303results about How to "Fast iteration" patented technology

The present invention discloses a co-processor IP core of a programmable convolutional neural network. The invention aims to realize the arithmetic acceleration of the convolutional neural network on a digital chip (FPGA or ASIC). The co-processor IP core specifically comprises a global controller, an I / O controller, a multi-level cache system, a convolution unit, a pooling unit, a filling unit, a full-connection unit, an internal interconnection logical unit, and an instruction set designed for the co-processor IP. The proposed hardware structure supports the complete flows of convolutional neural networks diversified in scale. The hardware-level parallelism is fully utilized and the multi-level cache system is designed. As a result, the characteristics of high performance, low power consumption and the like are realized. The operation flow is controlled through instructions, so that the programmability and the configurability are realized. The co-processor IP core can be easily applied to different application scenes.

The invention provides a word vector training method and device. The method comprises the following steps: an internet webpage is captured, and training language materials are acquired and stored in a corpus; each training language material in the corpus is subjected to word segmentation, and an orderly word set corresponding to each training language material is obtained; a word list is established according to pre-collected user query logs; the training language materials stored in the corpus are distributed to nodes of a distributed word vector learning model; the distributed word vector learning model is configured to perform periodic word vector training on each word in the word list, and the word vector corresponding to the word in the word list is obtained. According to the word vector training method and device, the word vectors obtained through training can be well applied to search business, and fast iterative high-quality word vector training can be realized.

The invention relates to an information recommendation method and device. The method comprises the following steps: determining target information, and obtaining user historical behavior data and user information associated with the target information; according to the user information and the user historical behavior data, determining training set data, and calculating the feature value vector of the target information; according to the attribute value list of a user to be screened and the feature value vector, determining a target user; and recommending the target information to the target user. By use of the information recommendation method and device, the technical problem that information can not be accurately recommended is solved, accurate information recommendation is carried out on the user through the analysis of the user information and the user historical behavior data, time for the user to obtain useful information is saved, and an effect on improving calculation efficiency is achieved.

A system and method for measuring elemental concentrations of a material from a sample containing several elements by LIBS analysis. The material is heated to generate plasma and its chemical composition is determined from spectral analysis of its radiation. The spectral lines of interest are identified among those emitted by the constituents of each element composing sample. The intensities of the spectral lined identified are measured. From an estimate of temperature, electron density and relative concentration values, the chemical composition of the plasma is calculated. The absorption coefficient according to wavelength is calculated for the spectral zones of the lines of interest. From an estimate of the plasma width, the spectral radiance of the plasma is calculated for the same spectral zones and then a comparison of the intensity and shape of the spectrum thus calculated with those of the spectrum measured is performed. These calculations and this comparison are repeated iteratively in order to adjust the temperature, electron density, relative values of the elemental concentrations and width of the plasma.

The invention discloses a high-precision map-based cloud control automatic driving task generating method. The method comprises the following steps: acquiring GPS information of a road by using centimeter-level high-precision positioning equipment, performing coordinate rotating conversion on a GPS track, performing curve fitting and selecting high-precision points to acquire the GPS sequence of a lane level of each road, establishing a high-precision map suitable for automatic driving and storing in a cloud server; and in the actual automatic driving stage, receiving an initial position and destination combined high-precision map and vehicle information in a global range from a vehicle by a cloud side, generating the route of the lane level, planning a driving task from the route of the lane level to form a task sequence, and issuing to the vehicle to perform execution and realize automatic driving. By the high-precision map-based cloud control automatic driving task generating method, hardware calculation resources of the internet and the cloud side are effectively utilized, and the complexity and the cost of the vehicle are reduced; and the cloud side masters the global traffic information, application of cloud processing, big data and artificial intelligent technology is benefited, and the automatic driving technology can be iterated and improved rapidly.

The invention discloses a software development platform, which belongs to the technical field of software development. The platform is a PaaS platform of an enterprise level. Enterprises can be helpedto go to DevOps practice. A business capability supermarket is constructed, an enterprise is enabled to get rid of complex infrastructure management and more concentrate on business research and development and delivery, an internet distributed architecture is adopted for design and development, normalization, standardization and internet technicalization of development and operation and maintenance processes are realized through a PaaS platform, and the development efficiency and the operation and maintenance capability are improved; design is carried out based on a micro-service architecture, high reliability and high expansion capability are taken as basic attributes, and use scenes such as single-point faults and large concurrency are dealt with; pROVIDING AUTOMATED DEPLOYMENT OF STANDS, rapid iteration and automatic operation and maintenance are used as basic attributes; the software development platform supports scenes of rapid development, smooth upgrading and the like, so thatthe user experience of the software development platform can be improved, the software development speed and the software quality are improved, one-stop service is provided for the software development process, smooth and continuous development of software and dynamic analysis and positioning during operation are facilitated, and the customer satisfaction is improved.

