Search

This paper studies a new learning paradigm for noisy labels, i.e., noisy correspondence (NC). Unlike the well-studied noisy labels that consider the errors in the category annotation of a sample, the NC refers to the errors in the alignment relationship of two data points. Although such false positive pairs are common especially in the data harvested from the Internet, which however are neglected by most existing works. By taking cross-modal retrieval as a showcase, we propose a method called learning with noisy correspondence (LNC). In brief, the LNC first roughly obtains the clean and noisy subsets from the original data and then rectifies the false positive pairs by using a novel adaptive prediction function. Finally, the LNC adopts a novel triplet loss with soft margins to endow cross-modal retrieval the robustness to the NC. To verify the effectiveness of the proposed LNC, we conduct experiments on six benchmark datasets in image-text and video-text retrieval tasks. Besides the effectiveness of the LNC, the experimental results show the necessity of the explicit solution to the NC faced by not only the standard model training paradigm but also the pre-training and fine-tuning paradigms. [ABSTRACT FROM AUTHOR]

Image deblurring is an essential problem in computer vision. Due to highly structured and special facial components (e.g. eyes), most general image deblurring methods and face deblurring methods failed to yield more explicit structure and facial details, resulting in too smooth, uncoordinated and distorted face structure. Considering the unique face texture and sufficient facial details, we present an effective face deblurring network by exploiting more regularized structure and enhanced texture information (RSETNet). We first incorporate the face parsing network with fine-tuning to obtain more accurate face structure, and we present the feature adaptive denormalization (FAD) to regularize the facial structure as a condition of auxiliary to generate more harmonious and undistorted face structure. Meanwhile, to improve the generated facial texture information, we propose a new Laplace depth-wise separable convolution (LDConv) and multi-patch discriminator. Compared with existing methods, our face deblurring method could restore face structure more accurately and with more facial details. Experiments on two public face datasets have demonstrated the effectiveness of our proposed methods in terms of qualitative and quantitative indicators. [ABSTRACT FROM AUTHOR]

Computer-aided drug design (CADD) has emerged as a highly effective and indispensable tool for streamlining the drug discovery process, leading to significant reductions in cost and time. The integration of CADD with machine learning (ML) and deep learning (DL) technologies further enhances its potential and promises novel advancements in the field. In this article, we provide a review of the computational methods employed in the development of novel anesthetics, outlining their respective advantages and limitations. These techniques have demonstrated their utility across various stages of drug discovery, encompassing the exploration of target-ligand interactions, identification and validation of new binding sites, de novo drug design, evaluation and optimization of absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties in lead compounds, as well as prediction of adverse effects. Through an in-depth exploration of computational approaches and their applications, this article aims to help relevant researchers develop safer and more effective anesthetic drugs. [ABSTRACT FROM AUTHOR]

Circular RNAs (circRNAs) have been extensively studied in tumor development and treatment. CircZNF609 (hsa_circ_0000615) has been shown to serve as an oncogene in all kinds of solid tumors and may act as the novel biomarker in tumor diagnosis and therapy in tumor early diagnosis and therapy. However, the underlying character and mechanism of circZNF609 in cisplatin chemosensitivity and bladder cancer (BCa) development were unknown. The expression level of cell division cycle 25B (CDC25B), microRNA 1200 (miR-1200), and circZNF609 in BCa cells and tissues depended on quantitative real-time PCR (qRT-PCR). CDC25B protein level was assayed with Western blot. Functional assays in vitro and in vivo had been conducted to inspect the important role of circZNF609 on BCa progression and cisplatin chemosensitivity in BCa. RNA sequencing and online databases were used to predict the interactions among circZNF609, miR-1200, and CDC25B. Mechanistic exploration was confirmed by RNA pull-down assay, RNA fluorescence in situ hybridization (FISH) and Dual luciferase reporter assay. CircZNF609 expression was increased significantly in BCa cell lines and tissues. For BCa patients, increased expression of circZNF609 was correlated with a worse survival. In vitro and in vivo, enforced expression of circZNF609 enhanced BCa cells proliferation, migration, and cisplatin chemoresistance. Mechanistically, circZNF609 alleviated the inhibition effect on target CDC25B expression by sponging miR-1200. CircZNF609 promoted tumor growth through novel circZNF609/miR-1200/CDC25B axis, implying that circZNF609 has significant potential to act as a new diagnostic biomarker and therapeutic target in BCa. Enhancing cisplatin sensitivity is an important direction for bladder cancer management. 1. This research reveals that circZNF609 improves bladder cancer progression and inhibits cisplatin sensitivity by inducing G1/S cell cycle arrest via a novel miR-1200/CDC25B cascades. 2. CircZNF609 was confirmed associated with worse survival of bladder cancer patients. 3. CircZNF609 act as a prognostic biomarker for bladder cancer treatment. [ABSTRACT FROM AUTHOR]

