Search

Detecting signet ring cells on histopathologic images is a critical computer-aided diagnostic task that is highly relevant to cancer grading and patients' survival rates. However, the cells are densely distributed and exhibit diverse and complex visual patterns in the image, together with the commonly observed incomplete annotation issue, posing a significant barrier to accurate detection. In this article, we propose to mitigate the detection difficulty from a model reinforcement point of view. Specifically, we devise a Classification Reinforcement Detection Network (CRDet). It is featured by adding a dedicated Classification Reinforcement Branch (CRB) on top of the architecture of Cascade RCNN. The proposed CRB consists of a context pooling module to perform a more robust feature representation by fully making use of context information, and a feature enhancement classifier to generate a superior feature by leveraging the deconvolution and attention mechanism. With the enhanced feature, the small-sized cell can be better characterized and CRDet enjoys a more accurate signet ring cell identification. We validate our proposal on a large-scale real clinical signet ring cell data set. It is shown that CRDet outperforms several popular convolutional neural network-based object detection models on this particular task.

Object detection is an essential computer vision task that possesses extensive application prospects in on-road applications. Copious novel methods have been proposed in this branch recently. However, the majority of them have high computational cost, making them intractable to be deployed on embedded devices. In this paper, taking YOLOv5s, the smallest model in the YOLOv5 family, as the baseline, we explore a bag of tricks that improve the detection performance for a specified on-road application, under the premise of ensuring that it does not increase the computational cost of YOLOv5s. Specifically, we introduce relevantly external data to deal with the problems of sample imbalance. Meanwhile, knowledge distillation is employed to transfer knowledge from a cumbersome model to a compact model, where a united distillation scheme is developed to enhance the effectiveness. In addition, a pseudo-label based training strategy is utilized to further learn from the biggest YOLOv5 model. We have applied the above tricks to the Embedded Deep Learning Object Detection Model Compression Competition for Traffic in Asian Countries held in conjunction with ICMR 2021. The experiments have shown that all the tricks are useful. Their combination have built an accurate and slim detection model. It is highly competitive and has been ranked 2nd place in the competition. We believe the tricks are also meaningful for building other application-oriented object detectors.

Automatic product image classification is a task of crucial importance with respect to the management of online retailers. Motivated by recent advancements of deep Convolutional Neural Networks (CNN) on image classification, in this work we revisit the problem in the context of product images with the existence of a predefined categorical hierarchy and attributes, aiming to leverage the hierarchy and attributes to improve classification accuracy. With these structure-aware clues, we argue that more advanced deep models could be developed beyond the flat one-versus-all classification performed by conventional CNNs. To this end, novel efforts of this work include a salient-sensitive CNN that gazes into the product foreground by inserting a dedicated spatial attention module; a multiclass regression-based refinement that is expected to predict more accurately by merging prediction scores from multiple preceding CNNs, each corresponding to a distinct classifier in the hierarchy; and a multitask deep learning architecture that effectively explores correlations among categories and attributes for categorical label prediction. Experimental results on nearly 1 million real-world product images basically validate the effectiveness of the proposed efforts individually and jointly, from which performance gains are observed.

As a well-established multidrug combinations schema, traditional Chinese medicine (herbal prescription) has been used for thousands of years in real-world clinical settings. This paper uses a complex network approach to investigate the regularities underlying multidrug combinations in herbal prescriptions. Using five collected large-scale real-world clinical herbal prescription datasets, we construct five weighted herbal combination networks with herb as nodes and herbal combinational use in herbal prescription as links. We found that the weight distribution of herbal combinations displays a clear power law, which means that most herb pairs were used in low frequency and some herb pairs were used in very high frequency. Furthermore, we found that it displays a clear linear negative correlation between the clustering coefficients and the degree of nodes in the herbal combination network (HCNet). This indicates that hierarchical properties exist in the HCNet. Finally, we investigate the molecular network interaction patterns between herb related target modules (i.e., subnetworks) in herbal prescriptions using a network-based approach and further explore the correlation between the distribution of herb combinations and prescriptions. We found that the more the hierarchical prescription, the better the corresponding effect. The results also reflected a well-recognized principle called “Jun-Chen-Zuo-Shi” in TCM formula theories. This also gives references for multidrug combination development in the field of network pharmacology and provides the guideline for the clinical use of combination therapy for chronic diseases.

Image-to-image translation is an important and challenging problem in computer vision. It aims to learn the mapping between two different domains, with applications ranging from data augmentation, style transfer, to super-resolution, etc. With the success of deep learning methods in visual generative tasks, researchers have applied deep generative models, especially generative adversarial networks (GANs), to image-to-image translation since the year of 2016 and gained fruitful progress. In this survey, we have conducted a comprehensive review of the literature in this field, covering supervised and unsupervised methods, among which unsupervised approaches include one-to-one, one-to-many, many-to-many categories and some latest theories. We highlight the innovation aspect of these methods and analyze different models employed and their components. Besides, we summarized some commonly used normalization techniques and evaluation metrics, and finally, present several challenges and future research directions in this area.

