221 results about "Model architecture" patented technology

Presented herein are embodiments of state-of-the-art speech recognition systems developed using end-to-end deep learning. In embodiments, the model architecture is significantly simpler than traditional speech systems, which rely on laboriously engineered processing pipelines; these traditional systems also tend to perform poorly when used in noisy environments. In contrast, embodiments of the system do not need hand-designed components to model background noise, reverberation, or speaker variation, but instead directly learn a function that is robust to such effects. A phoneme dictionary, nor even the concept of a “phoneme,” is needed. Embodiments include a well-optimized recurrent neural network (RNN) training system that can use multiple GPUs, as well as a set of novel data synthesis techniques that allows for a large amount of varied data for training to be efficiently obtained. Embodiments of the system can also handle challenging noisy environments better than widely used, state-of-the-art commercial speech systems.

The invention discloses a government affair information resource sharing system based on a block chain, and belongs to the technical field of information sharing based on the block chain. By combiningP2P network, the cryptology and the artificial intelligence theory, the overall model architecture based on the block chain technology is provided, so that a government affair information resource sharing and exchange network system, a government affair information resource sharing and exchange block chain directory system, a certificate and trust system and an information resource sharing security intelligent exchange system are constructed. The government affair information sharing and exchanging system effectively solves the problems of trust islands, data ownership, peer-to-peer management, standard consistency, non-real-time exchange and the like in government affair information resource sharing application, and is high in safety, trustable, real-time exchange, data standard consistency, traceability and wide in sharing range.

The invention discloses a medical image synthesis and classification method based on a conditional multi-judgment generative adversarial network. The method comprises the following steps: 1, segmenting a lesion area in a computed tomography (CT) image, and extracting a lesion interested area (Region of Interest, ROIs for short); 2, performing data preprocessing on the lesion ROIs extracted in thestep 1; 3, designing a Conditional Multi-Discriminant Generative Adversarial Network (Conditional Multi-) based on multiple conditions The method comprises the following steps: firstly, establishing aCMDGAN model architecture for short, and training the CMDGAN model architecture by using an image in the second step to obtain a generation model; 4, performing synthetic data enhancement on the extracted lesion ROIs by using the generation model obtained in the step 3; and 5, designing a multi-scale residual network (Multiscale ResNet Network for short), and training the multi-scale residual network. According to the method provided by the invention, the synthetic medical image data set with high quality can be generated, and the classification accuracy of the classification network on the test image is relatively high, so that auxiliary diagnosis can be better provided for medical workers.

The invention discloses an electronic government affair system construction method based on block chain technology. According to the method, a distributed decentralized data architecture which is developed along with digital encrypted currency and is called as a block chain and a data operation mode are applied on the field of electronic government affair. The block chain is obtained through linking many blocks with same structures. The blocks are utilized for finishing packaging of government affair data with time stamps, a distributed networking propagation mechanism and a consensus algorithm task of distributed nodes. Government affair standard management level and handling efficiency can be greatly improved. Furthermore, according to the method, a D-H asymmetric encryption algorithm is applied in the block chain; and a public key password and a private key password which are different from each other are used for performing encryption and decryption. Safety of a government affair information database can be remarkably improved. The whole model architecture and the operation mechanism of the block chain ensure wide application space of the block chain in the field of electronic government affair.

The invention discloses a strategy management model architecture applied to streaming media service experience quality (QoE) of a peer-to-peer network (P2P) in a mobile network, which is capable of providing a well service QoE guarantee for a node in a dynamically changed mobile network environment and on the premise of satisfying user demands. The model architecture mainly consists of the following modules: a mobile P2P (MP2P) streaming media server used for user service request processing, user information obtaining and QoE management strategy execution; a QoE evaluation module used for the mobile network environment information obtaining and user QoE dynamic evaluation; and a strategy server used for dynamically selecting and updating the QoE management strategy information. The model architecture disclosed by the invention can be applied to the QoE strategy management of P2P streaming media service based on a mobile cellular communication system, can effectively reduce the influence of network dynamism and terminal isomerism to the service quality and achieves the purpose of guaranteeing the QoE performance of the mobile P2P streaming media service.

