73results about How to "Enhance reasoning ability" patented technology

The invention discloses a medical interrogation dialogue system and a reinforcement learning method applied to the medical interrogation dialogue system, and relates to the technical field of medicalinformation. The system comprises a natural language understanding module used for classifying the intentions of users and filling slot values to form structured semantic frames; a dialogue managementmodule used for interacting with a user through a robot agent, inputting a dialogue state, performing action decision on the semantic frame through a decision network, and outputting final system action selection; a user simulator used for carrying out natural language interaction with the dialogue management module and outputting user action selection; a natural language generation module used for receiving system action selection and user action selection, enabling the user to check the selection through generating sentences similar to a human language by using a template-based method. According to the invention, the medical knowledge information between diseases and symptoms is introduced as a guide, and the inquiry historical experience is enriched through continuous interaction witha simulated patient. The reasonability of inquiry symptoms and the accuracy of disease diagnosis are improved, and the diagnosis result is higher in credibility.

The invention discloses a knowledge graph-based mechanical fault diagnosis knowledge base construction method, and belongs to the field of mechanical fault diagnosis. A mechanical fault diagnosis knowledge base reflects fault generation essences and domain expert experiences; and through a knowledge processing module, the fault generation essences and the domain expert experiences are stored in the knowledge base, thereby providing support for mechanical fault diagnosis. A conventional knowledge graph is represented in a network form; nodes represent entities; connection lines represent relationships; and for the representation form, a special graph algorithm needs to be designed for storing and utilizing a database, so that the disadvantage of time and labor waste exists. According to a representation learning technology represented by deep learning, a triple object is mapped to a vector space and represented as a dense low-dimensional vector, and efficient calculation and reasoning are realized through vector conversion. The knowledge graph-based mechanical fault diagnosis knowledge base construction method is established; mechanical fault diagnosis knowledge is represented as atriple, and the tripe is represented as the vector by utilizing a TransD model, so that the problems of inaccurate case representation, difficult maintenance and modification, low reasoning and calculation efficiency and the like of a conventional knowledge base can be optimized; and the method has important significance for the field of fault diagnosis.

The invention discloses an intelligent transformer fault diagnostic method based on an RBF (radial basis function) neural network. The intelligent method includes (1), acquiring three ratios of five gases C2H2 / C2H4, CH4 / H2, C2H4 / C2H6 as training sample data by the utilizing IEC (international electrotechnical commission) three ratio method; (2) performing fuzzily processing on the three ratios by utilizing a membership function; (3), coding fault types; (4), training the RBF neural network according to the training sample data until the RBF neural network meets precision requirements; (5), inputting to-be-diagnosed samples after being fuzzily processed; and (6), outputting diagnosed results. The intelligent transformer fault diagnostic method has good reasoning ability and high diagnosed precision, overcomes the defects of the IEC three ratio method, and can precisely display all transformer fault problems.

The invention discloses a data processing method, a knowledge base reasoning method and a related device. The data processing method includes the step that the following modes are adopted to determine the rule dependence relationship of the scheduling order of rules in a knowledge base reasoning process: data are selected from a knowledge base so as to form a first data set; the first data set is reasoned according to the rules of the knowledge base; and the rule dependence relationship is determined according to the reasoning results of the rules and corresponding input data. With the data processing method, the knowledge base reasoning method and the related device of the invention adopted, application scenes and flexibility of the reasoning technology of the knowledge base can be improved, and the efficiency of the execution of the rules and overall reasoning performance can be improved.

The invention relates to a mapping knowledge domain relation inference method and device, computer equipment, and a storage medium. The method comprises the steps: extracting an entity set and a relation set in a mapping knowledge domain, and obtaining a current triple set; extracting a triple, matched with a predefined knowledge rule, from the current triple set, and obtaining a current trainingset; Training a current teaching model through a first loss function according to the current training set, enabling the output of the teaching model to be fit with the output of a current learning model, and obtaining a trained teaching model and an updated current triple set; training the current learning model through a second loss function according to the updated current triple set, enablingthe output of the current learning model to be fit with the output of the current teaching model, obtaining a trained current learning model and an updated current training set, carrying out the training repeatedly till a training result meets a condition of convergence, and obtaining a target learning model; obtaining a target entity, and carrying out the inference according to the target learning model to obtain an inference result.