Single-cell multi-omics data integration aims to reduce the omics difference while keeping the cell type difference. However, it is daunting to model and distinguish the two differences due to cell heterogeneity. Namely, even cells of the same omics and type would have various features, making the two differences less significant. In this work, we reveal that instead of being an interference, cell heterogeneity could be exploited to improve data integration. Specifically, we observe that the omics difference varies in cells, and cells with smaller omics differences are easier to be integrated. Hence, unlike most existing works that homogeneously treat and integrate all cells, we propose a multi-omics data integration method (dubbed scBridge) that integrates cells in a heterogeneous manner. In brief, scBridge iterates between i) identifying reliable scATAC-seq cells that have smaller omics differences, and ii) integrating reliable scATAC-seq cells with scRNA-seq data to narrow the omics gap, thus benefiting the integration for the rest cells. Extensive experiments on seven multi-omics datasets demonstrate the superiority of scBridge compared with six representative baselines. Multi-omics data integration can be challenging in the event of cell heterogeneity. Here, the authors present scBridge, a method that exploits heterogeneous omics differences, to progressively integrate cells and narrows omics gap, leading to promising integration and label transfer results. [ABSTRACT FROM AUTHOR]

Hand gesture and action recognition have been extensively researched in the past two decades due to the emerging advanced acquisition and interaction technologies, which open the floodgates for a vast range of potential applications. Particularly, many spatial–temporal feature extractors have been proposed, such as RNNs-based models, temporal convolutional network (TCN), and 3D convolutional neural networks (3DCNN) for modeling long-term dependencies in sequential data. However, it remains challenging to obtain a high recognition rate because of the difficulty of effectively extracting spatial–temporal features and efficiently classifying them with noisy and complex skeleton sequences. Therefore, this paper proposes a deep ensemble framework called multi-model ensemble gesture recognition network (MMEGRN) for skeleton-based hand gesture recognition. Specifically, to establish effective feature extraction and accurate gesture recognition, we propose an architecture consisting of four sub-networks, three spatio-temporal features classifiers to leverage their various capabilities of extracting and classifying skeleton sequences. Through late feature fusion, the features resulted from the feature extractors of each sub-network are fused into a new fusion classifier. Each subnetwork is trained independently to perform the task of gesture recognition using only skeleton joints. The training is performed using the cyclic annealing learning rate to generate a series of models that are combined in an ensemble using the optimized weighted ensemble (OWE) method. The proposed framework combines deep learning and ensemble strengths to establish a new deep-learning network architecture for more accurate and efficient hand gesture recognition. Extensive experiments on three skeleton-based hand gesture recognition datasets have shown the effectiveness of the proposed framework and the superiority over other models in terms of recognition accuracy. [ABSTRACT FROM AUTHOR]