The invention relates to an embedded system reliability analysis and evaluation method. The method is technically characterized by comprising the following steps of: based on an AADL architecture model file and an AADL error model file, forming an AADL reliability model file; converting an AADL reliability model into a general stochastic Petri net (GSPN) reliability model; and performing quantitative analysis on the AADL reliability model by using a conventional GSPN reliability evaluation method. The method brings convenience to automation of software architecture reliability analysis, and brings the convenience to a user to analyze and evaluate the reliability of embedded software at an early stage of software design and evaluate the reliability of software at an architecture level; and if the model architecture cannot meet a requirement, then the software architecture can be modified in advance. Therefore, not only development cost can be saved, but also a development period can be shortened.

The invention discloses a construction method of a neural network model for realizing image classification. The construction method comprises the following steps: S1, constructing a unit structure search network, a system structure search network, an image training set and a random coding array; s2, generating a neural network model by using the unit structure search network, the system structuresearch network and the random coding array; s3, inputting the image training set into a neural network model to obtain an actual classification result; s4, judging whether an actual classification result meets a preset condition or not, and if not, performing a step S5; s5, updating the unit structure search network and the system structure search network according to the actual classification result and the theoretical classification of the image training set; s6, repeating S2; And S5, until the actual classification result obtained in the step S4 meets the preset condition. According to themethod disclosed by the invention, an original search space is converted into two spaces, namely a unit structure search space and an architecture search space, the optimal architecture of the architecture is searched in an automatic learning manner, and the flexibility of the generated model architecture is enhanced.

The invention provides a medical record text named entity recognition method based on an iterative expansion convolutional neural network. According to the method, the named entity recognition is carried out on a medical electronic medical record data set CCKS2017, a section of Chinese electronic medical record text is inputted, an iterative expansion convolutional neural network and a conditionalrandom field are used as a model architecture, the Chinese character components are used as features, and the named entities such as disease names and inspection means in the text are extracted.

The invention relates to the field of software technology development, in particular to elastic extensible multi-data-source mvc (model-view-controller) model architecture. The elastic extensible multi-data-source mvc model architecture is characterized in that each data source is provided with a corresponding model layer, namely, a business logical processing layer; different data sources are used for uniformly controlling interaction between a business layer and a presentation layer through a uniform controller layer; the presentation layer can be used for flexibly and dynamically selecting a multi-data-source business as required; different data sources correspond to sub-mvc modes; the presentation layer and the control layer are not designed independently; the control layer is common; the presentation layer is applied in a mixed way mostly, namely, different data source businesses can be called simultaneously in the same view interface. By adopting the elastic extensible multi-data-source mvc model architecture, the problem of difficulty in dynamical extension of other relevant businesses of multiple data sources in the conventional application is solved; the elastic extensible multi-data-source mvc model architecture can be applied to the development of Web applications.

The invention discloses a code generating method and system based on model driving. The system comprises a database model, a model architecture and code templates on a data model layer. In a code generating process, a model analysis module acquires model objects specified by programming personnel from database models and operational models, and invokes the corresponding code templates through a template engine according to the model objects; finally, a code generator cleans up the relations among the model objects, and generates final codes based on the model objects according to the model engine. The method and system provided by the invention adopt a positive code generating mode, a two-way derivation code generating mechanism and a plug-in type code generating architecture, and is high in efficiency and quality, so that higher-quality codes can be obtained according to the code generating technology.

The invention discloses a multi-field neural machine translation method based on a self-attention mechanism. The invention discloses a multi-field neural machine translation method based on a self-attention mechanism. The multi-field neural machine translation method comprises the following steps: carrying out two important changes on a Transformer, wherein the first change is a self-attention mechanism based on domain perception, and the domain representation is added to a key and a value vector of the original self-attention mechanism, the weight of the attention mechanism is the degree of correlation of the query and domain aware keys, the second change is to add a domain representation learning module to learn a domain vector. The method has the beneficial effects that a domain-aware NMT model architecture is provided on the basis of a neural network architecture Transformer representing the most advanced level at present. A self-attention mechanism based on domain awareness is provided for multi-domain translation. It is known that this is a first attempt on a multi-domain NMT based on a self-attention mechanism. Meanwhile, experiments and analysis also verify that the model can significantly improve the translation effect of each field and can learn the field information of training data.