The embodiment of the invention discloses a knowledge representation learning method and device, equipment and a storage medium, and relates to the technical field of natural language processing, deeplearning and knowledge graphs. A specific embodiment of the method comprises the following steps: sampling a knowledge graph sub-graph from a knowledge base; serializing the knowledge graph sub-graphto obtain a serialized text; and reading the serialized text according to the sequence on the knowledge graph sub-graph by using a pre-trained language model, and learning to obtain knowledge representation of each character in the serialized text. According to the embodiment, knowledge representation learning is oriented to entity and relationship representation learning in the knowledge base, semantic association of entities and relationships can be efficiently calculated in a low-dimensional space, the problem of data sparsity is effectively solved, and the knowledge acquisition, fusion and reasoning performances are remarkably improved. By utilizing the strong knowledge acquisition capability and context analysis capability of the pre-trained language model, the knowledge representation learned by the pre-trained language model can better represent the complex relationship in the knowledge base.

The embodiment of the invention provides a visual question-answering and visual question-answering model training method and device, electronic equipment and a computer storage medium. The visual question-answering model training method comprises the steps: receiving and inputting a training sample of a visual question-answering model through an input part, wherein the training sample comprises a sample image and a plurality of text questions corresponding to the sample image; through a feature extraction part of the visual question and answer model, performing feature extraction on the plurality of text questions to obtain a plurality of corresponding semantic vectors, and performing feature extraction on the sample image to obtain a corresponding image feature vector; in an expression learning part of the visual question and answer model, processing the image feature vector and the semantic vectors by using an attention mechanism to obtain an image feature expression vector and a question feature expression vector; finally, through an output part of the visual question and answer model, performing question result prediction according to the image feature expression vector and the question feature expression vector, and performing training of the visual question and answer model according to a question result prediction result.

The invention provides a dynamic reasoning network and method for multi-hop questions and answers, and the network comprises: a paragraph selector which receives paragraphs and questions, and selectssub-paragraphs related to answers of the questions from the paragraphs; the encoding module that is used for enhancing interaction between the problem and the sub-paragraphs by using collaborative attention, and calculating to obtain final vector representation of the sub-paragraphs and vector representation of the problem; the entity graph construction module that is used for constructing an entity graph; the dynamic reasoning module that is used for reasoning the entity graph, repeatedly reading texts to simulate the process of analyzing information by people and constructing a problem remodeling mechanism so as to repeatedly read problems and related important parts; and the answer prediction module that is connected with the encoding module, is connected with the dynamic reasoning module and is used for receiving the final vector representation of the sub-paragraphs and outputting to obtain four types of prediction. The network establishes a question remodeling mechanism, and the mechanism can repeatedly read questions to simulate the reading habits of people so as to improve the understanding and reasoning ability of the multi-hop reasoning question and answer model.

The invention discloses a visual question and answer method based on a combined relation attention network, aims at the problem that the existing visual question and answer method can only extract a simple visual relation, and innovatively constructs a self-adaptive relation attention module for fully extracting an accurate binary relation and a more complex ternary relation. The visual relationship between the relationship and the question can reveal deeper semantics, and the reasoning capability of the method when the method answers the question is enhanced. Meanwhile, the problem that an existing visual question and answer method cannot well fuse image features and position (relation) features of a target in an image is solved. According to the method, firstly, image features and position (relation) features of a target are extracted respectively, extraction of the image features of the target is independent of extraction of the relation features of the target, and then the two features are fused under guidance of a question, so that the two features are well fused together. By fully and accurately extracting the visual relationship and well fusing the image features and the relationship features, the accuracy of predicting the answers of the questions is improved.

The invention discloses a medical relationship extraction method and device, and the method comprises the following steps: carrying out the statistics of the occurrence frequency of a medical concept pair in a set time window from a medical electronic medical record, and obtaining two medical concept vectors in the medical concept pair; matching the two medical concepts with a knowledge base to obtain an incidence relation between the two medical concepts so as to construct a relation concept triple; mining a plurality of concept statements from a medical text set according to the relation concept triple; constructing a training sample set, wherein the training sample set comprises positive samples and negative samples, and each sample structure is composed of a relation concept triple, two medical concept vectors and a concept statement; training a fusion model by using the training sample set to obtain a trained fusion model; and performing medical relationship extraction by using the trained fusion model. According to the method, the relationship between the medical concepts can be continuously mined. Chapter titles are introduced to form concept statements, and effective training samples are increased.