Neural networks handling data imbalance heavily rely on resampling or reweighting strategies. However, existing resampling and reweighting approaches mainly focus on rebalancing known data, which ignore the essence of the data imbalance problem, namely, the problem of insufficient empirical representation of the minority class caused by the small number of samples. Therefore, we propose a new solution for neural networks classifying imbalanced data by sampling absent minority class samples. Specifically, an improved Metropolis Hasting (IMH) algorithm is developed to sample absent minority class samples by collecting samples rejected by the majority class approximation process. The sampled absent minority samples are then provided to neural networks to address the data imbalance problem. For IMH, in order to accelerate the sampling process and reduce the vague class definition of the sampled minority class samples, line segment transition kernel and class probability constraint are proposed. For neural networks, two boundary shifting strategies are supported to operate on different application modes of sampled absent minority class samples. In experiments, the proposed method is validated on 34 imbalanced datasets. Comparable AUC, G-MEAN, and MACC results are achieved. These results demonstrate the effectiveness of sampling absent minority class samples for neural networks solving the imbalanced data problem. [ABSTRACT FROM AUTHOR]

Blind face deblurring aims to recover a sharper face from its unknown degraded version (i.e., different motion blur, noise). However, most previous works typically rely on degradation facial priors extracted from low-quality inputs, which generally leads to unlifelike deblurring results. In this paper, we propose a multi-scale progressive face-deblurring generative adversarial network (MPFD-GAN) that requires no facial priors to generate more realistic multi-scale deblurring results by one feed-forward process. Specifically, MPFD-GAN mainly includes two core modules: the feature retention module and the texture reconstruction module (TRM). The former can capture non-local similar features by full advantage of the different receptive fields, which facilitates the network to recover the complete structure. The latter adopts a supervisory attention mechanism that fully utilizes the recovered low-scale face to refine incoming features at every scale before propagating them further. Moreover, TRM extracts the high-frequency texture information from the recovered low-scale face by the Laplace operator, which guides subsequent steps to progressively recover faithful face texture details. Experimental results on the CelebA, UTKFace and CelebA-HQ datasets demonstrate the effectiveness of the proposed network, which achieves better accuracy and visual quality against state-of-the-art methods. [ABSTRACT FROM AUTHOR]

In conversational systems, we can use external knowledge to generate more diverse sentences and make these sentences contain actual knowledge. Leveraging knowledge for conversation system is important but challenging. Firstly, the conversation system needs to find the appropriate knowledge. Secondly, the knowledge needs to be coded effectively and generated into fluent utterances. In this paper, we propose a knowledge-driven conversation system to address the above challenges. This system consists of three modules, namely topic predictor, knowledge selector, and dialogue generator. The topic predictor uses a combination of non-deep learning (coarse-grained) and deep learning (fine-grained) to form a rough recall and fine sorting process, and uses them to predict conversation topics. The knowledge selector aims to find the most appropriate knowledge based on the filtered topics. We propose the Bert2Transformer model as our dialogue generator, which can generate rich and fluent utterances based on contextual and relevant knowledge. On the public corpus KdConv, our system outperforms a strong baseline and achieves state-of-the-art results. In the ablation study, we analyze the effectiveness of the proposed components in detail and investigate the performance factors that may affect the knowledge-driven conversation generation. Experimental results show the proposed system achieves a significant improvement compared with traditional baseline methods. The average BLEU scores of our system achieve 35.92 and 23.24, respectively, given appropriate knowledge and without appropriate knowledge, while the Distinct-2 scores of our system achieve 16.32 and 15.93, respectively. The training corpus is publicly available (https://github.com/wulaoshi/dialogue_train_data). [ABSTRACT FROM AUTHOR]