The invention discloses a cognitive radio and edge calculation method based on an industrial wireless network. According to the method, a cognitive radio and edge calculation model architecture basedon the industrial wireless network is erected firstly. Secondly, system models are established, including a network model, a service model and a calculation model. Afterwards, distributed random optimization is described, including system state space, motion space and reward and optimization targets. Finally, optimal distribution motion minimizing the online calculation complexity is sought by useof dynamic random optimization and an augmented Markov decision process (MDP), wherein a state transition probability is calculated firstly, then a system reward is updated, and finally the augmentedMDP is reconstructed and joint resource management is performed. Through experimental tests, compared with a conventional method, the method can obtain higher system benefits, provide higher throughput and reduce system transmission delay under different spectrum arrival rates and numbers of CRECs.

The invention relates to a jet energy distribute simulating system based on griddling in the field of computer simulating domain. The system frame is positioned on the top layer of the system; the part module or interface is corresponding with the motor part module frame middle module and is used to transmit the data and control the simulating sequence between the modules; the motor part module frame is established on the simulating serve achieving frame which is used to seal the motor part and serve; the standard griddling serve interface and simulating serve achieving frame provide each network basic function to achieve the connection; the simulating serve achieving frame combines the computer basic facilities and the simulating source into a standard network computer basic facilities which is used to mask the network basic hard and software resource and to provide the standard network facilities interface to the motor part module frame.

The invention relates to a platform virtualization system comprising a CPU simulator, a memory virtualization module, and an external virtualization module. The CPU simulator reads an X86 architecture code instruction and judges whether an instruction basic block is translated or not; a binary translator is used for translation and comprises a translation engine and an execution engine; the translation engine translates an X86 architecture code into a Loongson platform code; the execution engine prepares the operational context of the Loongson platform code, locates the Loongson platform code corresponding to the X86 architecture code from a Loongson platform code cache and executes the code. The memory virtualization module uses a shadow page-table method. The external virtualization module establishes a corresponding device model for each external device. An X86 architecture virtual machine interacts with the external devices through the device models, thereby discovering and accessing the devices. The platform virtualization system allows information systems not matching with the domestic Loongson hardware platform yet to run in the domestic software-hardware environments in a virtualized manner, and contributions are made for the smooth transition between new and old technical systems in the automatic upgrading process of the information systems.

The invention relates to an embedded software testing method based on AADL (Architecture Analysis and Design Language) mode transformation relationship, which has the following steps of: constructing a mode transition diagram on the basis of mode information in an AADL model, and converting the diagram into a mode relationship tree required by a transformation test according to the improved depth-first traversing algorithm; constructing a source test case in the mode transformation relationship by traversing the mode relationship tree, generating a subsequent test case by means of the mode transformation relationship in the AADL model, and verifying the mode transformation relationship to obtain the conclusion of the transformation test. The embedded software testing method based on AADL mode transformation relationship solves the 'Oracle' problem existing in the embedded software test, is convenient for a user to test the embedded software at an early stage of software design and ensures the reliability of software at a system architecture level. If the model architecture can not meet corresponding requirements, the architecture of the software can be modified at an early stage of development, thus the development cost is saved, and meanwhile, the development cycle can also be shortened.

A method for implementing a model-driven architecture, including defining a principal model having a plurality of classes, references, attributes, and associations between any of the classes, the model configured to facilitate the automatic generation of at least one resource for use by a computer-executable application, where a change to the principal model subsequent to performing the automatic generation requires the automatic generation be performed again in order to effect the change for use by the application, defining a decoration model having a class, reference, and attribute for any corresponding one of the primary model classes, references, and attributes, where a change to the decoration model subsequent to performing the automatic generation does not require the automatic generation be performed again in order to effect the change for use by the application, mapping the decoration model to the principal model, and storing both of the models on a computer-readable medium.

The invention discloses an action set output method and system based on multi-agent reinforcement learning. The method comprises the following steps: S1, constructing a model architecture of a tree structure; s2, modeling each child node in the tree structure constructed in the step S1 as an intelligent agent, and modeling a multi-intelligent-agent reinforcement learning system through a hierarchical extended Markov game; s3, enabling all agents to interact with the environment, and carrying out reinforcement learning training to form an action set output model; and S4, scoring each action inthe action space to be processed by utilizing the multi-agent reinforcement learning action set output model, and generating a target action set for recommendation. According to the method, a multi-agent reinforcement learning method is used for processing an action set decision problem of a large-scale action space, so that good expandability and more accurate and faster training and reasoning speed can be obtained; according to the invention, the MCTS algorithm is used to increase the amount of information for decision making of the upper-layer agent, effective search can be carried out, anda more accurate decision can be obtained.

