177results about How to "Improve model performance" patented technology

A statistical classifier is constructed by estimating Naïve Bayes classifiers such that the conditional likelihood of class given word sequence is maximized. The classifier is constructed using a rational function growth transform implemented for Naïve Bayes classifiers. The estimation method tunes the model parameters jointly for all classes such that the classifier discriminates between the correct class and the incorrect ones for a given training sentence or utterance. Optional parameter smoothing and / or convergence speedup can be used to improve model performance. The classifier can be integrated into a speech utterance classification system or other natural language processing system.

The invention discloses a reinforcement dual-channel sequence learning-based dialog reply generation method and system. The method comprises the following steps of: (1) modeling a context to obtain acontext semantic vector; (2) carrying out combined learning on a current dialog and the context semantic vector by utilizing an encoder so as to current a current dialog vector and an encoder vector;(3) inputting the context semantic vector and the current dialog vector into a decoder so as to obtain a first channel dialog reply draft and a decoder vector; (4) inputting the encoder vector, the decoder vector and the first cannel dialog reply draft into an embellishing device to carry out embellishing, so as to generate an embellished dialog reply of a second channel; (5) optimizing a target function by utilizing a reinforcement learning algorithm; and (6) ending model training and generating and outputting a dialog reply. By utilizing the method and system, dialog generation models can grasp global information more deeply, and replies more according with dialog scenes and having substantial contents can be generated.

The invention discloses a cross-domain federal learning model and method based on a value iteration network, and the model comprises a data preparation unit which is used for taking a path planning field of a grid map as a training environment, and taking observation states of two different parts in the same map as inputs of the two fields of federal learning; an Federated-VIN network establishingunit is used for establishing a Federhead-VIN network structure based on a value iteration network; the VIN network structure constructs the full connection of the value iteration modules of the source domain and the target domain, and defines a new joint loss function about the two domains according to the newly constructed network; the iteration unit is used for carrying out forward calculationon the VI modules in the two fields in the training process, and value iteration is achieved for multiple times through the VI modules; and the backward updating unit is used for carrying out backward calculation to update the network parameters, and carrying out backward updating on the VIN parameters and the full connection parameters in the two fields according to the joint loss function.

The invention discloses a method for predicting the remaining service life of complex equipment based on a combined depth neural network. The main steps are as follows: acquiring multi-sensor data ofcomplex equipment; Obtaining effective measurement data by feature selection; obtaining A plurality of slice samples by preprocessing; Establishing the neural network regression model which combines the attention mechanism and depth neural network; The slice samples and their corresponding labels are inputted into the neural network regression model to train the neural network regression model offline. inputting The slice samples of multi-sensor data to be predicted into the trained neural network regression model, and the remaining service life of complex equipment is obtained. Considering the data characteristics of the multi-sensor signal, the invention fully excavates the local characteristics and the time sequence information in the data, has high prediction accuracy and wide applicability, and can be widely applied to various pieces of complex equipment.

An apparatus and associated method are utilized for monitoring an operation of a system characterized by operational parameters. A non-parametric empirical model generates estimates of parameter values in response to receiving a query vector of monitored parameters for a model characterizing the system. A distance estimation engine (a) determines robust distances between the query vector and each of a set of predetermined historical vectors for the non-parametric empirical model based on an implementation of an elemental kernel function; (b) determines weights for the monitored parameters based on the robust distances; and (c) combining the weights with the predetermined historical vectors to make predictions for the system.

The embodiment of the invention discloses a model training method, device and equipment, and the method comprises the steps: obtaining a pre-trained dialogue generation model which comprises an encoder and a decoder; constructing a dialogue system evaluation model, wherein the dialogue system evaluation model takes questions and replies as input and takes scores corresponding to the replies as output; initializing parameters of an encoder in the dialogue system evaluation model according to the parameters of the encoder in the dialogue generation model; and according to the first training sample set, training the initialized dialogue system evaluation model to obtain a dialogue system evaluation model meeting a training ending condition, each training sample in the first training sample set comprising a question, a reply and a label score corresponding to the reply. The dialogue system evaluation model obtained by training through the method can evaluate the reply quality of the dialogue system from the perspective of semantic correlation, and the reliability of dialogue reply evaluation is improved.

Frequency features to be used for binary classification of data using a linear classifier are selected by determining a set of hypotheses in a d-dimensional space using d-dimensional labeled training data. A mapping function is constructed for each hypothesis. The mapping functions are applied to the training data to generate frequency features, and a subset of the frequency are selecting iteratively. The linear function is then trained using the subset of frequency features and labels of the training data.