Low-cost live broadcasting is highly anticipated for soccer matches that involve heavy expenses in human and equipment currently. This demo showcases the autoSoccer system to approach the target. It takes a fix-view panoramic soccer video as the input and automatically generates its live broadcasting, which continuously focuses on the most interesting area in the soccer field. To this end, a novel pipeline is developed by leveraging both low-level motion and visual features, and high-level soccer-related semantics. By appropriately utilizing these clues, autoSoccer is capable of delivering a soccer match video analogous to human directed. Demo on school soccer show that autoSoccer produces videos with satisfactory watching experience. Moreover, it executes in real-time on a PC with Intel i7 CPU and one Nvidia Titan XP GPU.

As one of the most active research fields in computer vision, the performance of object detection has been boosted by recent advancement of deep learning. However, performance improvements are often accompany with more resource consumption, which in turn restricts the application. There has been a rising interest in building object detectors that run well on embedded systems, i.e., with a better tradeoff among accuracy, speed, resource consumed, etc. To this end, we analyze popular object detectors, with an emphasis on taking typical evaluation metrics from the above aspects to assess both the selection of backbone network and the effects of input size. With these empirical studies, we propose to integrate MobileNet series into SSD to generate efficient yet accurate models and discuss their training tricks. The models are verified on the embedded object detection model competition held in conjunction with ICME 2020, where we gain the best accuracy on scooter detection.

To maintain sustainable economic growth, China has created a national innovation system (NIS) and strengthened the central status of firms. Our data show that the effect of turnover growth in small and medium-sized enterprises (SMEs) on China&rsquo, s aggregate Gross Domestic Product (GDP)growth is significant, but the status of SMEs in the NIS and related policies is not significant. To determine whether there is a correspondence between the sustainability of innovation in SMEs and the support of China&rsquo, s NIS, we developed a framework for China&rsquo, s innovation policy under the NIS framework, taking into account its transition characteristics, to examine the texts of SME innovation policies and reveal the sustainability of SMEs&rsquo, innovation. The relevant national government policy texts were collected from the yearbooks of Chinese SMEs between 1999 and 2017 and government notices between 1994 and 2017. On this basis, we also compared with some other countries&rsquo, innovation systems. The findings indicate that China&rsquo, s NIS pays little attention to the sustainability of SMEs&rsquo, innovation activities for two reasons. First, the scope of the NIS is very narrowly defined. Second, the top-down, government-oriented Research and Development (R&amp, D) system that focuses on large state-owned firms leaves little room for innovation policies in SMEs.

Identifying signet ring cells on pathological images is an important clinical task that highly relevant to cancer grading and prognosis. However, it is challenging as the cells exhibit diverse visual appearance in the crowded cellular image. This task is also less studied by computational methods so far. This paper proposes a Classification Reinforcement Detection Network (CRDet) to alleviate the detection difficulties. CRDet is composed of a Cascade RCNN architecture and a dedicated devised Classification Reinforcement Branch (CRB), which consists of a dedicated context pool module and a corresponding feature enhancement classifier, aiming at extracting more comprehensive and discriminative features from the cell and its surrounding context. With the reinforced features, the small-sized cell can be well characterized, thus a better classification is expected. Experiments on a public signet ring cell dataset demonstrate the proposed CRDet achieves a better performance compared with popular CNN-based object detection models.

Semi-supervised community detection has gained a lot of attention by leveraging side information for better understanding network topologies. However, most of existing works select side information in a random manner. They usually require a great amount of side information to significantly improve the performance of community detection. Besides, they have to define the number of communities in advance. To address these issues, this paper proposed an active semi-supervised community detection method, called SK-rank-D. The key advantages of this framework are twofold: (1) Actively selecting a small amount of links as side information to“sharpen” the boundaries between communities and “compact” the connections within communities; (2) Automatically referring the number of communities by selecting informative nodes in communities. Empirical analysis on both synthetic and real-world networks showed the effectiveness and rationality of the newly proposed method in deciding community number. We also compared with competing semi-supervised community detection methods, the experimental results demonstrated the superior performance of our approach.