The invention discloses a micro-application / network transmission / physical (Micro-ANP) Micro-ANP communication protocol model architecture method of an underwater acoustic sensor network (UWSN). The method comprises three parts, i.e. Micro-ANP protocol stack hierarchy classification, Micro-ANP network transmission layer protocol composition and grouped format design, and the optimization of UWSN pack load length. Due to the concise protocol hierarchies and unique network grouping format of Micro-ANP, multiple addresses, first length, calibration, upper protocol type and other packaging, brought by a traditional hierarchy model, can be reduced, and the problems that in a sensor node, uplink protocol stacks cannot be too complicated due to computation, storage, energy and other extremely limited resources can be solved; and due to the optimized design of the UWSN pack load length, the energy and the consumption of end-to-end delay and other resources can be reduced, and simultaneously, the network throughput and the transmission reliability are improved. The Micro-ANP communication protocol model architecture method of the underwater acoustic sensor network is applicable to network communication of underwater data acquisition, environment monitoring, disaster prevention, resource monitoring and other application based on the UWSN, and has good application prospect as an effective and practical technical scheme.

The invention relates to a proton exchange membrane fuel cell modeling method. The method comprises the steps of activating an overvoltage model act, an ohmic overvoltage model ohmic, a concentrationovervoltage model con and a voltage drop model Eloss caused by internal current / permeation loss. According to the technical scheme, PEMFC voltage loss caused by hydrogen permeation or internal currentis used as a calculation part of the model; Through the PEMFC modeling method provided by the invention, the PEMFC model architecture is more reasonable to build, more conforms to the working mechanism of the proton exchange membrane fuel cell, and the model simulation calculation is more accurate.

The invention discloses a legal knowledge graph construction method and equipment based on entity relationship joint extraction. The construction method comprises the following steps: constructing a triple data set; designing a model architecture and training a model, wherein the model architecture comprises a model coding layer, a head entity extraction layer and a relation-tail entity extraction layer; judging a relationship between text sentences; and carrying out triple compounding and map visualization. According to the design of the model architecture, a Chinese bert pre-training model is adopted as an encoder, and the Chinese text encoding effect is good. According to the entity extraction part, two BiLSTM binary classifiers are adopted to judge the starting position and the ending position of an entity, and the entity in a phrase form in a text can be effectively extracted. According to the method, the head entity is firstly extracted, then the tail entity corresponding to the entity relationship is extracted from the extracted head entity, and when the entity relationship and the tail entity are extracted, not only is coded information of sentences used, but also coded information of the head entity is fused. According to the method, the legal knowledge graph with relatively high accuracy can be obtained.

The invention discloses a message-oriented middleware system and an implementation method, which realize the effects of high performance and low delay and can realize microsecond-level delay on the basis of not reducing the reliability index of the system. According to the technical scheme, a service data flow model architecture arranged in a message middleware system is a reliable low-delay system solution realized through technologies such as transmission strategy optimization, lock-free queues and memory zero copy in the field of financial derivative transaction. A unified API dynamic upgrading model architecture arranged in the system also solves the problem of frequent API updating caused by service change.

The invention discloses a sequential image prediction method based on LSTM and DCGAN, and the method combines the excellent feature capture capability of the DCGAN with the LSTM, can enable the predicted image data to be visualized, and facilitates the direct observation. The improved LSTM network has convolution characteristics inside, and two-dimensional spatial characteristics of image data canbe directly learned; In order to reduce the internal learning complexity, a traditional input image is changed into an input feature; Characteristics are extracted from DCGAN, and compared with an original image, the method has the advantages that the dimension is greatly simplified, and the whole network is controllable. According to the method, the feature dimension is well reduced through theDCGAN, and the problem that the high dimension cannot be calculated is solved; The improved LSTM can better learn time sequence characteristics, so that more accurate prediction is realized; The wholenetwork structure complies with a stack type cascading strategy in the connection method, and guarantees are provided for controlling the network depth. The sequential image prediction model architecture provided by the invention is theoretically suitable for all sequential images.

In some embodiments, a method and apparatus for simulating segmented addressing on a flat memory model architecture are described. In one embodiment, the method includes the translation of instructions from a source instruction set architecture (ISA) having a segmented memory addressing model into a target ISA having a non-segmented memory addressing model. The conversion of instructions into translated instructions for execution within the target ISA is performed by simulating a segmented memory addressing model within the target ISA for the translated instructions. In one embodiment, hardware components and data structures of the target ISA are allocated to simulate the segmented memory addressing model within the target ISA. Accordingly, translated code utilizes allocated flags, such as predicate registers, in order to convert translated program statements having logical address expressions into linear addressing expressions supported by the target ISA. Other embodiments are described and claimed.