The invention relates to a tree modeling method based on skeleton point cloud. In the method, scaffold branches of the tree and outline of the crown which are sketched manually are taken as input automatic constructed tree model, comprising the following steps: extracting two-dimensional skeleton from the sketched strokes through pixel analysis; constructing a three-dimensional skeleton point cloud with two two-dimensional skeletons; expanding the two-dimensional skeleton into a three-dimensional scaffold branches skeleton under the guidance of the three-dimensional point cloud information; and constructing twigs and leaves based on the outline of the crown. The invention has easy application, simple algorithm and high modeling efficiency, and can create tree models with sense of reality. The modeling results of the invention have significant application values in fields of computer games, three-dimensional films, network roaming, urban landscape design and the like.

Disclosed are a method and apparatus for processing a non-linear signal in an electronic device. The electronic device includes an audio input module, an audio output module, and a processor. The processor is configured to identify a first audio signal that will be outputted through the audio output module, provide a first signal into which the first audio signal is processed and a second signal into which the first audio signal is processed, and output the first audio signal through the audio output module, and acquire an external audio signal comprising the first audio signal of the electronic device, and acquire a first output value through a first input channel of an audio filter, and acquire a second output value through a second input channel of the audio filter, and provide a second audio signal, based at least on a first difference value between the magnitude value corresponding to the first frequency of the external audio signal and the first output value and a second difference value between the magnitude value corresponding to the second frequency of the external audio signal and the second output value.

The invention provides a multi-task classification model training method and device, and a multi-task classification method and device. The multi-task classification model training method comprises the following steps: inputting preset information into a pre-training model, wherein the preset information comprises a plurality of information units; calling a parameter sharing layer, performing global vector representation processing on each information unit, and determining a global semantic representation vector of each information unit; calling a plurality of classifiers, performing classification processing on the preset information according to each global semantic representation vector, and determining a classification prediction result of the preset information; based on the classification prediction result, the first quantity, the second quantity and the labeling result, calculating to obtain a loss value; and under the condition that the loss value is within a preset range, taking a target pre-training model obtained by training as a multi-task classification model. According to the multi-task classification model training method, a good multi-task classification model can be obtained on the basis of a small amount of training data, and only a small amount of annotation training data needs to be added under the condition that a new task exists, and the annotation cost can be reduced.

The invention discloses a visualization method and device for a random forest model and a storage medium, and relates to the technical field of machine learning, and the method comprises the steps ofscreening a target training sample meeting a preset condition from a training sample set corresponding to each decision tree of the random forest model, so as to form a target training sample set forconstructing a classification tree; obtaining the variable importance degree of each characteristic variable in each decision tree, and carrying out descending sorting on all the characteristic variables according to the variable importance degrees; according to the target training sample set and all the feature variables after descending sorting, starting from a root node of the classification tree, optimal feature variables and optimal segmentation values corresponding to all the nodes in the classification tree are sequentially determined by taking the Gini coefficient as a splitting rule,so that the classification tree is constructed; and generating a tree-shaped visual graph corresponding to the classification tree and outputting the tree-shaped visual graph. According to the invention, the decision process of the random forest model can be visually displayed, and the interpretability of the model is improved.

The invention provides an image processing method and device based on an adversarial generation network. The method comprises the steps of S1, obtaining a first face sample image; obtaining a first face multi-attribute condition; S2, inputting the first face sample image and the first face multi-attribute condition into a trained generative adversarial network to obtain a first synthetic image; and step S3, taking the first synthetic image as a face image conforming to the first face multi-attribute condition, and outputting the face image.

The embodiment of the invention discloses a similarity determining method and device based on an individualized deep neural network. The method includes the steps of obtaining an inquiry text and user individualized information input by a user, wherein the user individualized information is determined according to the historical search behavior of the user or the attribute information of an intelligent terminal held by the user; conducting deep neural network processing on the inquiry text, search items and the user individualized information, and determining the similarity between the inquiry text and the search items according to the deep neural network processing result. According to the technical scheme, the user individualized information is fused in the deep neural network learning, the model effect of traditional semantic similarity determination is improved, and therefore the accuracy of the similarity between the inquiry text and the search items is improved.

The invention provides a network training method and device, an action recognition method and device, equipment and a computer readable storage medium. The method comprises the following steps: updating model parameters of a pre-training model by utilizing a first sequence data set of a human skeleton point sequence and a visual angle label corresponding to each piece of first sequence data in thefirst sequence data set; initializing model parameters of a human body action recognition model based on the updated model parameters of the pre-trained model; wherein the pre-training model and thehuman body action recognition model have feature extraction networks with the same structure; and updating model parameters of the human body action recognition model by utilizing the second sequencedata set of the human body skeleton point sequence and the action category label corresponding to each second sequence data in the second sequence data set to obtain a trained human body action recognition model. Through the method and the device, the action recognition precision of the human body action recognition model can be improved, the model training time can be reduced, the dependence on strong annotation data can be reduced, and the manual workload is further reduced.