Recently, many studies have been carried out on model compression to handle the high computational cost and high memory footprint brought by the implementation of deep neural networks. In this paper, model compression of convolutional neural networks is constructed as a multiobjective optimization problem with two conflicting objectives, reducing the model size and improving the performance. A novel structured pruning method called Conventional-based and Evolutionary Approaches Guided Multiobjective Pruning (CEA-MOP) is proposed to address this problem, where the power of conventional pruning methods is effectively exploited for the evolutionary process. A delicate balance in pruning rate and model accuracy has been automated achieved by a multiobjective optimization evolutionary model. First, an ensemble framework integrates pruning metrics to establish a codebook for further evolutionary operations. Then, an efficient coding method is developed to shorten the length of chromosome, thus ensuring its superior scalability. Finally, sensitivity analysis is automatically carried out to determine the upper bound of pruning rate for each layer. Notably, on CIFAR-10, CEA-MOP reduces more than 50% FLOPs on ResNet-110 and improves the relative accuracy. Moreover, on ImageNet, CEA-MOP reduces more than 50% FLOPs on ResNet-101 with negligible top-1 accuracy drop. [ABSTRACT FROM AUTHOR]

Chinese couplet consists of two syntactically symmetric sentences which are concise but meaningful and equal in length. Creating Chinese couplets is a challenging task even in the artificial intelligence field. Couplet automatic generation approaches based on maximum likelihood estimation (MLE) have been studied. However, MLE-based methods are limited by exposure bias, and the token-level probability maximizing generation makes the model ignore the global control of sentences, while couplets also have strict requirements at the sentence level between two sentences, such as semantic and sentence pause. In this paper, we propose a GAN-based model to fill this gap, named CoupGAN. Specifically, we introduce a discriminative model to evaluate the global-level quality of generated sentences. Meanwhile, to deal with discrete text data, CoupGAN is trained by a policy gradient-based method. In detail, this method rewards the generator to mitigate exposure bias, which could encourage the model to generate more realistic samples. Experiments among large Chinese couplet corpora datasets demonstrate that our proposed model is superior to existing methods in various evaluation metrics. [ABSTRACT FROM AUTHOR]

Synchronous parameters algorithms with data parallelism have been successfully utilized to accelerate the distributed training of deep neural networks (DNNs). However, a prevalent shortcoming of the synchronous methods is computation waste resulted from the mutual waiting among the computational workers with different performance and the communication delays at each synchronization. To alleviate this drawback, we propose a novel method, free local stochastic gradient descent (FLSGD) with parallel synchronization, to eliminate the waiting and communication overhead. Specifically, the process of distributed DNN training is firstly modeled as a pipeline which assembly consists of three components: dataset partition, local SGD, and parameter updating. Then, a novel adaptive batch size and dataset partition method based on the computational performance of the node is employed to eliminate the waiting time by keeping the load balance of the distributed DNN training. The local SGD and the parameter updating including gradients synchronization are parallelized to eliminate the communication cost by one-step gradient delaying, and the stale problem is remedied by an appropriate approximation. To our best knowledge, this is the first work focusing on decreasing both distributed training load balancing and communication overhead Extensive experiments are conducted with four state-of-the-art DNN models on two image classification datasets (i.e., CIFAR10 and CIFAR100) to demonstrate that the effectiveness of FLSGD outperforms the synchronous methods. [ABSTRACT FROM AUTHOR]

Multi-view spectral clustering has drawn much attention due to the effectiveness of exploiting the similarity relationships among data points. These methods typically reveal the intrinsic structure using a predefined graph for each view. The predefined graphs are fused to a consensus one, on which the final clustering results are obtained. However, such common strategies may lead to information loss because of the inconsistency or noise among multiple views. In this paper, we propose to merge multi-view information in partition level instead of the raw feature space where the data points lie. The partition of each view is treated as a perturbation of the consensus clustering, and the multiple partitions are integrated by estimating a distinct rotation for each partition. The proposed model is formulated as a joint learning framework, i.e., with the input data matrix, our model directly outputs the final discrete clustering result. Hence it is an end-to-end single-stage learning model. An iterative updating algorithm is proposed to solve the learning problem, in which the involved variables can be optimized in a mutual reinforcement manner. Experimental results on real-world data sets illustrate the effectiveness of our model. [ABSTRACT FROM AUTHOR]