The invention discloses an electrical simulation modeling method and device and a readable storage medium, and the method comprises the following steps: obtaining an original design file, and carryingout the classification of the original design file; carrying out information extraction on the classified original design files to obtain an electrical entity list and a key attribute list, and carrying out association matching to form an electrical parameter database; and importing the electrical parameter database into electrical simulation software to complete modeling. By classifying originaldesign files, extracting to obtain an electrical entity list and a key attribute list, performing association matching to obtain an electrical parameter database, and packaging into a simulation datapacket according to a data interface and a protocol, and finally, importing the simulation data packet into electrical simulation software to complete modeling, an implementation scheme is provided for an intelligent establishment function of the electrical simulation model from the problems of links such as electrical simulation model architecture and establishment.

The invention discloses a lithium ion battery residual life prediction method based on fusion of improved particle filtering and a double-exponential decay empirical physical model. Aiming at the problem that the precision of a method based on data driving seriously depends on the perfection accuracy degree of a model architecture, the lithium ion battery residual life prediction method adopts a nonlinear least square method to carry out parameter identification on a double-exponential model, utilizes methods such as analogue simulation and test measurement to verify a specific research objectbattery and optimize an empirical model, meanwhile, adopts a statistical correlation coefficient theory to improve a resampling strategy, utilizes a path similarity degree threshold value to correctthe particle weight again, and abandons state smooth estimation to solve the problem of particle degradation in a standard PF algorithm. Based on this, a complete set of lithium ion battery remaininglife prediction systematic research method integrating an improved particle filtering algorithm based on a correlation coefficient theory and a parameter identification double-exponential recession empirical model with scientific and accurate architecture is proposed, and high-precision and high-timeliness prediction of battery health management is fully realized.

The embodiment of the invention provides a training method, a detection method and a system for a malicious webpage detection model. The training method comprises the following steps of: obtaining a data set of a webpage and judgment result data of whether the webpage is a malicious webpage or not; Processing the data set; Establishing a training model architecture; And taking the processing result of the webpage data set as input data, and taking the judgment result as output data to train the training model architecture, so as to form a detection model capable of predicting whether the webpage is a malicious webpage based on the input webpage data processing result. According to the training method in the embodiment of the invention, the detection model for simply and efficiently predicting whether the corresponding webpage is the malicious webpage based on the input network data can be trained, so that the precision is higher when the webpage is predicted through the detection model, and meanwhile, the prediction operation of a user on the webpage is greatly simplified.

The invention discloses a training method of a personalized model of distillation-based semi-supervised federated learning, which adopts a knowledge distillation technology, and a client side can select a self-designed model architecture by uploading model prediction rather than model parameters, so that privacy information of the client side about the model is well protected, and the shared data and the local data of the client are used for training together, so that the generalization ability of the model is greatly improved. In addition, the aggregation scheme can perform dynamic aggregation according to the importance degree of knowledge provided by each client, so that the aggregated model prediction better fuses the model knowledge of the client. And after the server finishes aggregation, the model prediction distribution information of the public data is not the pseudo label information, so that the communication transmission efficiency is further improved by utilizing the mode.

An ocean single-element observation quality control method based on multi-model fusion adopts a four-layer model architecture combining statistical analysis and a single classification algorithm to detect whether historical observation data of a certain element of an ocean site is abnormal or not. The method comprises the following steps: S1, an input layer, which constructs three time windows from far to near for historical observation data of a certain element of a marine site, extracts statistical features, fits features and classification features, and constructs a detection sample; S2, a statistical analysis layer, which filters 70% of positive samples by using a statistical discrimination algorithm, reduces the scale of an abnormal candidate set, and effectively relieves the influence caused by imbalance of the positive and negative samples; S3, a single classification layer, which further detects the suspected abnormal observation data points by using a single classification model; and S4, the output layer, which is used for comprehensively making a final judgment according to results of the statistical analysis layer and the single classification layer, and evaluating a detection result. According to the method, the detection results of various models are comprehensively considered to make an optimal decision, so that the accuracy of the detection method is effectively improved.