The invention provides a generation method and device and an application method and device for a knowledge graph of a bank financial system. The method comprises the following steps: identifying a core entity of an ontology layer in the bank financial system and a relationship between the core entities based on domain modeling by adopting a process from top to bottom; analyzing and identifying entities in the bank financial system and relationships among the entities by adopting a bottom-to-top process and integrating various system requirements and design documents; and fusing the relationship between the core entities identified by the process from top to bottom with the relationship between the entities identified by the process from bottom to top to obtain the knowledge graph of the bank system. The invention provides a generation method and device and an application method and device for a knowledge graph of a bank financial system. Construction modes from top to bottom and from bottom to top are used as complementary schemes, and the reliability of a model can be regressed and verified by matching the relationship between a natural entity and a data entity through the design.

A distributed ecological intelligent internet pension service system comprises a government-level multi-department unified command and supervision layer, a region-level central management and control layer, a community-level service guarantee layer and a user terminal function layer. The user terminal function layer acquires user health basic data and sends the data to the region-level central management and control layer; the region-level central management and control layer receives the data and community comprehensive service guarantee service configuration information, builds a user database according to the information, generates region user group reports and distributes the reports to corresponding organizations; the government-level multi-department unified command and supervision layer sends related decisions and instructions to the region-level central management and control layer. The distributed ecological intelligent internet pension service system is an intelligent service system based on the internet of things, the internet and pension, and has the advantages of data standardization, structure networking, system self-evolution, operation disintermediation, high information security degree and the like.

The invention provides a method for quickly developing a heterogeneous parallel program, and relates to performance analysis of a CPU (central processing unit) serial program and transplantation of a heterogeneous parallel program. The method includes: firstly, performing performance and algorithm analysis on the CPU serial program, and positioning a performance bottleneck and parallelizability of the program; secondly, inserting an OpenACC pre-compilation command on the basis of an original code to obtain a heterogeneous parallel code which can be executed in heterogeneous parallel environment; compiling and executing the code according to specified parameters of a hardware and software platform, and determining whether further optimization is needed or not according to a program run result. Compared with the prior art, the method has the advantages that existing codes need not to be reconstructed; multilanguage support is realized, and languages such as C / C++ and FORTRAN (formula translator) are supported; cross-platform and cross-hardware are realized, operating systems such as Linux, Windows, Mac and the like are supported, and hardware such as Nvidia, GPU of AMD and Intel Xeon Phi is supported. By the method which is high in practicality and easy to popularize, existing programs can be parallelized efficiently, and the programs are enabled to make full use of computing power of a heterogeneous system.

The invention discloses a machine learning system. The system comprises clients used to collect to-be-recovered data and a cloud-end server connected with the client via network. Each client comprises a first processor provided with an improved intelligent algorithm; and the cloud end server comprises a second processor provided with a BP neural network model, the second processor processes the to-be-recovered data via the BP neural network model and optimizes initial parameters value of the BP neural network via the improved intelligent algorithm, and the initial parameter values include an initial weight and an initial threshold. The machine learning system is higher in learning speed and learning precision.

A mine car unmanned driving system comprises a cloud computing and remote service device (1), a communication device (2), a sensing device (3) and a control device (4), the cloud computing and remoteservice device (1) is used for generating a global path and a work task, and the global path and the work task are sent to the control device (4) through the communication device (2); the sensing device (3) is used for sensing the surrounding environment of the vehicle and sending the surrounding environment of the vehicle to the control device (4), and the sensing device (3) is composed of more than one sensing sensor; and the control device (4) is used for controlling the vehicle to run according to the global path, the work task and the vehicle surrounding environment.

