Search

Multi-modal Contrastive Representation (MCR) learning aims to encode different modalities into a semantically aligned shared space. This paradigm shows remarkable generalization ability on numerous downstream tasks across various modalities. However, the reliance on massive high-quality data pairs limits its further development on more modalities. This paper proposes a novel training-efficient method for learning MCR without paired data called Connecting Multi-modal Contrastive Representations (C-MCR). Specifically, given two existing MCRs pre-trained on (A, B) and (B, C) modality pairs, we project them to a new space and use the data from the overlapping modality B to aligning the two MCRs in the new space. Meanwhile, since the modality pairs (A, B) and (B, C) are already aligned within each MCR, the connection learned by overlapping modality can also be transferred to non-overlapping modality pair (A, C). To unleash the potential of C-MCR, we further introduce a semantic-enhanced inter- and intra-MCR connection method. We first enhance the semantic consistency and completion of embeddings across different modalities for more robust alignment. Then we utilize the inter-MCR alignment to establish the connection, and employ the intra-MCR alignment to better maintain the connection for inputs from non-overlapping modalities. We take the field of audio-visual contrastive learning as an example to demonstrate the effectiveness of C-MCR. We connect pre-trained CLIP and CLAP models via texts to derive audio-visual contrastive representations. Remarkably, without using any paired audio-visual data and further tuning, C-MCR achieves state-of-the-art performance on six datasets across three audio-visual downstream tasks., Comment: Demos are available at \url{https://c-mcr.github.io/C-MCR/}

Immunohistochemistry (IHC) is a vital tool to distinguish tumor metastases from primary lesions in addition to morphologic analysis. In this study, a 64-year-old female with a past surgical history of lung adenocarcinoma 11 years ago was presented with recurrence of liver nodular lesions after multiple surgical procedures, including the Whipple procedure for pancreatic head adenocarcinoma and cytoreductive surgery for liver metastasis. Liver biopsy and review of the previous specimens, based on IHC analyses, suggested heterochronous metastases of lung adenocarcinoma to the digestive systems in a long-time span, instead of primary pancreatic adenocarcinoma. This case demonstrates the potential for misdiagnoses from morphologic analysis alone and suggests the necessity of IHC analyses to avoid misjudgment on tumor phenotypes, when a previous oncologic history is presented.

Automatic diagnosis systems have attracted increasing attention in recent years. The reinforcement learning (RL) is an attractive technique for building an automatic diagnosis system due to its advantages for handling sequential decision making problem. However, the RL method still cannot achieve good enough prediction accuracy. In this paper, we propose a Generative Adversarial regularized Mutual information Policy gradient framework (GAMP) for automatic diagnosis which aims to make a diagnosis rapidly and accurately. We first propose a new policy gradient framework based on the Generative Adversarial Network (GAN) to optimize the RL model for automatic diagnosis. In our framework, we take the generator of GAN as a policy network, and also use the discriminator of GAN as a part of the reward function. This generative adversarial regularized policy gradient framework can try to avoid generating randomized trials of symptom inquires deviated from the common diagnosis paradigm. In addition, we add mutual information to enhance the reward function to encourage the model to select the most discriminative symptoms to make a diagnosis. Experiment evaluations on two public datasets show that our method beats the state-of-art methods, not only can achieve higher diagnosis accuracy, but also can use a smaller number of inquires to make diagnosis decision.

Medical entity relation verification is a crucial step to build a practical and enterprise medical knowledge graph (MKG) because high-precision medical entity relation is a key requirement for many MKG-based applications. Existing relation verification approaches for general knowledge graphs are not designed for considering medical domain knowledge, although it is central to achieve high-quality entity relation verification for MKG. To this end, in this paper, we introduce a system for medical entity relation verification with large-scale machine reading comprehension. The proposed system is tailored to overcome the unique challenges of medical relation verification including high variants of medical terms, the high difficulty of evidence searching in complex medical documents, and the lack of evidence labels for supervision. To deal with the problem of variants of medical terms, we introduce a synonym-aware retrieve model to retrieve the potential evidence implicitly verifying the given claim. To better utilize the medical domain knowledge, a relation-aware evidence detector and a medical ontology-enhanced aggregator are developed to improve the performance of the relation verification module. Moreover, to overcome the challenge of providing high-quality evidence due to the lack of labels, we introduce an interactive collaborative-training method to iteratively improve the evidence accuracy. Finally, we conduct extensive experiments to demonstrate that the performance of our proposed system is superior to all comparable models. We also demonstrate that our system can significantly reduce the annotation time by medical experts in real-world verification tasks. It can help to improve the efficiency by nearly 300%. In particular, our system has been embedded into the Baidu Clinical Decision Support System.