The invention mainly discloses a system and a method for monitoring a short message reaching effect. The technical scheme is as follows: the system for monitoring the short message touch effect is composed of a visualization system, a short chain system and a distributed data analysis system. The method for monitoring the short message touch effect comprises the following five steps of: S1, task initialization, S2, mobile phone number coding, S3, generating of a dynamic short chain; s4, collecting of short link skip logs; s5, log data analysis. The system and the method are based on an application scene of short message marketing, can effectively monitor a click condition of a short link in the short message copy after the marketing short message is sent, and can analyze which mobile phonenumber users click links in the short message copy. On one hand, full-process data conversion of the marketing scene is facilitated, and on the other hand, a model optimizer can conveniently optimizethe model effect by collecting positive sample users, so that the efficiency and the economic benefits in the marketing scene are remarkably improved.

A predictive modeling system and method that improves revenue management for a cargo business, preferably the air cargo business, by bridging a bidding stage and a decision stage by jointly learning dual predictive models, wherein it leverages the intrinsic co-clusters of originations and destinations (OD) to enable information sharing among different OD pairs. The predictive modeling method effectively leverages the block structure of the OD pairs thus increasing revenue.

The invention relates to a tree modeling method based on the depth search, and the method is to automatically build a tree model by taking a tree main branch and a tree crown outline drawn by hands as input. The method comprises the following main steps of: extracting two-dimension frameworks from drawn strokes through pixel analysis; directly converting the two two-dimension frameworks into a three-dimension main branch framework through the depth search; and building withes and leaves on the basis of restriction of the tree crown outline. The method can efficiently build the tree model which has the reality sense and accords with input restriction by simple algorithm. The modeling result obtained by the tree modeling method has the significant application value in fields of computer games, three-dimension movies, network roaming, city view design and the like.

The invention discloses a model processing method and system, terminal equipment and a server. According to the method, the technical concept that a to-be-adjusted model (the model is constructed by use of model adjustment information sent by the server by use of at least one set of terminal equipment) acquired from the server is adjusted by use of data of a terminal on the terminal equipment, model adjustment information corresponding to the adjusted model is sent to the server, so that the server adjusts the model on the basis of the model adjustment information from the terminal is proposed. By use of the scheme, model training by use of terminal data can be realized, the trained model has a better model effect with no need of uploading terminal data to the server by the terminal equipment, and the privacy of user data in the terminal equipment is guaranteed.

The embodiment of the invention discloses a labeled sample determination method and device based on artificial intelligence, equipment and a storage medium, and the method comprises the steps: obtaining a sample pair which comprises an unlabeled sample and a labeled sample; taking the unlabeled sample and the labeled sample in the sample pair as two paths of inputs of a sample evaluation model respectively to obtain an output result of the sample evaluation model; wherein the sample evaluation model is used for measuring the similarity between two paths of input samples; determining the availability of unlabeled samples in the sample pair according to the output result; and when the availability meets a preset condition, determining an unlabeled sample in the sample pair as a to-be-labeledsample. According to the method, paired learning is introduced into a sample selection process, feature extraction and learning are carried out on an unlabeled sample and a labeled sample by utilizing a sample evaluation model when the labeling value of the unlabeled sample is measured, and the labeling value of the unlabeled sample is measured based on the inter-domain difference of the unlabeled sample and the labeled sample.

Embodiments of this application disclose a classification model training method, a classification method, a device, and a medium. An initial classification model is first trained by using a first sample set including a large quantity of first samples, to obtain a pre-trained model, each first sample including a social text and an emoticon label corresponding to the social text; and the pre-trained model is then trained by using a second sample set including a small quantity of second samples, to obtain a social text sentiment classification model that uses a social text as an input and use a sentiment class probability distribution corresponding to the social text as an output. In this method, the model is trained by combining a large quantity of weakly supervised samples with a small quantity of supervised samples, to ensure that the model obtained through training has better model performance without increasing manually labeled samples.

The invention relates to the technical field of image processing, and particularly relates to a three-dimensional facial expression base establishing method and device, a computer readable storage medium and electronic equipment. The method comprises the steps that target key points of a target face image in input data are extracted, a camera projection matrix is determined according to the targetkey points and initial key points, corresponding to the target key points, in an average face model, and the average face model is a reference three-dimensional face model obtained according to facemodel data set processing; the camera projection matrix is used to calculate a target three-dimensional face model corresponding to the target face image according to the target key point, the averageface model reference three-dimensional face model and the initial key point; and a target three-dimensional facial expression base corresponding to the target human face image is generated by using the target three-dimensional human face model. According to the technical scheme, the problems that in the prior art, an obtained expression base model is poor in effect and large in calculated amountcan be solved.

