability and specificity of data. Learning the adaptive graph jointly with the optimized projections, DGPAGE can generate an embedded subspace that has better pattern discrimination for image classification. Results derived from extensive experiments on image data sets have shown that DGPAGE outperforms the state-of-the-art graph-based dimensionality reduction methods. The ablation studies show that it is beneficial to have an integrated framework, like DGPAGE, that brings together the advantages of manual/adaptive graph construction.

Collaborative Knowledge Distillation via Multiknowledge Transfer.

Knowledge distillation (KD), as an efficient and effective model compression technique, has received considerable attention in deep learning. The key to its success is about transferring knowledge from a large teacher network to a small student network. However, most existing KD methods consider only one type of knowledge learned from either instance features or relations via a specific distillation strategy, failing to explore the idea of transferring different types of knowledge with different distillation strategies. Moreover, the widely used offline distillation also suffers from a limited learning capacity due to the fixed large-to-small teacher-student architecture. In this article, we devise a collaborative KD via multiknowledge transfer (CKD-MKT) that prompts both self-learning and collaborative learning in a unified framework. Specifically, CKD-MKT utilizes a multiple knowledge transfer framework that assembles self and online distillation strategies to effectively: 1) fuse different kinds of knowledge, which allows multiple students to learn knowledge from both individual instances and instance relations, and 2) guide each other by learning from themselves using collaborative and self-learning. Experiments and ablation studies on six image datasets demonstrate that the proposed CKD-MKT significantly outperforms recent state-of-the-art methods for KD.

A class-specific mean vector-based weighted competitive and collaborative representation method for classification.

Collaborative representation-based classification (CRC), as a typical kind of linear representation-based classification, has attracted more attention due to the effective and efficient pattern classification performance. However, the existing class-specific representations are not competitively learned from collaborative representation for achieving more informative pattern discrimination among all the classes. With the purpose of enhancing the power of competitive and discriminant representations among all the classes for favorable classification, we propose a novel CRC method called the class-specific mean vector-based weighted competitive and collaborative representation (CMWCCR). The CMWCCR mainly contains three discriminative constraints including the competitive, mean vector and weighted constraints that fully employ the discrimination information in different ways. In the competitive constraint, the representations from any one class and the other classes are adapted for learning competitive representations among all the classes. In the newly designed mean vector constraint, the mean vectors of all the class-specific training samples with the corresponding class-specific representations are taken into account to further enhance the competitive representations. In the devised weighted constraint, the class-specific weights are constrained on the representation coefficients to make the similar classes have more representation contributions to strengthening the discrimination among all the class-specific representations. Thus, these three constraints in the unified CMWCCR model can complement each other for competitively learning the discriminative class-specific representations. To verify the CMWCCR classification performance, the extensive experiments are conducted on twenty-eight data sets in comparisons with the state-of-the-art representation-based classification methods. The experimental results show that the proposed CMWCCR is an effective and robust CRC method with satisfactory performance.

Regularization on Augmented Data to Diversify Sparse Representation for Robust Image Classification.

Image classification is a fundamental component in modern computer vision systems, where sparse representation-based classification has drawn a lot of attention due to its robustness. However, on the optimization of sparse learning systems, regularization and data augmentation are both powerful, but currently isolated. We believe that regularization and data augmentation can cooperate to generate a breakthrough in robust image classification. In this article, we propose a novel framework, regularization on augmented data (READ), which creates diversification in the data using the generic augmentation techniques to implement robust sparse representation-based image classification. When the training data are augmented, READ applies a distinct regularizer, l1 or l2 , in particular, on the augmented training data apart from the original data, so that regularization and data augmentation are utilized and enhanced synchronously. We introduce an elaborate theoretical analysis on how to optimize the sparse representation by both l1 -norm and l2 -norm with the generic data augmentation and demonstrate its performance in extensive experiments. The results obtained on several facial and object datasets show that READ outperforms many state-of-the-art methods when using deep features.

Weighted discriminative collaborative competitive representation for robust image classification.

Collaborative representation-based classification (CRC) is a famous representation-based classification method in pattern recognition. Recently, many variants of CRC have been designed for many classification tasks with the good classification performance. However, most of them ignore the inter-class pattern discrimination among the class-specific representations, which is very critical for strengthening the pattern discrimination of collaborative representation (CR). In this article, we propose a novel CR approach for image classification, called weighted discriminative collaborative competitive representation (WDCCR). The proposed WDCCR designs the discriminative and competitive collaborative representation among all the classes by fully considering the class information. On the one hand, we incorporate two discriminative constraints into the unified WDCCR model. Both constraints are the competitive class-specific representation residuals and the pairs of class-specific representations for each query sample. On the other hand, the constraint of the weighted categorical representation coefficients is introduced into the proposed model for further enhancing the power of discriminative and competitive representation. In the weighted constraint, we assume that the different classes of each query sample should have less contribution to the representation with the small representation coefficients, and then two types of weight factors are designed to constrain the representation coefficients. Furthermore, the robust WDCCR (R-WDCCR) is proposed with l1-norm representation fidelity for recognizing noisy images. Extensive experiments on six image data sets demonstrate the effective and robust superiorities of the proposed WDCCR and R-WDCCR over the related state-of-the-art representation-based classification methods.

Sexual dimorphism in the Chinese endemic species Pachyhynobius shangchengensis Fei, Qu and Wu, 1983 (Urodela: Hynobiidae).

Sexual dimorphism (SD) is a widespread phenomenon in most vertebrate species and is exhibited in a myriad of ways. In amphibians, sexual size dimorphism, in which females are larger than males, is the most common type, and sexual shape dimorphism varies among species. Different selection forces (sexual selection, fecundity selection, and ecological selection) that act differently upon the sexes form the consequence of SD. Thus, studies of SD provide information about the general intersexual divergence of the same species and allow insights into the impact of selective forces on the sexes. In this study, we analyzed morphometric data of the Shangcheng stout salamander, Pachyhynobius shangchengensis, an endemic and poorly known Chinese salamander, to examine sexual dimorphism in size and shape. The morphometric data included 15 characteristics of 68 females and 55 males which were analyzed using univariate and multivariate methods. A significant difference was found between the sexes in terms of both body size (snout-vent length) and some body shapes (e.g., head length and width, tail length and width, distance between limbs, and limb length and width) in this salamander. The longer snout-vent length in males may be attributed to sexual selection, longer and wider head in males may contribute to male-male competition, longer and wider tail in males may be attributed to energy storage and reproductive success, the larger distance between limbs in females is likely due to a fecundity advantage, and longer and more robust limbs in males may be related to reproductive or competitive behaviors. These results demonstrated that sexual dimorphism of different morphological traits is the consequence of different selection forces that act differently upon the sexes.

Maximum neighborhood margin discriminant projection for classification.

We develop a novel maximum neighborhood margin discriminant projection (MNMDP) technique for dimensionality reduction of high-dimensional data. It utilizes both the local information and class information to model the intraclass and interclass neighborhood scatters. By maximizing the margin between intraclass and interclass neighborhoods of all points, MNMDP cannot only detect the true intrinsic manifold structure of the data but also strengthen the pattern discrimination among different classes. To verify the classification performance of the proposed MNMDP, it is applied to the PolyU HRF and FKP databases, the AR face database, and the UCI Musk database, in comparison with the competing methods such as PCA and LDA. The experimental results demonstrate the effectiveness of our MNMDP in pattern classification.