The invention provides a social network representation method based on bidirectional distance network embedding, and belongs to the technical field of data mining and networks. The method comprises the following steps: firstly, reading nodes in a social relation network and encoding; secondly, reading a concerned and concerned relationship, respectively generating an upper text neighbor node sequence and a lower text neighbor node sequence with the window size of k for each node, and recording a directed distance from each neighbor node to the node; constructing a three-layer network embeddingmodel; learning by taking the node coding set as input, and continuously adjusting model hyper-parameters; and finally, taking the weight matrix of the hidden layer as a final network embedding result, and taking the vector representation of each row as the vector representation of the node. According to the method, the problems that the structure and topology information of the existing social relation network are inaccurate in representation, the capability of restoring the real social relation is low, network data cannot be effectively processed, and the development of events cannot be accurately and effectively controlled are solved. The method can be used for social network representation.

The invention provides a neural network model reasoning method and device, electronic equipment and a readable medium, and relates to the technical field of neural networks. The method comprises the following steps of preprocessing a to-be-identified picture by adopting an assembly line method; loading a pre-trained neural network model into a neural network reasoning acceleration engine; and reasoning the preprocessed to-be-identified picture by adopting the loaded neural network reasoning acceleration engine to obtain a reasoning result. Picture preprocessing is subjected to parallel processing through an assembly line method, CPU resources are fully utilized, the reasoning performance is improved, a neural network reasoning acceleration engine is adopted for reasoning the acceleration engine, and the reasoning speed of a neural network model is increased.

The invention discloses a deep reinforcement learning air combat game interpretation method based on a fuzzy decision tree, and the method comprises the steps: carrying out the air combat game througha trained deep reinforcement learning model, and obtaining a training set and a feature set; constructing a membership function of each feature in the feature set, and fuzzifying the features one byone to obtain a fuzzy feature set after fuzzification of the feature set; establishing a fuzzy decision tree according to the training set and the fuzzy feature set; pruning the fuzzy decision tree byminimizing a loss function of the decision tree; traversing all paths of the pruned fuzzy decision tree, wherein each path represents an air combat game rule; storing the input and output of the deepreinforcement learning model during air combat game as to-be-processed data, and inputting the to-be-processed data into the pruned fuzzy decision tree to obtain a corresponding air combat game rule,thereby completing air combat game interpretation. According to the method, the problems of poor interpretability and non-intuitive result of a deep reinforcement learning algorithm are solved.

The invention provides a multi-step prediction method and a multi-step prediction system. The multi-step prediction method comprises the steps of performing self-adaptive fuzzification processing to trend items of prediction elements and calculating self-adaptive membership degree vectors of the trend items; combining with observed values in the past to generate learning samples with fuzzification features according to the self-adaptive membership degree vectors of the trend items and establishing time series prediction models; and utilizing the time series prediction model to perform multi-step prediction. The self-adaptive prediction method with an embedded fuzzy set state and the self-adaptive prediction system can achieve prediction with high accuracy rate and low error rate.

The invention relates to a human body posture estimation method based on motion feature constraints. According to the method, a video is divided into a plurality of human body space-time windows, image features of each frame under the windows are extracted, human body motion features of the windows are extracted according to the image features contained in the windows, and then human body postureestimation of a single frame of picture is restrained through the motion features. According to the invention, the architecture of the full convolutional neural network is adopted, and the dynamic convolution is used to adaptively adjust the single-frame attitude estimation according to the situation information contained in the video, so that the common problems of motion blur, limb occlusion andthe like in a human body attitude estimation task can be better solved, the attitude estimation can be better carried out on the video, and the human body posture estimation accuracy in a video sceneis improved.