There are contrasting results concerning the effect of reactive school closure on SARS-CoV-2 transmission. To shed light on this controversy, we developed a data-driven computational model of SARS-CoV-2 transmission. We found that by reactively closing classes based on syndromic surveillance, SARS-CoV-2 infections are reduced by no more than 17.3% (95%CI: 8.0–26.8%), due to the low probability of timely identification of infections in the young population. We thus investigated an alternative triggering mechanism based on repeated screening of students using antigen tests. Depending on the contribution of schools to transmission, this strategy can greatly reduce COVID-19 burden even when school contribution to transmission and immunity in the population is low. Moving forward, the adoption of antigen-based screenings in schools could be instrumental to limit COVID-19 burden while vaccines continue to be rolled out. Reactive school class closures have been widely implemented to mitigate COVID-19 outbreaks. Here, the authors show that, compared to symptom-prompted PCR testing, screening for cases in schools with antigen tests leads to greater reductions in infection rates in both students and the wider community. [ABSTRACT FROM AUTHOR]

In the context of human-robot and robot-robot interactions, the better cooperation can be achieved by predicting the other party's subsequent actions based on the current action of the other party. The time duration for adjustment is not sufficient provided by short term forecasting models to robots. A longer duration can by achieved by mid-term forecasting. But the mid-term forecasting models introduce the previous errors into the follow-up forecasting and amplified gradually, eventually invalidating the forecasting. A new mid-term forecasting with error suppression based on restricted Boltzmann machine(RBM) is proposed in this paper. The proposed model can suppress the error amplification by replacing the previous inputs with their features, which are retrieved by a deep belief network(DBN). Furthermore, a new mechanism is proposed to decide whether the forecasting result is accepted or not. The model is evaluated with several datasets. The reported experiments demonstrate the superior performance of the proposed model compared to the state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Heart sound segmentation (HSS) aims to detect the four stages (first sound, systole, second heart sound and diastole) from a heart cycle in a phonocardiogram (PCG), which is an essential step in automatic auscultation analysis. Traditional HSS methods need to manually extract the features before dealing with HSS tasks. These artificial features highly rely on extraction algorithms, which often result in poor performance due to the different operating environments. In addition, the high-dimension and frequency characteristics of audio also challenge the traditional methods in effectively addressing HSS tasks. This paper presents a novel end-to-end method based on convolutional long short-term memory (CLSTM), which directly uses audio recording as input to address HSS tasks. Particularly, the convolutional layers are designed to extract the meaningful features and perform the downsampling, and the LSTM layers are developed to conduct the sequence recognition. Both components collectively improve the robustness and adaptability in processing the HSS tasks. Furthermore, the proposed CLSTM algorithm is easily extended to other complex heart sound annotation tasks, as it does not need to extract the characteristics of corresponding tasks in advance. In addition, the proposed algorithm can also be regarded as a powerful feature extraction tool, which can be integrated into the existing models for HSS. Experimental results on real-world PCG datasets, through comparisons to peer competitors, demonstrate the outstanding performance of the proposed algorithm. [ABSTRACT FROM AUTHOR]

As a written language used for thousands of years, Ancient Chinese has some special characteristics like complex semantics as polysemy and the one-to-many alignment with Modern Chinese. Thus it may be translated in a large number of fully different but equally correct ways. In the absence of multiple references, reference-dependent evaluations like Bilingual Evaluation Understudy (BLEU) cannot identify potentially correct translation results. The explore on automatic evaluation of Ancient-Modern Chinese Translation is completely lacking. In this paper, we proposed an automatic evaluation metric for Ancient-Modern Chinese Translation called DTE (Dual-based Translation Evaluation), which can be used to evaluate one-to-many alignment in the absence of multiple references. When using DTE to evaluate, we found that the proper nouns often could not be correctly translated. Hence, we designed a new word segmentation method to improve the translation of proper nouns without increasing the size of the model vocabulary. Experiments show that DTE outperforms several general evaluations in terms of similarity to the evaluation of human experts. Meanwhile, the new word segmentation method promotes the Ancient-Modern Chinese translation models perform better on proper nouns' translation, and get higher scores on both BLEU and DTE. [ABSTRACT FROM AUTHOR]