Text-based diagnosis classification is a critical problem in AI-enabled healthcare studies, which assists clinicians in making correct decision and lowering the rate of diagnostic errors. Previous studies follow the routine of sequence based deep learning models in NLP literature to deal with clinical notes. However, recent studies find that structural information is important in clinical contents that greatly impacts the predictions. In this paper, a novel sequence-to-subgraph framework is introduced to process clinical texts for classification, which changes the paradigm of managing texts. Moreover, a new classification model under the framework is proposed that incorporates subgraph convolutional network and hierarchical diagnostic attentive network to extract the layered structural features of clinical texts. The evaluation conducted on both the real-world English and Chinese datasets shows that the proposed method outperforms the state-of-the-art deep learning based diagnosis classification models.

The zones where tunnels pass through faults are considered to be severely damaged during earthquakes. Site investigation of tunnels crossing faults revealed that there were different types of damage patterns after earthquakes. This article proposed a new seismic design concept of “guiding” and “yielding.” Two types of joints, multilevel brittle-flexible joint and flexible joint, were based on this concept and were investigated in this study. A series of shaking table tests were conducted on reduced scale tunnel models under two steps of the loading process, fault movement and subsequent seismic excitation with increasing intensities. The results showed that both types of joints clearly reduced the seismic responses of the tunnel lining. The idealized behavior—step-like deformation—appeared along the longitudinal direction of the tunnel in the two tests. No shear failure of the tunnel linings was found, and the longitudinal cracks in the crown, arch springing, and invert were common in the tests. The multilevel brittle-flexible joint was verified to be more appropriate for tunnels subject to massive earthquakes.

The lesion detection on medical images has been a challenging problem in computer vision due to the lack of high-quality annotations at the bounding box level and the difficulty to identify the lesion on gray-scale medical images which are commonly used in radiology tests like X-ray, CT and MRI. In this paper, we propose a novel framework of medical image enhancement based on the class-aware attention weight extraction and the deep colorization of medical images with generative adversarial networks motivated by human visual characteristics and cell staining. The evaluation conducted on the real public medical image datasets proves that, the performance of lesion detection based on the existing detectors is improved after enhancing the medical images with the proposed enhancement framework.

Importance . The last decade has witnessed the advances of cognitive computing technologies that learn at scale and reason with purpose in medicine studies. From the diagnosis of diseases till the generation of treatment plans, cognitive computing encompasses both data-driven and knowledge-driven machine intelligence to assist health care roles in clinical decision-making. This review provides a comprehensive perspective from both research and industrial efforts on cognitive computing-based CDSS over the last decade. Highlights . (1) A holistic review of both research papers and industrial practice about cognitive computing-based CDSS is conducted to identify the necessity and the characteristics as well as the general framework of constructing the system. (2) Several of the typical applications of cognitive computing-based CDSS as well as the existing systems in real medical practice are introduced in detail under the general framework. (3) The limitations of the current cognitive computing-based CDSS is discussed that sheds light on the future work in this direction. Conclusion . Different from medical content providers, cognitive computing-based CDSS provides probabilistic clinical decision support by automatically learning and inferencing from medical big data. The characteristics of managing multimodal data and computerizing medical knowledge distinguish cognitive computing-based CDSS from other categories. Given the current status of primary health care like high diagnostic error rate and shortage of medical resources, it is time to introduce cognitive computing-based CDSS to the medical community which is supposed to be more open-minded and embrace the convenience and low cost but high efficiency brought by cognitive computing-based CDSS.

The automatic diagnosis has been suffering from the problem of inadequate reliable corpus to train a trustworthy predictive model. Besides, most of the previous deep learning based diagnosis models adopt the sequence learning techniques (CNN or RNN), which is difficult to extract the complex structural information, e.g. graph structure, between the critical medical entities. In this paper, we propose to build the diagnosis model based on the high-standard EMR documents from real hospitals to improve the accuracy and the credibility of the resulting model. Meanwhile, we introduce the Graph Convolutional Network into the model that alleviates the sparse feature problem and facilitates the extraction of structural information for diagnosis. Moreover, we propose the mutual attentive network to enhance the representation of inputs towards the better model performance. The evaluation conducted on the real EMR documents demonstrates that the proposed model is more accurate compared to the previous sequence learning based diagnosis models. The proposed model has been integrated into the information systems in over hundreds of primary health care facilities in China to assist physicians in the diagnostic process.

