- Data-anonymous Encoding for Text-to -SQL Generation
papers link: https://www.microsoft.com/en-us/research/publication/data-anonymous-encoding-for-text-to-sql-generation/
across Research is an important issue Text-to-SQL column name field is to identify natural language statements mentioned in the table, and the value of the cell. This paper proposes a framework based on intermediate variables and multi-task learning, while trying to solve the table entity recognition and semantic parsing problem, and achieved good results. Paper presented at the EMNLP 2019 meeting.
-
Towards Complex Text-to-SQL in Cross-domain Database
papers link: https://www.microsoft.com/en-us/research/publication/towards-complex-text-to-sql-in-cross-domain-database -with-intermediate-representation /
computer executable language (such as SQL statements and stored closely related structure) there is a mismatch problem with natural language, semantic parser to complex problems makes it difficult. To solve this problem, the paper designed an intermediate language. First natural language into an intermediate language, and then converted into an intermediate language SQL, you can improve the accuracy of the semantic parser. The paper was published in ACL 2019 meeting. -
Leveraging Adjective-Noun Phrasing Knowledge for Comparison Relation Prediction in Text-to-SQL
papers link: https://www.microsoft.com/en-us/research/publication/leveraging-adjective-noun-phrasing-knowledge-for-comparison -relation-prediction-in-text-
to-sql / natural language understanding, application of knowledge is very important. In this paper, Adjective-Noun Phrasing Knowledge as a starting point to try to use language knowledge in Text-to-SQL to improve the accuracy of the language understanding. Paper presented at the EMNLP 2019 meeting. -
FANDA: A Novel Approach to Perform Follow -up Query Analysis
Papers link: https://www.microsoft.com/en-us/research/publication/fanda-a-novel-approach-to-perform-follow-up-query -analysis /
multi-wheeled dialog, the dialog statements often refers to the presence or omitted, the need to understand the context of the current sentence. This article analyzes and summarizes the common omission occurs in conversational data analysis or to refer to the phenomenon, and made a full complement of the current statement method. Paper presented at the AAAI 2019. -
A Split-and-Recombine Approach for Follow-up Query Analysis
Papers link: https://www.microsoft.com/en-us/research/publication/a-split-and-recombine-approach-for-follow-up- query-analysis /
herein proposes a split-recombine frame processing context, the statement can be used to effectively treatment session context omitted or present problems frequently refer. This framework can be used both for the current statement full complement (restate), can also be generated directly logic form (such as SQL). Published in EMNLP 2019. -
QuickInsights: Quick and Automatic Discovery of Insights from Multi-Dimensional Data
Papers link: https://www.microsoft.com/en-us/research/uploads/prod/2019/05/QuickInsights-camera-ready-compliant.pdf
the thesis put forward innovative abstract definition of multidimensional data insights (insights) a generally applicable, and systematically put forward an effective insight for large-scale multi-dimensional data mining algorithms. Published in SIGMOD 2019. Appropriate technologies from 2015 to Microsoft Power BI, Office 365 and other products. -
TableSense: Spreadsheet Table Detection with Convolutional Neural Networks
Thesis link: https://www.microsoft.com/en-us/research/uploads/prod/2019/01/TableSense_AAAI19.pdf
article proposed based on the depth learning model TableSense technology, and a detection region can be appreciated that the spreadsheet table structure, and converts it to an automatic analysis of multidimensional data structure. This technology has been transformed into Microsoft's Office 365 products, with the Ideas in Excel full-featured on-line. Published in AAAI 2019. -
Text-to-Viz: Automatic Generation of Infographics Proportion-Related Natural language Statements from
paper links: https://www.microsoft.com/en-us/research/publication/text-to-viz-automatic-generation-of- infographics-from-proportion-related-
natural-language-statements / the paper was published in the IEEE VIS 2019, pioneered the technology generated automatically by the natural language data-information (Infographics) a. This technology makes it possible to design very easy to get a lot of data map, the data used to enhance the expression of the story. -
DataShot: Automatic Generation of Fact Sheets from Tabular Data