The invention relates to the technical field of computer vision, and discloses a face image quality evaluation method based on a lightweight regression network, and the method comprises the steps of collecting a face image data set; performing data preprocessing on the face image data set by using a face detection algorithm; utilizing a feature extraction algorithm to generate a quality score label, training, verifying and testing the deep learning regression network, and generating a face quality evaluation model; and performing quality evaluation on the face ID to be subjected to quality evaluation by utilizing the face quality evaluation model. According to the invention, the cosine similarity and the face confidence coefficient are used for marking the data, errors caused by manual marking are reduced, the marking speed is high, the lightweight deep learning network is used for regression of the quality score of the face image, the regression precision is guaranteed, the reasoningperformance of the face quality evaluation model is improved, the face image can be evaluated more comprehensively, and the recognition accuracy and the system operation efficiency of the face recognition system are improved.

The invention relates to the field of industrial Internet, and provides a port network intrusion detection method based on a Bayesian network, which comprises the following steps: S1, collecting, acquiring and preprocessing an abnormal port network flow data set to obtain a network flow characteristic set; S2, constructing and obtaining a Bayesian network model by utilizing the network data packetfeature set; S3, inputting a training set and training parameters of the Bayesian network model, and obtaining conditional probability parameters of the Bayesian network model by using a Bayesian theorem; and S4, detecting an input prediction set by using a conditional probability parameter and the Bayesian theorem to obtain a detection result. The invention discloses a network intrusion detection method based on a Bayesian network model. On the basis of a Bayesian network model, network intrusion detection is realized by modeling network traffic behaviors and characteristic attributes, and online dynamic adjustment can be performed on a detection model to cope with the change of a network environment, so that the accuracy of detecting and protecting network intrusion and the robustness of the model are improved, and finally, a remarkable effect is achieved.

The invention relates to a knowledge graph construction method fusing an inference engine, and belongs to the field of knowledge graphs. The method comprises the following steps: manually constructing a knowledge graph schema according to the actual condition of a service, accelerating the definition of entity attributes and relationships by utilizing an inference engine, importing semi-structured data to perform knowledge extraction, and cleaning and normalizing and disambiguating knowledge graph data; and then integrating rule design to carry out deep reasoning processing on knowledge to form a final rule knowledge base. According to the method, mass information is integrated and managed based on the structure data of the knowledge graph, rule reasoning is combined, the knowledge reasoning effect can be improved, and a more accurate reasoning result and visual display are obtained. According to the knowledge graph, multi-source heterogeneous data can be fused, the organization and relevance of the data are enhanced, and mass data can be conveniently stored and managed; wherein, an inference engine is introduced in the construction process of the knowledge graph to carry out rule design, so that deep inference and knowledge integration are effectively carried out on knowledge, the construction process of the knowledge graph is accelerated, and the data accuracy is improved.

The invention discloses an intention recognition method and device, equipment and a medium, and relates to a natural language processing technology. According to the specific implementation scheme, the method comprises steps of shielding each character and at least one first matching keyword in each pre-training sample according to a preset probability, wherein the first matching keyword is a word matched with an entity in a pre-established knowledge graph in the pre-training sample, the knowledge graph comprises entities related to each intention category; based on each shielded pre-training sample, carrying out pre-training on a pre-built language model; performing fine tuning on the pre-trained language model to obtain a target language model, and performing intention recognition based on the target language model. According to the embodiment of the invention, the knowledge graph is fused in the pre-training stage of the language model, so that the model can learn semantic representation of knowledge, the representation capability and reasoning capability of the model are enhanced, and the prediction accuracy of the model is improved.

The invention discloses an intelligent automobile single-target tracking method based on a twin network. After a template and an image to be detected are extracted through a twin network formed by a Mish-channel-dark network to obtain feature information corresponding to the template and the image to be detected, then the feature information is input into a similarity judgment branch and a qualityevaluation branch, finally, the similarity is determined through cross-correlation calculation, and template updating is completed. The Mish-channel-park network is composed of a mish convolution module, an m-Residual module, a channel attention module and a structural body. According to the method, the reasoning capability and the anti-interference capability of the tracking algorithm are improved, the channel attention module is added, the tracking precision of the network is remarkably improved, the time for retrieving the object when the target disappears and reappears is shortened, and the practicability of the network is effectively improved. The twin network designed by the invention has better performance, and can achieve higher speed only by lower configuration during deployment.