Multimodal image-to-image translation based on generative adversarial networks (GANs) shows suboptimal performance in the visual domains with high internal variability, e.g., translation from multiple breeds of cats to multiple breeds of dogs. To alleviate this problem, we recast the training procedure as modeling distinct distributions which are observed sequentially, for example, when different classes are encountered over time. As a result, the discriminator may forget about the previous target distributions, known as catastrophic forgetting, leading to non-/slow convergence. Through experimental observation, we found that the discriminator does not always forget the previously learned distributions during training. Therefore, we propose a novel generator regulating GAN (GR-GAN). The proposed method encourages the discriminator to teach the generator more effectively when it remembers more of the previously learned distributions, while discouraging the discriminator to guide the generator when catastrophic forgetting happens on the discriminator. Both qualitative and quantitative results show that the proposed method is significantly superior to the state-of-the-art methods in handling the image data that are with high variability. [ABSTRACT FROM AUTHOR]

Computer technology provides new conditions and possibilities for art creation and research, and also expands the forms of artistic expression. Computer-created art has thus become one of the important forms of art. In this paper, we proposed a novel method of generating abstract paintings. We used the public painting dataset WikiArt and designed a K-Means algorithm that automatically finds the optimal K value to perform color segmentation on these images, and divide the picture into different color blocks. We proposed the concept of the collection of color block (CoCB), which records all color block information of the segmented image and serves as an intermediate vector for the generation of abstract painting. We extracted the CoCB as an empirical sample and used a learning model based on deep learning to automatically generate brand-new CoCBs. We then converted the CoCBs into an abstract painting, so that the generated abstract painting also followed certain aesthetic rules. Experiments showed that the resulting abstract painting have great visual impact, and some of them have been installed as decorations in public and private spaces, as well as art institutions. Also, some artists and designers have used the results in their work. [ABSTRACT FROM AUTHOR]

Nasopharyngeal carcinoma (NPC) is one of the most common cancers of the nasopharynx. A structural analysis of NPC can provide vital insights into methods of treatment. However, manually marking the boundaries of NPC in images is tedious, time-consuming, and prone to error. It has become necessary to use computer-based automatic segmentation algorithms to accurately locate NPC. However, this remains a challenging task owing to the high variation (in shape and size) in the structure of the nasopharynx across subjects. Moreover, the nasopharyngeal area is small, and this causes severe imbalance in the foreground and background categories. In this paper, we propose a 3D convolutional neural network with long-range skip connection and multi-scale feature pyramid (SFP) for the segmentation of images of NPC. Unlike the traditional skip connection in residual blocks, which only considers the feature transfer and feature fusion between the same convolutional layer, long-range skip connection with original features from the first convolution in our network is passed to each down-sampling stage using element-wise sum to effectively increase reuse of low-level features and to solve the problems of gradient disappearance and explosion. The multi-scale feature pyramid with a varying atrous rate adapts to images of different sizes to learn multi-scale features, and hierarchical contextual information regarding NPC. To accelerate the convergence of our network, we use deep supervision to generate three auxiliary segmentation maps and merge the weighted loss into the objective function. And we fuse these auxiliary segmentation maps to refine the final segmentation result. In our experiments, the proposed network was trained and tested on 3D magnetic resonance imaging (MRI) images of 120 clinical patients using 5-fold cross-validation. The average dice similarity coefficient (DSC) and average symmetric surface distance (ASSD), used as evaluation metric, were 0.737 and 1.214 mm, respectively. This shows that in terms of results, our method is superior to five state-of-the-art networks and equivalent to the judgment of an experienced physician. [ABSTRACT FROM AUTHOR]