A new concept decomposition method WordCom is proposed.It creates concept vectors by identifying semantic word communities from a weighted word co-occurrence network.It is not only robust to the sparsity of short texts but also overcomes the curse of dimensionality.It scaling to a large number of short text inputs due to the concept vectors being obtained from term-term space.Experimental tests have shown that the proposed method outperforms state-of-the-art algorithms. Short text clustering is an increasingly important methodology but faces the challenges of sparsity and high-dimensionality of text data. Previous concept decomposition methods have obtained concept vectors via the centroids of clusters using k-means-type clustering algorithms on normal, full texts. In this study, we propose a new concept decomposition method that creates concept vectors by identifying semantic word communities from a weighted word co-occurrence network extracted from a short text corpus or a subset thereof. The cluster memberships of short texts are then estimated by mapping the original short texts to the learned semantic concept vectors. The proposed method is not only robust to the sparsity of short text corpora but also overcomes the curse of dimensionality, scaling to a large number of short text inputs due to the concept vectors being obtained from term-term instead of document-term space. Experimental tests have shown that the proposed method outperforms state-of-the-art algorithms.

Underwater coral reef fish detection is topic receiving increasingly attention due to its importance in various applications like fish biodiversity monitoring, marine resource managements, etc. However, compared with studies on generic object detection, existing methods on this task are not mature so far where advanced deep models and technologies are seldom considered. This paper presents FFDet, a fully convolutional network for coral reef fish detection by layer fusion. FFDet consists of a single shot multibox detector (SSD)-based backbone, but different with SSD, it devises a novel feature fusion module to aggregate adjacent prediction layers for enhanced feature representation. Thus, instead of using the prediction layers one-by-one, the enhanced features each combining information from multiple layers, are leveraged to detect fishes at different scales. We argue that the proposed module is capable of encoding both strong semantics and detail context information. Experimental results on SeaCLEF dataset show that FFDet not only outperforms SSD in performance by sacrificing only a little efficiency, but also better than another two popular end-to-end deep models in both detection performance and speed, especially on detecting large-sized fishes.

Large-scale product classification is an essential technique for better product understanding. It can provide support to online retailers from a number of aspects. This paper discusses the CNN based product classification with the existence of class hierarchy. A SaCNN-MCR method is developed to settle this task. It decomposes the classification into two stages. Firstly, a spatial attention based CNN model that directly classifies a product to leaf classes is proposed. Compared with traditional CNNs, the proposed model focuses more on product region rather than the whole image. Secondly, the outputted CNN score together with class hierarchy clues are jointly optimized by employing a multi-class regression (MCR) based refinement, which provides another kind of data fitting that further benefits the classification. Experiments on nearly one million real-world product images show that, based on the two innovations, SaCNN-MCR steadily improves the classification performance over CNN models without these modules. Moreover, it is demonstrated that CNN features characterize product images much better than traditional features, whose classification performance outperforms those of the traditional features by a large margin.

Community detection is a very important problem in social network analysis. Classical clustering approach, K -means, has been shown to be very efficient to detect communities in networks. However, K -means is quite sensitive to the initial centroids or seeds, especially when it is used to detect communities. To solve this problem, in this study, we propose an efficient algorithm K -rank, which selects the top- K nodes with the highest rank centrality as the initial seeds, and updates these seeds by using an iterative technique like K -means. Then we extend K -rank to partition directed, weighted networks, and to detect overlapping communities. The empirical study on synthetic and real networks show that K -rank is robust and better than the state-of-the-art algorithms including K -means, BGLL, LPA, infomap and OSLOM.

Community structure is one of the most important properties for understanding the topology and function of a complex network. Recently, the rank reduction technique, non-negative matrix factorization (NMF), has been successfully used to uncover communities in complex networks. In the machine learning literature, the algorithm Alternating Constraint Least Squares (ACLS) is developed to perform NMF with sparsity constraint for clustering data and showed good performance, but it is not used in detecting communities in networks. In this study, we first test the ACLS algorithm on several synthetic and real networks to show its performance on community detection. Then we extend ACLS to orthogonal nonnegative matrix factorization, propose ALSOC, in which orthogonality constraint is added into NMF to discovery communities. The experimental results show that NMF with orthogonality constraint is able to improve the performance of community detection, meanwhile it has ability to maintain the sparsity of matrix factors.

Background. Symptoms and signs (symptoms in brief) are the essential clinical manifestations for individualized diagnosis and treatment in traditional Chinese medicine (TCM). To gain insights into the molecular mechanism of symptoms, we develop a computational approach to identify the candidate genes of symptoms.Methods. This paper presents a network-based approach for the integrated analysis of multiple phenotype-genotype data sources and the prediction of the prioritizing genes for the associated symptoms. The method first calculates the similarities between symptoms and diseases based on the symptom-disease relationships retrieved from the PubMed bibliographic database. Then the disease-gene associations and protein-protein interactions are utilized to construct a phenotype-genotype network. The PRINCE algorithm is finally used to rank the potential genes for the associated symptoms.Results. The proposed method gets reliable gene rank list with AUC (area under curve) 0.616 in classification. Some novel genes like CALCA, ESR1, and MTHFR were predicted to be associated with headache symptoms, which are not recorded in the benchmark data set, but have been reported in recent published literatures.Conclusions. Our study demonstrated that by integrating phenotype-genotype relationships into a complex network framework it provides an effective approach to identify candidate genes of symptoms.