The invention discloses a method for calculating stray current and neutral point direct current of transformer caused by stray current in a metro. The method comprises the following steps: 1. The structure of rail-drain network-earth in metro system is equivalent to a three-layer resistance network model, and the resistance network is divided into several sections equally; 2, that flow path of theDC bias magnetic field of the transformer caused by stray current is analyzed, and a current branch of the DC bias magnetic field of the transformer caused by stray current is added into the resistance network model; 3, KVL equations along respective loop are arranged to obtain loop current equations for that built resistance network model; 4, a loop current equation is resolved by using the LU decomposition method, and the stray current of the rail leakage and the neutral point current of the transformer is calculated. This method can be used to calculate the distribution curves of neutral current and stray current of transformer under subway operation, which provides a theoretical basis for evaluating the influence of stray current on DC magnetic bias of transformer.

Provided is a browser plug-in generating, issuing, loading and updating method and system. The browser plug-in generating method includes: respectively defining virtual application program interfaces (API) for one or a plurality of specific functions; defining business logic codes for calling at least one virtual API; building an API function suitable for a specific browser for the at least one API; building a plug-in starting interface for positioning the business logic codes; and building a browser plug-in suitable for the specific browser, wherein the codes for calling the plug-in starting interface, the codes for loading the business logic codes and the built API function suitable for the specific browser are packaged in the browser plug-in.

The invention discloses a hard acceleration method and device for a neural network model of first electronic equipment. The method and the device relate to the technical field of deep learning in artificial intelligence. The method includes the following steps: obtaining to-be-recognized data and configuration parameters of the neural network model; performing hard acceleration of convolution calculation matching the neural network model according to the configuration parameters, and obtaining a convolution result of the to-be-recognized data by the neural network model; performing the hard acceleration of function calculation on the convolution result by calling at least one function module matching the neural network model from at least one preset function module based on the configuration parameters, and obtaining a recognition result of the to-be-recognized data by the neural network model. The method and the device can support the neural network model established by various open source development environments, and also support a user-defined neural network model; and when an algorithm of the neural network model is updated, only the parameters of the first electronic equipment need to be reconfigured without changing hardware design.

The invention discloses a web application test data flow tracking method and system based on an IAST test platform. The data flow tracking method comprises the following steps: installing an agent program on a server where the tested application program is located, wherein the agent program is used for communicating with a virtual machine running the application program; after the application program is started, inserting monitoring codes into key functions in the input stage and the output stage of the application program through the agent program and the byte code instrumentation tool; and when the monitoring code is instrumented, introducing a class file name, a method name, a method parameter and a return value of a to-be-instrumented function into the monitoring code, obtaining a propagation path of stain data in the application program through the monitoring code, wherein the stain data is data input by a user in the testing process. by adopting the data flow tracking method, real-time tracking of taint data can be realized, and an execution method chain in a request-response process can be accurately obtained based on an actual request.

Described is a modeling and analysis design environment allowing the specification of an architectural lighting system, composed of both natural and artificial lighting elements and lighting controls. The modeling environment allows users to create 3D models through a series of plan and section drawings. Its glyph language also provides for quick specification of elements such as windows, luminaires, and control systems. The analysis workbench provides both visual and robust way of analyzing multidimensional data, characteristic of lighting simulation. One aspect of the invention is a method for evaluating combinations of artificial and natural lighting to optimize lighting quality and energy cost. This method includes using integrated Plan / Section approach for specification of 3D lighting models, glyph language for quick specification of geometry in Plan / Section, a calculation manager, and visual, spreadsheet-like language for managing spatial and temporal data.

The invention discloses an App automation continuous integration system and method based on an iOS system. The App automation continuous integration system includes a MAC client, a certificate module,a component module, a Git module, a compilation module, a resource distribution module and a test module. A required certificate, a component, a component version and a source code are uploaded to the Mac client; a compilation instruction is initiated from the Mac client to a compilation server, and the compilation server downloads the required source code from a Git server and parses a lock file, downloads and compiles the required certificate and component into an ipa program package, then executes packaging and uploads the package to a distribution server; the distribution server executesthe App distribution to a test server through the instruction of the Mac client, and / or uploads Apps to an AppStore and / or downloads the Apps according to different purposes of the Apps; and the testserver sends execution test cases to an iOS mobile phone client and / or an iOS simulator, and generates and feeds back test reports to developers. According to the invention, the APPs can be continuously integrated in the unit of days, compilation problems can be quickly found, therefore, the development cycle is shortened, and the version is rapidly iterated.