Papers link: https://www.microsoft.com/en-us/research/publication/datashot-automatic-generation-of-fact-sheets-from-tabular-data/
the paper presented at the IEEE VIS 2019, we propose a technique to automatically generate a table of data from a combination of data from a plurality of data poster map made of. -
Towards Automated Infographic Design: Deep Learning- based Auto-Extraction of Extensible Timeline
paper links: https://www.microsoft.com/en-us/research/publication/towards-automated-infographic-design-deep-learning-based- auto-extraction-of-extensible-
timeline / this paper proposes a technique for automatically extracting the template data from the picture of FIG. Using computer vision, image time axis into a plurality of visual design elements and re-combination, so that the visual axis image design reuse becomes possible. The paper was published in the IEEE VIS 2019. -
Visualization Assessment: A Machine Learning Approach
Paper Link: https://www.microsoft.com/en-us/research/publication/visualization-assessment-a-machine-learning-approach/
The paper was published in the IEEE VIS 2019, explored automatic evaluation method to visualize the picture characteristics, such as the degree of memory, aesthetics, let the machine learning algorithm play a role in the visualization of the generated recommendation. -
Supporting Story Synthesis: Bridging the Gap between Visual Analytics and Storytelling
paper links: https://www.microsoft.com/en-us/research/publication/supporting-story-synthesis-bridging-the-gap-between-visual-analytics -and-storytelling-2 /
the paper defines a framework for generating a new story, the results demonstrate that the process to analyze the data to a common story abstract generation process. The framework supports interactively generate the average reader can make the story understandable from a complex visual analysis results. Published in TVCG 2019. -
Cross-dataset Time Series Anomaly Detection for Cloud Systems
papers link: https://www.microsoft.com/en-us/research/publication/cross-dataset-time-series-anomaly-detection-for-cloud-systems/
article proposed set of anomaly detection based on the frame data migration across learning and active learning, can migrate between different time series data sets efficiently, only 1% to 5% of the labeled amount of the sample detected with high accuracy can be achieved. Articles published in the field of top-level system meeting USENIX ATC 2019. -
Robust Log-based Anomaly Detection on Unstable Log Data
Papers link: https://www.microsoft.com/en-us/research/publication/robust-log-based-anomaly-detection-on-unstable-log-data/
article LogRobust proposed model based on the depth of learning technology, which can effectively overcome the log instability, achieved outstanding results in practical industrial data fast iteration, the study published in the field of software engineering top conference FSE 2019. -
An Intelligent, End-To-End Analytics Service for Safe Deployment in Large-Scale Cloud Infrastructure
papers link: https://www.microsoft.com/en-us/research/publication/an-intelligent-end-to-end- analytics-service-for-safe-
deployment-in-large-scale-cloud-infrastructure / the paper presents a spatio-temporal correlation model, comparing the system state before and after the failure to provide clues for troubleshooting on the dual dimensions of time and space, the model has achieved a high accuracy in security deployments, the research results will be published in the top-level meeting NSDI 2020 systems. -
Outage Prediction and Diagnosis for Cloud Service Systems
papers link: https://www.microsoft.com/en-us/research/publication/outage-prediction-and-diagnosis-for-cloud-service-systems/
This paper proposes a kind of large-scale intelligent interrupted early warning mechanism AirAlert, AirAlert signal monitoring system to collect all the entire cloud system, using robust gradient boosting tree algorithms make predictions, and Bayesian network diagnostic analysis. Related research papers published in the WWW 2019. -
Prediction-Guided Design for Software Systems
papers link: https://www.microsoft.com/en-us/research/publication/prediction-guided-design-for-software-systems/
paper presents intelligent buffer management methods, based on expected guide (prediction-guided) framework for machine learning prediction engine as the core operating platform workloads and can monitor cluster deployed, these loads to predict the probability of failure and the new capacity growth in demand, dynamically adjusted reserve buffer. The method has been successfully integrated into the Microsoft Azure, increasing the robustness of capacity configuration, reducing the enormous costs. Research will be published in AAAI 2020 Workshop. -