This study aims to evaluate the epidemiological characteristics of mother-to-child transmission (MTCT) of HIV and identify the possible factors leading to infant HIV infection using a retrospective cohort study of early infant diagnosis (EID). Information on a total of 3,145 exposed infant-mother pairs was collected from the EID platform from July 2014 to December 2019. The MTCT rate was 2.1%. Spatial-temporal maps showed that rates varied by year and by region, with four districts (Baise, Guigang, Guilin, and Hechi) maintaining rates of2.0% in 2019. The rate of antiretroviral therapy (ART) use was 94.4%, with a gradual increase in prescriptions of highly active ART (HAART) from 83.0% in 2014 to 92.4% in 2019. A majority of 99.5% of infants were receiving artificial feeding. Factors associated with MTCT were ART use (odds ratio [OR] = 0.065, confidence interval [95% CI] = 0.035-0.121) and artificial feeding (OR = 0.091, 95% CI = 0.018-0.452). HAART was more helpful in decreasing the risk of MTCT compared with monotherapy (OR = 0.115, 95% CI = 0.014-0.933). ART during the postpartum period correlated with an increased risk (OR = 11.579, 95% CI = 1.402-95.960) compared with use of ART during pregnancy. This study indicates that MTCT rate of HIV is decreasing meaningfully in Guangxi. Some areas still face challenges in elimination of MTCT and need further resources and interventions. Future program planning should take into consideration the fact that ART use-in particular the use of HAART or ART during pregnancy-and replacement feeding may contribute to the prevention of MTCT.

The automatic text-based diagnosis remains a challenging task for clinical use because it requires appropriate balance between accuracy and interpretability. In this paper, we attempt to propose a solution by introducing a novel framework that stacks Bayesian Network Ensembles on top of Entity-Aware Convolutional Neural Networks (CNN) towards building an accurate yet interpretable diagnosis system. The proposed framework takes advantage of the high accuracy and generality of deep neural networks as well as the interpretability of Bayesian Networks, which is critical for AI-empowered healthcare. The evaluation conducted on the real Electronic Medical Record (EMR) documents from hospitals and annotated by professional doctors proves that, the proposed framework outperforms the previous automatic diagnosis methods in accuracy performance and the diagnosis explanation of the framework is reasonable.

Spaceborne synthetic aperture radar (SAR) polarimetric application, operating at P-band, is significantly influenced by ionospheric Faraday rotation (FR). Various researches have been published to evaluate the FR effect on polarimetric scattering matrix, nevertheless all of which are based on the narrow-band hypothesis and ignore the polarimetric dispersion (PD) effect caused by FR for its high disturbance within the bandwidth. The existent model is no longer well applicable to the future P-band ultra-wideband system with benefits of the high-resolution detection of the subsurface or obscured targets. A novel analytical model of PD based on the frequency-domain range pulse compression is proposed and analysed in this Letter, and the simulation results indicate the spaceborne P-band ultra-wideband SAR system is seriously influenced by the PD effect.

The zones where tunnels through the active faults are proved to be seriously damaged because of the fault dislocation which could be due to strong earthquakes. A steel reinforced rubber joint with flexibility and ductility is proposed in this paper. An experimental investigation of the responses and the damage patterns of the tunnel lining with the joints under normal faulting is presented. The results showed the damage of the lining with the joints in the model test was mainly concentrated in the fault areas because of the joints’ adjustment. No obvious shear failure was found in this test. Then, a nonlinear three-dimensional finite element (FE) model, using concrete damaged plasticity model, was performed to simulate the tunnel linings with the proposed joints and without joints under normal faulting. The numerical results show that the tunnel with the joints had step-like displacements without big relative displacement compared to the tunnel without joints. The joints efficiently decreased the maximum relative displacement between lining segments. The extent of the tensile damage area of the lining affected by normal fault was longer than that of the compressive damage area. And the tensile damage area was larger than the compressive damage area for either the lining with the joints or the lining without joints. There was no obvious difference in damage to the lining with and without joints in areas under compression loading. The joints efficiently decreased the extent of the damage in the zone with tensile stresses. The comparisons between the numerical predictions and experimental results show a good agreement.