Latent community discovery that combines links and contents of a text-associated network has drawn more attention with the advance of social media. Most of the previous studies aim at detecting densely connected communities and are not able to identify general structures, e.g., bipartite structure. Several variants based on the stochastic block model are more flexible for exploring general structures by introducing link probabilities between communities. However, these variants cannot identify the degree distributions of real networks due to a lack of modeling of the differences among nodes, and they are not suitable for discovering communities in text-associated networks because they ignore the contents of nodes. In this paper, we propose a popularity-productivity stochastic block (PPSB) model by introducing two random variables, popularity and productivity, to model the differences among nodes in receiving links and producing links, respectively. This model has the flexibility of existing stochastic block models in discovering general community structures and inherits the richness of previous models that also exploit popularity and productivity in modeling the real scale-free networks with power law degree distributions. To incorporate the contents in text-associated networks, we propose a combined model which combines the PPSB model with a discriminative model that models the community memberships of nodes by their contents. We then develop expectation-maximization (EM) algorithms to infer the parameters in the two models. Experiments on synthetic and real networks have demonstrated that the proposed models can yield better performances than previous models, especially on networks with general structures.

Network modeling and analysis have been developed as one of the promising approaches for exploring the regularities behind the phenomena of complex organization and interactions in many significant fields. Traditional Chinese medicine (TCM) is a kind of holistic medical science, usually in whose clinical setting herb prescriptions consisting of several distinct herbs were used for individualized patients to get the maximum effectiveness. Detecting the significant herb interactions with effectiveness for some specific disease conditions is an important issue for both TCM clinical treatment and novel drug development. By modeling herb prescriptions as herb interaction network, in this paper, we propose a network comparison method based on multiscale backbone algorithm (msbNC) to discover the herbal interactions from one herb network that differ significantly with respect to a referenced herb network according to a null model. This method could easily be used to find the significant effective herbal interactions while incorporating appropriate outcome variables to construct coupled herb networks (one network is constructed from herb prescriptions with good outcome, while another one is from herb prescriptions with bad outcome). Using two herb prescription data sets from the outpatient cases of highly-experienced TCM physicians for insomnia treatment, we applied msbNC method to detect significant herbal interactions in the herb prescriptions of two TCM physicians and two distinct outcomes. The results showed that msbNC could distinguish clinically meaningful herbal interactions from these data sets. Therefore, the proposed method: msbNC coupled with network modeling could be used as a promising approach for effective herb interactions discovery from large-scale clinical data.

Relational Markov networks (RMNs) are a joint probabilistic model for an entire collection of related entities. The model is able to mine relational data effectively by integrating information from content attributes of individual entities as well as the links among them, yet the prediction accuracy is greatly affected by the definition of the relational clique templates. Maximum likelihood estimation (MLE) is used to estimate the model parameters, but this can be quite costly because multiple rounds of approximate inference are required over the entire dataset. In this paper, we propose constructing RMNs basing on the community structures of complex networks, and present a discriminative maximum pseudolikelihood estimation (DMPLE) approach for training RMNs. Experiments on the collective classification and link prediction tasks on some real-world datasets show that our approaches perform well in terms of accuracy and efficiency.

Community structure is one of the most important features of complex networks, it uncovers the internal organization of the nodes. Affinity propagation (AP) is a recent proposed powerful cluster algorithm as it costs much less time and reaches much lower error. But it was shown that AP displayed severe convergence problems for identifying communities on the majority of unweighted protein-protein interaction (PPI) networks. On the contrary, AP was shown to achieve great success for identifying communities in benchmark artificial and social networks. So, in this study, we use AP to identify communities on artificial, social and unweighted PPI networks for finding the problem of the conflict. And we compare AP with Markov cluster (MCL), which was shown to outperform a number of clustering algorithms for PPI networks. The experimental results have shown that AP performs well without oscillations when similarity matrixes are chosen properly, and MCL is more accurate than AP but it runs slower than AP in large scale networks.

K-means clustering is widely used due to its fast convergence, but it is sensitive to the initial condition.Therefore, many methods of initializing K-means clustering have been proposed in the literatures. Compared with Kmeans clustering, a novel clustering algorithm called affinity propagation (AP clustering) has been developed by Frey and Dueck, which can produce a good set of cluster exemplars with fast speed. Taking the convergence property of K-means and the good performance of affinity propagation, we presented a new clustering strategy which can produce much lower squared error than AP and standard K-means: initializing K-means clustering using cluster exemplars produced by AP. Numerical experiments indicated that such combined method outperforms not only AP and original K-means clustering, but also Kmeans clustering with sophisticated initial conditions designed by various methods.