An Empirical Investigation of Incident Triage for Online Service Systems
papers link: https://www.microsoft.com/en-us/research/publication/an-empirical-investigation-of-incident-triage-for-online-service-systems /
expand the Microsoft article 20 large-scale online service system based on case studies, we found fault assignment error results in additional time overhead, and then verify the effectiveness of existing software Bug assignment algorithm in fault Dispatched scene. This is the first study of fault assigned practice in the industrial large-scale online service system, research published in the ICSE SEIP 2019. -
Continuous Incident Triage for Large-Scale Online Service Systems
papers link: https://www.microsoft.com/en-us/research/publication/continuous-incident-triage-for-large-scale-online-service-systems/
the This article proposes DeepCT assignment algorithm based on continuous automated fault depth learning. DeepCT incorporates a new screening policies based on attention mechanisms, gating cell cycle model and improved loss function, you can gradually accumulate knowledge from engineers to discuss the problem and dispatch optimization results. Related results were published in the ASE 2019. -
Neural Feature Search: A Neural Architecture for Automated Feature Engineering
Papers link: https://www.microsoft.com/en-us/research/publication/neural-feature-search-a-neural-architecture-for-automated-feature- engineering /
paper presents a neural identity search (neural feature Search, NFS), based on recurrent neural network (Recurrent neural network, RNN) controller, by changing the rules of the most promising features of each original transformation achieved over the prior performance automatic feature engineering methods. The achievement has been in the field of data mining conference ICDM 2019 published, established a new level of technology in automatic feature engineering research.
-
Repairing Man-Made Meshes via Visual Driven Global Optimization with Minimum Intrusion
paper links: http://haopan.github.io/mesh_repair.html
method proposed article to repair the defect model ShapeNet, ModelNet other large 3D data sets. This article was published at SIGGRAPH Asia 2019. -
Learning Adaptive Hierarchical Cuboid Abstractions of 3D Shape Collections
paper link: https://isunchy.github.io/projects/cuboid_abstraction.html
artificial objects such as furniture generally has structural features, which human can be easily abstracted as a simple geometric objects the combination of shapes, such as rectangular, to facilitate understanding and analysis of the object. The paper by unsupervised learning on the same object, to generate rectangular abstract expression and adaptive hierarchical. Article published in SIGGRAPH Asia 2019. -
A Scalable Galerkin Multigrid Method for Real- time Simulation of Deformable Objects
paper link: http://tiantianliu.cn/papers/xian2019multigrid/xian2019multigrid.html
A over unstructured grid Galerkin multigrid method, which greatly accelerate the performance of the prior art flexible body simulation. The real-time simulation method can be flexible body finite element model with nearly a million of the people in the virtual world can interact with the complexity of the model and to promote one to two orders of magnitude. The paper was published in SIGGRAPH Asia 2019. -
Deep Inverse Rendering for High-resolution SVBRDF Estimation from an Arbitrary Number of Images
Papers link: https://gao-duan.github.io/
This paper proposes a method to optimize the textures intrinsic space, achieved against any number of input images textures modeled. Visually gives reasonable results when a given number of small images, but with the increase in the number of inputs, gradually more accurate reconstruction results. The paper was published in SIGGRAPH 2019. -
Synthesizing 3D Shapes from Silhouette Image Collections using Multi-Projection Generative Adversarial Networks
papers link: https://arxiv.org/abs/1906.03841
two-dimensional outline of a three-dimensional image learning object creation. The method requires a large number of two-dimensional contour image only for a certain type of object, does not require any correspondence relationship, which is distributed by the contour of the object classes in different directions is also characterized, learning and training data is generated that meets these three-dimensional distribution body. The paper was published in CVPR 2019.