The invention discloses a deep learning sample enhancement system, comprising a video module used for recording and providing a video sequence, a detector used for obtaining an optimized SSD network from the video sequence, a sample module including original samples of labeled data, a sampling module used for carrying out sampling detection and statistical analysis on the video sequence, a screening module, and a labeling module. The invention further discloses an operation method of the deep learning sample enhancement system. Through the deep learning sample enhancement system, training samples are automatically selected, the diversity and complexity of training samples are enhanced, the redundancy of training samples is reduced, and the training effect and generalization ability of thealgorithm are improved. Moreover, the operation method of the deep learning sample enhancement system can directly and significantly improve the model effect without algorithm-level optimization, andcan reduce the invalid workload of image labeling.

The subject matter disclosed herein provides methods, apparatus, systems, techniques, and articles for false positive reduction in abnormality detection models. A date and time of a first transaction by a transaction entity and associated with a transaction characteristic can be stored. Data representing subsequent transactions associated with the transaction characteristic can be stored. A history marker profile specific to the transaction characteristic and transaction entity can be generated and can include the transaction characteristic, the date and time of the first transaction, and maximum and mean abnormality scores. A date and time of a current transaction can be determined. A current abnormality score for the current transaction can be received. A tenure of the observed transaction characteristic can be computed. The current abnormality score can be recalibrated from the transaction entity abnormality detection system according to the maximum, mean, and current abnormality scores and a length of the tenure.

An event subject identification method comprises the following steps: identifying an entity in a target text by adopting a predetermined entity identification model; marking the identified entity in the target text by adopting a first predetermined symbol to obtain a marked target text; obtaining an embedded vector of each character in the target text according to the marked target text; inputtingthe obtained embedded vector of each character in the target text into a named entity recognition prediction model to obtain an output label corresponding to each character in the target text; and according to the obtained output label corresponding to each character in the target text, identifying an event main body in the target text. According to the invention, the identification accuracy canbe improved.

The invention provides a high-dimensional data feature selection method based on a graph neural network and spectral clustering. The method comprises the steps: taking each gene as a node to establisha gene relation graph structure model, taking gene correlation data as side information to be added into a gene relation graph, taking a graph neural network model for obtaining feature vector representation of the nodes, and after the feature vector representation of each node is obtained, starting a link prediction stage, generating a new edge based on the genetic relationship graph, and obtaining a new genetic relationship graph; finally, selecting a node with the highest weight from the new genetic relationship graph based on spectral clustering to serve as a feature node. According to the invention, the finally selected gene has small redundancy, a good model effect is achieved, and the interpretability of the biological angle is supported.

Examples of the present invention include air charge estimation models using linear programming support vector regression (LP-SVR). Air model parameters may be updated during operation of a vehicle, allowing the air model performance to be improved in the presence of part-to-part variation and part aging.

The invention provides a pedestrian re-identification method and system fused with part texture three-dimensional mapping, a medium and a terminal, and the method comprises the following steps: extracting the pedestrian texture information of a pedestrian image, and obtaining a pedestrian texture map containing the pedestrian texture information; preprocessing the pedestrian picture and the pedestrian texture map; and training a pedestrian re-identification network model based on the pre-processed pedestrian image and the pedestrian texture map so as to perform pedestrian re-identification based on the trained pedestrian re-identification network model. By constructing the pedestrian re-identification feature extraction method with attitude robustness and the deep learning network taking the features as auxiliary data, the change of pedestrian attitude and environment can be effectively coped with, an attention mechanism is combined, the information fusion under different spatial features is realized, and the pedestrian re-identification efficiency and accuracy are improved.

The embodiment of the invention discloses a real-time cloth defect detection method and system based on deep learning, and the method comprises the steps: 1, collecting different types of cloth defect images, and constructing a defect data set; 2, performing data expansion firstly, and then performing data expansion by means of a generative adversarial network; 3, carrying out labeling processing on the expanded defect data set; 4, constructing a deep learning target detection network to perform cloth defect detection; 5, training a cloth defect detection network; and 6, capturing images of the cloth in real time by using a camera, inputting the captured images into the trained cloth defect detection network, judging whether defects exist in the images, determining the types of the defects, positioning the defects, and finally storing a result into an output file. According to the method, manual design of features can be omitted, the robustness of a defect detection system is improved, the detection performance is greatly improved, manpower can be liberated, and the intelligent degree of the textile industry is further improved.