As an important format of multimedia, music has filled almost everyone's life. Automatic analyzing of music is a significant step to satisfy people's need for music retrieval and music recommendation in an effortless way. Thereinto, downbeat tracking has been a fundamental and continuous problem in Music Information Retrieval (MIR) area. Despite significant research efforts, downbeat tracking still remains a challenge. Previous researches either focus on feature engineering (extracting certain features by signal processing, which are semi-automatic solutions); or have some limitations: they can only model music audio recordings within limited time signatures and tempo ranges. Recently, deep learning has surpassed traditional machine learning methods and has become the primary algorithm in feature learning; the combination of traditional and deep learning methods also has made better performance. In this paper, we begin with a background introduction of downbeat tracking problem. Then, we give detailed discussions of the following topics: system architecture, feature extraction, deep neural network algorithms, data sets, and evaluation strategy. In addition, we take a look at the results from the annual benchmark evaluation—Music Information Retrieval Evaluation eXchange—as well as the developments in software implementations. Although much has been achieved in the area of automatic downbeat tracking, some problems still remain. We point out these problems and conclude with possible directions and challenges for future research. [ABSTRACT FROM AUTHOR]

During data acquisition, samples in a time series may contain noise, such as inconsistent data ranges, inconsistent data, and incomplete data. Therefore, the classification model requires improved robustness to correctly classify the sequence of human motion. This paper presents a classification model with improved robustness performance based on the factored gated restricted Boltzmann machine to effectively overcome the various aforementioned data problems. The proposed model acquires the deformation code of each action first and integrates the deformation codes together to be an integrated deformation code of the entire sequence. Then, the model determines the classification from the integrated deformation code. This approach mainly focuses on the deformation relations among action samples in the extraction sequence, and it ignores the data expression in the sequence samples. Experiments show that the proposed model performs better than state-of-the-art approaches in terms of the robustness of time series classification with noise. [ABSTRACT FROM AUTHOR]

The main challenge for most image–text tasks, such as zero-shot, is the way to measure the semantic similarity between visual and textual feature vectors. The common solution is to map the image feature vectors and text feature vectors into the Hilbert space and then rank the similarity by the inner product between feature vectors. In this paper, we learn the feature representation of images and their sentence descriptions by different deep neural networks to learn about the inner-modal correspondences between visual and language data. We then use a joint embedding structure based on angle calculation for measuring the semantic similarity between visual and textual features. In the proposed method, a constant factor b keeps the similarities of positive samples and negative samples at a certain distance. Since the proposed cosine similarity method involves both normalization and vectors computation, we also develop the learning algorithm on neural networks for expressing the semantic features of texts and images. We applied the angle-based method to the challenging Caltech-UCSD Birds and the Oxford-102 Flowers datasets. The experiments demonstrate good performances on both recognition and retrieval tasks. [ABSTRACT FROM AUTHOR]

Manifold learning (ML) is a research topic of great interest in the field of machine learning that aims to determine the appropriate low-dimensional embeddings of data. The embeddings should preserve the intrinsic structure of the data manifold. Many ML techniques have been proposed to learn the underlying manifold of data. It is crucial to effectively evaluate the quality of the corresponding embedding results when selecting an appropriate ML technique in practice. However, there is a lack of effective embedding quality assessment (EQA) criteria to evaluate the embedding quality. In this paper, a new local included angles preservation (LUNA) criterion is proposed to evaluate the embedding quality. Unlike previous EQA methods that mainly focus on local neighborhood preservation performance or the preservation of the global geometric structure, the proposed LUNA criterion incorporates both an assessment of the neighborhood preserving capacity and local included angle holding performance. By introducing an effective evaluation of the performance of local included angle preservation, we show that the LUNA criterion could provide a more reasonable quality assessment than conventional criteria. The implementation of the LUNA criterion is direct and simple. To the best of our knowledge, this is the first EQA method that explicitly takes into account the preservation of local included angles. The effectiveness of the LUNA criterion is experimentally supported by its outstanding performance on a series of benchmark datasets. [ABSTRACT FROM AUTHOR]