The invention provides a computing method for an unsteady flow field of rotary impeller machinery based on a three-dimensional dynamic mesh. The method is applied to the technical field of rotary impeller machinery, is based on the technology of three-dimensional dynamic mesh, realizes the compute of the unsteady flow field of rotary impeller machinery, and has higher iteration speed compared with the traditional sliding mesh method.

The invention relates to a novel ankle joint rehabilitation robot. The robot includes an electric control box, a base, a drive motor, an encoder, a 3-RRR spherical parallel mechanism and a foot pedal; the 3-RRR spherical parallel mechanism is located between the base on the bottom and the foot pedal on the upper portion and has three identical motion branch chains, each branch chain is composed of three revolute pairs and two connecting rods, the 3-RRR spherical parallel mechanism has three pure rotational freedom degrees, all rotating axes are intersected at one point, the point is the rotation central point O of the 3-RRR spherical parallel mechanism, and axes of the revolute pairs of all the motion branch chains are also intersected at the point O. Since the 3-RRR spherical parallel mechanism has three rotational freedom degrees, the work space can meet the range requirement of all ankle joint movement angles; meanwhile, the physiological characteristics that the ankle joint is high in rotation flexibility and strong in bearing capacity are also achieved.

The invention discloses an online compensation method for a spatial error of a multi-axis linkage CNC machine tool. An AC pendulum five-axis CNC machine tool is connected with a servo driver through aspatial error compensation PC, the spatial error compensation PC includes a grating signal input interface, spatial error compensation software and a grating signal output interface, the spatial error compensation software acquires a grating signal of the AC pendulum five-axis CNC machine tool through the grating signal input interface, according to the position coordinate fed back by the gratingsignal and an error value in an error parameter file in the spatial error compensation software, a spatial error of the current position and a compensation value of each axis of the AC pendulum five-axis CNC machine tool are calculated, after a feedback signal of a grating is modified according to the compensation value, the feedback signal is sent to the servo driver through the grating signal output interface. The method is advantaged in that the online error compensation idea and method are realized by modifying the grating feedback signal, iterative compensation calculation is carried outin two steps, iteration is made to converge quickly, and computational efficiency is high.

The invention discloses a method and device for achieving a low delay CORDIC trigonometric function based on a carry-save summator. The method comprises the following steps of initializing an X data path, a Y data path and a Z data path; predicting the rotating direction of the Z data path, and outputting the rotating direction to the X data path and the Y data path, compressing the front N / 2 circulation of the X data path and the Y data path into the N / 8 level, and completing iteration for each level in one beat by adopting a carry-save adder; conducting parallel iteration on the last N / 2 circulation of the X data path and the Y data path through the carry-save summator. The device comprises an initialization module, a rotating direction predicting module, an iteration compressing module and a parallel iteration module, and the carry-save summator is adopted in both an iteration compressing unit and a parallel iteration unit. The method and device have the advantages that hardware expenditure is small, a period is short, delay is short, throughput is large, and precision is high.

The invention provides a middleware-based interface implementation method, a middleware-based interface implementation system, a computer device, and a computer readable storage medium. The middleware-based interface implementation method comprises the following steps: establishing an interface adapted to multiple protocols to receive operation request data of different formats of a client; converting the operation request data of different formats into standard data of preset formats according to pre-established standard data templates; establishing a service logic of the interface accordingto the converted standard data based on the middleware of information data transfer; and executing the service logic to implement the release of the interface service. By adoption of the middleware-based interface implementation method, the modularization of various stages of interface development can be achieved, therefore the development of different services is maximized, the requirements of rapid iteration, complex structure and diversified modes are met.

Generative adversarial networks (GAN) display a great development prospect in image generation and semi-supervised learning, and have been developed to Triple-GAN. However, there are two problems requiring solving based on the classification method based on Triple-GAN, wherein the two problems comprise a problem that KL-based divergence distribution construction is easy to generate gradient disappearance to cause unstable training and a problem that the Triple-GAN employs a manual mode to perform mark of tags for samples to cause too large workload of manual marking and non-uniform marking. Based on this conditions, RandomForests are employed to perform classification of real samples to perform automatic marking of tags of blade nodes, and the ideal of the least square generative adversarial networks (LSGAN) are employed to construct a loss function to avoid gradient disappearance.