The invention discloses an automatic vector optimization method for width inconsistency of a deep learning framework compiler, which is based on a heterogeneous platform, and comprises the following steps that: S1: a front end of the framework compiler identifies a sub-graph capable of carrying out vector optimization in a calculation graph, S2: a middle end of the framework compiler fuses operators in the sub-graph capable of carrying out vector optimization marked in the step S15, S3, the rear end of the framework compiler carries out vector optimization with inconsistent widths on the bottom-layer IR obtained in the step S2 according to vector widths of a control core and a calculation core of the heterogeneous many-core processor; and S4, a code generation module of the frame compilerconverts the underlying IR after vector optimization obtained in the step S32 into a high-level language code specified by a user, and generates a platform target code after vector optimization through a basic compiler. The instruction set parallel performance of the deep learning load is further mined, and the vectorization degree of the deep learning load is improved, so that the reasoning performance of the deep learning load on the heterogeneous many-core platform is improved.

The invention discloses an improved computing method for resolving ontology concept semantic similarity based on semantic distance. The method comprises the following steps: computing the semantic distance d1 between the ontology concepts (g1, g2) through an initialized ontology concept module and computing a distance factor d2 based on the semantic distance so as to construct the semantic similarity sim1(g1, g2); constructing the semantic similarity sim2(g1, g2) according to the corresponding depth and density of two ontology concept nodes; and finally obtaining the concept semantic similarity sim2(g1, g2). Compared with the traditional method for resolving the semantic similarity based on the information theory method and the semantic distance method, the accuracy of the semantic similarity resolved by use of the method disclosed by the invention is higher; and the semantic distance computation has a certain theory preciseness; the obtained result is more accurate by integrating the information theory and the semantic distance method; the multi-inheritance problem of the node in the ontology tree is solved; this method for computing the semantic similarity is more close to the experience point of an expert on the quantization concept; the ontology inference effect is better improved, and more extensive application research value is realized.

The invention discloses a target pose estimation method and system based on an attention mechanism and Hough voting. The method comprises the following steps: acquiring a color image and a depth image; segmenting and cutting the color image to obtain color and depth image blocks of each target object; using two strategies to estimate the six-dimensional pose of a target object, for a three-dimensional rotation matrix, based on a feature extraction network of bidirectional space attention, performing robust feature extraction by utilizing target surface two-dimensional feature constraint, and then using a multi-layer sensing network to regress the three-dimensional rotation matrix of the target; and for a three-dimensional translation vector, reconstructing a target object point cloud and normalizing point cloud data, estimating a point cloud three-dimensional translation direction vector point by point by adopting a Hough voting network, finally establishing a translation center straight line set, and solving a space nearest point to obtain a target three-dimensional translation vector. According to the method, the rotation matrix and the translation vector are estimated respectively, the execution speed is high, and precision is high.

The invention provides a spine centrum information automatic recognition method and system based on a medical image, a terminal and a storage medium. The method comprises the steps that a spine part radiation plain film is acquired and preprocessed; determining standard annotation data according to annotation results of specialists on the spinal radiation plain films; inputting the standard annotation data into a preset deep learning network model for training to obtain a vertebral body segment recognition model; inputting the standard annotation data into a preset deep learning network modelfor training to obtain a vertebral body segment classification inference joint model; training the vertebral body segment recognition model and the vertebral body segment classification and inferencecombined model by adopting cross validation to obtain a cross training vertebral body segment classification and inference combined model; inputting the preprocessed to-be-labeled data into a vertebral body segment recognition model, carrying out cross training on a vertebral body segment classification inference joint model, and predicting to obtain a segment frame, a segment type and a positionnumber of a vertebral body; and prediction and recognition of the types and the position numbers of the vertebral body segments can be realized.