Amplitude and phase statistics of SAR complex interferogram are significant in the study of interferometry and polarimetry. To reduce statistical variations, multi-look processing is adopted by averaging spatially the complex interferogram. In this study, we derive and validate three kinds of probability density functions (PDFs) of multi-look interferogram for different surface feature scenes. For simple homogeneous areas with the gamma distribution intensity, a concise product-form interferometry phase PDF is derived, which is equivalent to a conventional Gauss hypergeometric PDF. For complicated areas with the K and G0 distributions intensity, two new interferometry amplitude PDFs named as Gamma-K and Gamma-G are proposed, and their phase PDFs are approximately preserved. Finally three typical areas including grass, mountain, and city are picked out from a pair of RADARSAT-2 SAR images and studied. Experimental results indicate good agreement between the computed histograms and the theoretical distributions. The results obtained can be applied to the feature classification of polarisation SAR data and the estimation of decorrelation effect of interferometric SAR. Science Journal, Vol. 64, No. 6, November 2014, pp.564-570, DOI:http://dx.doi.org/10.14429/dsj.64.4747

For spaceborne Synthetic Aperture Radar(SAR) processing, Doppler property influences the azimuth performance and the imaging accuracy directly. In this paper, the Doppler centroid frequency expression based on elliptical orbit is derived analytically. The influences of the yaw steering and the pitch steering are analyzed respectively. Then, a new method called elliptic orbit total Zero Doppler steering is proposed, and a simulation for TerraSAR-X parameters shows its advantages. The simulation results indicate that this method reduces the Doppler centroid 100 times smaller than currently applied yaw steering methods, and 5 times smaller than the circular orbit total Zero Doppler steering, which illuminates the applicability of the method.

This article focuses on the relationships between Chinese and German enterprises in the CSR field, and the stakeholders involved in their cooperation are mentioned and described. The background of environmental regulation in both countries, as well as the differences in German and Chinese companies, is provided—plus the differences in the understanding of the concepts of CSR and sustainability. After describing the current situation of the stakeholders, the author comments the opportunities and challenges of the partnerships between companies of both countries, and gives suggestions for their future cooperation.

With the smart grid construction proceeding further, requirements increase on model and information sharing between control center and substation, and the current commonly used communication protocol between control center and substation has been unable to meet the demand of communication. This paper proposes a real-time service communication protocol, which is based on IEC61850, directly mapped the abstract communication service interface (ACSI) to the TCP/IP communication protocol stack, to build up a communication protocol in smart grid construction needs, meet the intelligent seamless communication requirements between control center and substation. Also the conversion model method and calibration method is also proposed.

This paper presents a new efficient phase unwrapping algorithm for interferometric synthetic aperture radar (InSAR) phase images based on nonlinear phase model. Firstly, based on the analysis of statistical properties and signal characteristics of the interferometric phase, the concept of nonlinear phase model is proposed. It uses the nonlinear phase model of reflecting the topographic outline to approximate the interferometric phase change appeared in the local window. The lower amplitude bound of principal vector is defined and its value acquire method be given, which can effectively and adaptively extract the nonlinear components of the phase. Secondly, the nonlinear phase model is introduced to phase unwrapping, which increased the reliability of integration path and the accuracy of phase gradient, and improved effectively the performance of phase unwrapping. Real data processing results demonstrate the validity of the proposed nonlinear phase model and its corresponding processing method.