The invention discloses a currency credit big data monitoring analysis system. The system comprises a client side, a server, a service end and a data end. The client side comprises a client terminal and a software client running on the client terminal, and is used for importing operation data of each financial institution and carrying out collection, cleaning, verification and sorting processing on imported data. The server side is connected with the client side and the data side and used for business logic processing and data processing among the client side, the data side and the server side. The data end is used for realizing distributed storage of the acquired and processed data and providing a high-speed access engine for data extraction and query of the server. The problems of narrowcoverage, low expansibility, insufficient flexibility, poor timeliness, low security and the like in the existing currency credit management and data analysis services can be solved, currency policyand credit policy data acquisition and big data processing can be realized, and efficient utilization and monitoring of the currency policy and credit policy data are realized.

The invention belongs to the technical field of vehicle license plate recognition and correction and discloses a vehicle license plate image vertical tilt correction method for determining a vertical tile angle based on cross correlation. According to the method, based on the principles of envelope and cross correlation, the vertical tile angle of a vehicle license plate binary image is determined through calculating the ratio of an offset of a horizontal coordinate corresponding to the cross correlation maximum value of upper envelope and lower envelope on a vehicle license plate binary image, and the vertical tile of the vehicle license plate binary image is corrected through shear transformation. According to the method, an iterative solution way is employed, a vertical tilt angle error caused the asymmetry of the upper envelope and lower envelope is overcome, the vehicle license plate image vertical tilt correction method has the advantages of fast operation speed and high accuracy, and the real-time requirement by a vehicle license plate recognition system is fully satisfied.

The invention provides a sample screening method, system and device and a medium. The method comprises the steps of: obtaining part of unlabeled samples from a target sample set to form a candidate sample set; utilizing the classification model to predict each unlabeled sample in the candidate sample set for multiple times, and calculating an uncertainty value of each unlabeled sample in the candidate sample set according to multiple prediction results; and screening the first K unlabeled samples with the highest uncertainty values from the candidate sample set to serve as training samples ofthe classification model. According to the invention, a small number of most representative samples can be selected from mass data for manual annotation, and the problem that in the prior art, when full-amount samples need to be annotated, the cost is huge is solved. In addition, the number of data required to be manually annotated when the classification model is trained can be reduced, the laborcost of manual annotation is reduced, and the annotation efficiency is effectively improved. And meanwhile, the model can be quickly iterated and optimized at the minimum data annotation cost, so that an optimal model can be trained by using less data.

The invention discloses a low-profile holographic imaging antenna based on a Fabry-Perot resonant cavity structure. The antenna is based on a Fabry-Perot resonant cavity structure, operates in a leakywave mode, mainly includes a feed source antenna, an artificial magnetic conductor and an encoding part reflective surface. The transmitting phase of the encoding part reflective surface is arrangedcorrespondingly by a phase retrieval algorithm, such that the holographic imaging can be performed at a corresponding position in the near field. Compared with the prior art, the low-profile holographic imaging antenna has a low profile, is easy to design and machine, can be used to realize a low-profile two-dimensional beam deflection and multi-beam outgoing antenna, and can be used to achieve low-profile beam scanning and novel millimeter-wave wireless communication systems in cooperation with hardware, such as an FPGA and the like.

A method and a computer software program for predicting and modeling ion flux colliding with an ion-propelled vehicle traversing a magnetic field predicts a colliding ion path based upon an ion exhaust velocity, an ion charge-to-mass ratio, and a localized magnetic field vector. An ion-propelled vehicle path is predicted based upon a selected trajectory and a vehicle velocity vector. An ion plume is transformed based upon the predicted colliding ion path. The ion-propelled vehicle is projected into the transformed ion plume. The flux is calculated at a silhouette of the projected vehicle.

The invention discloses a DNA storage coding optimization method based on a K-means clustering multivariate universe algorithm, and the method specifically comprises the steps: constructing an optimalDNA coding sequence meeting a combination constraint condition, firstly constructing a certain number of DNA sequences as an initial population, and carrying out the evaluation and sorting of the fitness of the population; secondly, optimizing the obtained DNA coding sequence by using a k-means clustering algorithm and wormhole intersection to obtain a DNA coding sequence with higher fitness; then, judging whether to add an alternative solution set or not according to constraints through constraint comparison; and finally, outputting an optimal DNA coding sequence. According to the method, DNA coding sequences with relatively good quantity can be searched.