The invention relates to a time sequence knowledge graph reasoning method and device based on an attention mechanism, computer equipment and a storage medium. The method comprises the steps of obtaining neighborhood information of each entity in each time period by constructing a knowledge graph snapshot of each time period in a time sequence knowledge graph; aggregating neighborhood information corresponding to all relations of the plurality of entities through a neighborhood aggregator to obtain neighborhood feature representation of each entity; an attention weight matrix containing multi-head information is determined through a time sequence event encoder based on an attention mechanism according to the neighborhood feature representation of a target entity at the current moment and the neighborhood feature representation of the target entity at the historical moment, and then a time entity representation sequence of the historical information is selectively concerned; obtaining the implicit vector representation of the target entity updated by the time sequence event encoder at the current moment; and performing coding scoring on the time sequence event encoder according to the implicit vector representation through the feedforward neural network and the multi-classification-layer network to realize time sequence knowledge graph reasoning.

The invention relates to an intelligent control system for plant growth factors and a control method. The intelligent control system comprises a plurality of growth factor information collection units, a human-computer interaction module, an inference engine, a unified control coordination module and an external electrical execution control unit, a signal output end of each growth factor information collection module and a signal output end of the human-computer interaction module are connected with a signal input end of the inference engine, a signal output end of the inference engine is connected with an input end of the unified control coordination module, an output end of the unified control coordination module is connected with the external electrical execution control unit, and a signal output end of the external electrical execution control unit is connected with a plant growth factor control valve. The intelligent control system can improve the control quality for various growth factors in the plant growth environment.

The invention provides a data processing method for sketch retrieval. A knowledge preserving mechanism is introduced; on the basis of a domain loss function, a triple loss function and a semantic lossfunction, a knowledge preserving loss function is additionally constructed, and all the loss functions are used for training; in the knowledge migration process of the pre-training model, new knowledge of a target domain can be learned, knowledge learned from a source domain can be kept, and sketch retrieval precision is improved.

The invention discloses a knowledge graph construction method integrating teaching feedback and learned understanding. The method comprises the following steps: 1) data acquisition: converting behaviors of students in a classroom into a text for describing the understanding degree of the students on current knowledge; 2) named entity recognition and relation extraction: performing entity recognition and relation extraction on classroom text data at the same time to obtain a related entity triple; 3) embedding the entity and the relation, generating information of neighbor nodes by a weighted graph convolutional network, learning richer semantic expressions of the entity and the relation to form a final entity embedding expression; 4) scoring alternative tail entities in the knowledge graph triple through a multi-scale convolutional neural network, selecting the tail entity with the highest score as a reasoning result, further reasoning implicit knowledge, and updating the knowledge graph; wherein the alternative tail entities are all entities updated in the step 3). According to the invention, a specific knowledge graph of each student can be constructed, and learning and teaching evaluation feedback can be carried out.

The invention provides a part assembly quality digital detection system and method based on machine learning. The system comprises an image collection module which is used for obtaining the image data of at least one part, an image analysis module which is used for performing image information analysis on the image data to obtain a plurality of quality feature elements corresponding to the image data, and a quality detection module which is used for constructing a target quality detection model based on particle swarm parameter optimization, and performing drilling and / or riveting quality detection on the part by using the target quality detection model according to quality characteristic elements to obtain a quality detection result of the part. According to the part assembly quality digital detection system and method based on the machine learning of the invention, the quality detection of hole making and riveting in the part assembling process is realized in a process-oriented, modularized, automatic and efficient manner, the accuracy of a quality detection result is improved, and the manpower and time cost is greatly saved.

The invention discloses an equipment fault diagnosis method based on collaborative case-based reasoning and semantic model-based reasoning, which comprises the following steps of S1, collecting cases,and constructing a case library; S2, combining fuzzy logic and knowledge extracted by an FMEA analysis method, and constructing a fault diagnosis ontology model by a fuzzy ontology development methodology process; and S3, on the basis of knowledge obtained in the ontology model, generating a corresponding SWRL rule in combination with expert experience, and performing conflict detection on the generated SWRL rule to form a fault diagnosis rule base; S4, performing fault detection according to the constructed fault diagnosis ontology model, rule base and case base. According to the method, onthe basis of combination of CBR and RBR, knowledge extracted by fuzzy logic and an FMEA analysis method is fused into construction of the ontology model, so that the integrity of the ontology model isimproved, and definition of uncertain knowledge is more reasonable; meanwhile, diagnosis rules are constructed by utilizing shallow knowledge and deep knowledge, and the integrity and accuracy of a rule base are improved, so that the reasonability of a diagnosis framework is improved.