The traditional way of distributed modeling is based on CIM/XML model file, uses model combination to generate uniform power grid model. The disadvantages of this way includes complex model combination process, low efficiency, and the real-time synchronization of power grid model among control centers cannot be achieved. Because CIM/XML-based model file is difficult to extend content, Model combination can only support physical topology model, and cannot achieve the comprehensive requirements of application services, such as real-time monitoring, dispatching planning, security checking and dispatching management. The content of model combination is not completed, so several necessary messages such as formula define, Signboard information cannot be described. In the paper, a new distributed modeling mechanism was proposed, and its mechanism is based on physical-fit storage of power grid model, can achieve the goal for maintaining power model at its source and sharing immediately by all control centres, and generate complete power grid model among control centres. The main working principle of this mechanism is to use the method of area-partition identification for model objects, all model objects are coded by area-partition, so every model object is uniqueness among all related control centres, then the goal for physical-fit storage of power grid model can be achieved. with this method, one control centre can independently maintain its own power grid model, and synchronization these model to other centres. So an unitive model management mode among control centres is not necessary. After power model is maintained at its source, this power model can be shared immediately by all control centers. Based on unique identification of model objects and physical-fit storage of power grid model, This new mechanism can achieve the goal of automatic model synchronization among control centers, with no need for model combination or manual intervention. The model information which needed to share can be custom made, so can achieve the goal for completeness, comprehensiveness and no need for extend. All of these are to achieve the goal to assure the flexibility, reliability and instantaneity for maintaining power model at its source and sharing by all control centers. physical-fit storage, area-partition identification, distributed modeling, sharing immediately.

The error in the estimation of the ground interferometric phase will reduce the accuracy of the inversion of the vegetation height in three-stage vegetation inversion method. Aiming at this problem, the new vegetation height inversion method based on the BP neural network is proposed. The new method directly fits the nonlinear mapping relationship between the complex correction coefficients and the vegetation height, so it reduces the height inversion error caused by the error in the estimated ground interferometric phase. The new method has better performance than the three-stage vegetation height inversion method, and the experiment results validate the superiority of the new method.

The background and existing problems about data exchanging among multi-systems in the control center are introduced. The best way of data exchanging among systems are setting up a common information data platform system (CIDPS) CIDPS, which is located in the II and III sections, ensures the data to be shared by the multi-subsystems in the control center. Two methods of realizing CIDPS are DATA-PROXY and DATA-ENTITY. After discussing, the conclusion is come to that the DATA-ENTITY CIDPS method is preferred for the dispatching automation systems. Standard IEC 61970 is the theoretical base of CIDPS and it is the interactive technology standard among CIDPS and other system. A distributed object technology is the technology guarantee of CIDPS. The power model of CIDPS is built according to CIM Standard. CIDPS offers CIS interfaces for other systems accessing data. The sub-system can get the various power system models, parameters or real-time data by GDA or HSDA interfaces, and for a graphics centric method, SVG (scalable vector graphics) is a good candidate for the exchange format. Lastly, an application of CIDPS named OPEN3000 is presented particularly.

GMD is a WYSWYG (What You See is What You Get) method for power system modeling that makes the graph, model and database unified; the traditional GMD is based on the special and private information model and models cannot be exchanged or merged between different sources. With the adoption of standard IEC 61970, GMD should be based on the common information model (CIM) to enable development the "plug-in" applications and the interoperable systems. This method is called CIM-based GMD. This paper presents the basic principle of GMD and compares the traditional GMD with CIM-based GMD at first, then discusses the key technologies of CIM-based GMD. Finally, the paper points out that CIM-based GMD had better use the realtime database with orient-object, relational and hierarchy characteristics, form much more information automatically as possible and validate models with artificial intelligence knowledge to make sure that applications process models correctly.

The distributed object technology and its application in the next generation of a power dispatching automation system are introduced. Standard IEC 61970 is to found common information model (CIM) for the power system objects and to form component interface specifications (CIS) for logical information exchanges among all applications. Standard IEC 61970 is the theoretical basis of data models and distributed object technology is the technology guarantee of implementing this system. Each application, each service or each sub-system can be described as one or more components. All components are interconnected by an integration framework, which is composed of object request broker (ORB) bus, namely the core of common object request broker architecture (CORBA). In this paper, the background of developing a next generation of a power dispatching automation system complying with Standard IEC 61970 is reviewed. On the basis of a lot of testing data about CORBA performance, the feasibility of adopting CORBA-based distributed object technology is discussed. Constructing integration framework of the system and encapsulating the existing applications or programs are two main methods about CORBA applications. Lastly, the whole system structure and its design scheme of the dispatching automation system are presented and analyzed.

An optical fiber industrial network is the key to data exchange between upper administration layer and lower control layer. A high-speed network can realize all kinds of data, such as control commands, field signals and multimedia signals, to be transferred together. Good availability is the basis of ensuring optical fiber industrial network normal running. By analyzing the multi-level single-attachment network availability, the conclusion is that a dual-level single-attachment network is the best topology of the optical fiber industrial network. Finally, the application of this network in a transformer substation is introduced.