Many models are used to solve classification problems in machine learning. The classification restricted Boltzmann machine (ClassRBM) is a type of self-contained network model that is widely used in various classification applications. To implement classification, the ClassRBM updates the model parameters constantly during the training phase in terms of their class labels so that the model parameters learned from the ClassRBM are different from those learned from the conventional restricted Boltzmann machine (RBM), which is trained by unsupervised learning. In this paper, we demonstrate that the reconstruction errors of the ClassRBM are larger than those of the conventional RBM because of the label information. We then propose a classification model of the restricted Boltzmann machine based on these reconstruction errors. The reconstruction errors are used to train the proposed model to improve the classification performance of the ClassRBM. Extensive experiments are carried out to verify the proposed model. The experimental results demonstrate that the proposed model can improve the classification performance of the ClassRBM. [ABSTRACT FROM AUTHOR]

This paper presents an angle and density-based data preprocessing method. It can be used to simultaneously identify outliers and boundary points (called uniformly boundary points). Detecting boundary points is often more interesting than detecting normal points, since they represent valid, interesting, and potentially valuable patterns. An efficient local geometry-based method is proposed for detecting such points by both angle and density measures. The unified measure is adaptive and stable by combining multiple features (angles and density), which can be used to evaluate to what degree a given point is a boundary point. Compared with two related state-of-the-art approaches, our method better reflects the characteristics of the data and provides similar but accuracies for more data set. Experimental results obtained for a number of synthetic and real-world data sets demonstrate the effectiveness and efficiency of our method. [ABSTRACT FROM AUTHOR]

Restricted Boltzmann machines (RBMs) have been successfully applied in unsupervised learning and image density-based modeling. The aim of the pre-training step for RBMs is to discover an unknown stationary distribution based on the sample data that has the lowest energy. However, conventional RBM pre-training is sensitive to the initial weights and bias. The selection of initial values in RBM pre-training will directly affect the capabilities and efficiency of the learning process. This paper uses principal component analysis to capture the principal component directions of the training data. A set of initial parameter values for the RBM can be obtained by computing the same reconstruction of the data. Experiments on the Yale and MNIST datasets show that the proposed method not only retains a strong learning ability, but also significantly accelerates the learning speed. [ABSTRACT FROM AUTHOR]

In the field of cranio-maxillofacial (CMF) surgery, surgical simulation is becoming a very powerful tool to plan surgery and simulate surgical results before actually performing a CMF surgical procedure. Reliable prediction of facial soft tissue changes is in particular essential for better preparation and to shorten the time taken for the operation. This paper presents a surgical simulation system to predict facial soft tissue changes caused by the movement of bone segments during CMF surgery. Two experiments were designed to test the feasibility of this simulation system. The test results demonstrate the feasibility of fast and good prediction of post-operative facial appearance, with texture. Our surgical simulation system is applicable to computer-assisted CMF surgery. [ABSTRACT FROM AUTHOR]

A crucial step in many manifold clustering and embedding algorithms involves selecting the neighbors of a data point. In this paper, a non-negative representation learning algorithm is proposed to select the neighbors of a data point, so that they are almost always lying in the tangent space of the manifold at the given query point. This is very important to multi-manifold clustering and manifold embedding. Convex geometry theory states that a data point on a face of a convex polytope can be a convex combination of the vertices of the polytope, and that the non-zero combination scalars correspond to the vertices of the face the point lies on. The intent of this paper is to use points that that are near to a point as the vertices of a convex polytope, and select the vertices on the same face as this point to be the neighbors. Clearly, the point can be a convex combination of its neighbors. We can observe that these neighbors almost always lie in the tangent space of the manifold at the query point and can preserve the local manifold structure well. Based on this basic idea, we construct an objective function. Moreover, an augmented Lagrange multipliers method is proposed for solving it to derive a non-negative representation, and then the neighbors of a data point can be obtained. The result of our neighbor selection algorithm can be used by other clustering methods such as LEM or manifold embedding methods such as LLE. We demonstrate the effectiveness and efficiency of the proposed method through experiments on synthetic data and a real-world problem. [ABSTRACT FROM AUTHOR]