61 results about "Score" patented technology

The technology disclosed relates to machine learning based anomaly detection. In particular, it relates to constructing activity models on per-tenant and per-user basis using an online streaming machine learner that transforms an unsupervised learning problem into a supervised learning problem by fixing a target label and learning a regressor without a constant or intercept. Further, it relates to detecting anomalies in near real-time streams of security-related events of one or more tenants by transforming the events in categorized features and requiring a loss function analyzer to correlate, essentially through an origin, the categorized features with a target feature artificially labeled as a constant. It further includes determining an anomaly score for a production event based on calculated likelihood coefficients of categorized feature-value pairs and a prevalencist probability value of the production event comprising the coded features-value pairs.

Methods and systems for determining the importance of each of the variables, or combinations of variables, that contribute to the overall score generated by a predictive statistical model are presented. In a specialized case, for each variable in the model, an importance is calculated based on the calculated slope and deviance of the predictive variable. In a more general case, for each variable in the model, an importance is calculated based on setting that variable to have the average value for the data set, and then calculating the change in score. The totality of variables (or combinations thereof) is then ranked by the Δscore, or a magnitude of it, such as |Δscore|.

A method and system for analyzing time series data. In an embodiment, a loop is executed and terminated upon a specified maximum number of iterations of the loop being performed or upon a difference between scores in successive iterations of the loop not being greater than a specified tolerance, wherein the score in each iteration is calculated as function of an absolute value of a difference between respective cumulative probability values of first and second cumulative probability distributions which are generated from respectively first and second time series data sets. In an embodiment, time series data is processed in a sequence of time periods, wherein a combined cumulative probability distribution is generated in each time period by combining a cumulative probability distribution of new time series data with previously combined cumulative probability distribution data according to a ratio of the number of new to previous observed values.

The invention provides a multi-model malicious code detection system based on reliability probability interval. Each machine learning detection model corresponds to a distribution of the underlying data, and various threshold-based detection models can be integrated into the statistical platform, so that the distribution of the semantic code data is detected from the multi-angle view, and the model degradation problem caused by the concept drift is relieved. The detection system changes the prediction mode of 0 or 1 of the existing machine learning detection model, calculates the score based on the existing detection model, carries out statistical analysis, and establishes a isotonic regression function for the score distribution of the sample and the label of the sample. For an unknown sample, according to the score given by the existing detection model, the calculated isotonic regression function is input, the reliability probability interval of a certain label can be given, and theprobability interval can relieve the problem of over-fitting of the fixed threshold to the training data set, the adaptive ability of the detection model to the current dynamic data is improved, and the concept drift phenomenon is found in advance.

The invention belongs to the technical field of knowledge maps, and discloses a knowledge map information representation learning method, a system, equipment and terminal, and the knowledge map information representation learning method comprises the steps: carrying out the preprocessing according to a path constraint resource distribution method; calculating the reliability of all paths, and outputting the reliability to a training set and a test set; initializing the model and setting parameters; generating a triple according to an iterator, and randomly replacing head and tail entities; calculating a loss function of the triple according to the score function; calculating a loss function of an additional path according to the path reliability; performing parameter optimization by using an Adam method; and performing model verification by using entity prediction and relation prediction. According to the method, rich path information contained in the knowledge graph is considered, the modeling effect of entities and relationships is improved, the modeling of the relationships can be optimized by inputting vectors into a complex plane and using rotation to represent the vectors, and the method can be used for link prediction and recommendation systems.

In a system for enabling configuration of an ensemble of several solvers, such that the ensemble can efficiently solve a constraint problem, for each one of several candidate configurations, an array of scores is computed. The array corresponds to a statistical parameter related to a problem solution, and the computation is based on, at least in part, a set of features associated with the problem. One candidate configuration is assigned to a solver, and based on the array of scores associated with that candidate configuration the same or a different candidate configuration is assigned to a another solver. A system for dynamically reconfiguring an ensemble of solvers obtains runtime data from several solvers, and a new configuration is determined by applying a machine learning and / or heuristic analysis procedure to the runtime data. The configuration of a solver may be updated according to the new configuration while that solver is running.

The invention relates to the technical field of artificial intelligence, and provides a machine learning model optimization effect evaluation method and device, a terminal and a storage medium, and the method comprises the steps: segmenting a historical sample data set into a plurality of test sample data sets; predicting the plurality of test sample data sets by using a plurality of machine learning models to obtain a plurality of prediction values; constructing an evaluation function based on the business indexes and the technical indexes, and calculating a plurality of evaluation scores according to the plurality of prediction values and the evaluation function; constructing an evaluation matrix based on the plurality of evaluation scores; and evaluating an optimization effect value ofeach machine learning model according to a preset optimization effect evaluation model and the evaluation matrix. Each machine learning model subjected to multi-iteration optimization can be evaluatedby combining the business indexes and the technical indexes.

An analytics processing system for generating a decision-making algorithm based on a prescribed set of pre-defined data points describing one or more characteristics of an entity to achieve an objective. The objective is modelled by an underlying base algorithm. The system includes a user interface to receive initial data concerning the objective from a client. It also includes a decision engine comprising a pipeline of modules including a validation module, a retrospect module, a refinement module and a comparison module. These modules perform the following functions in iterative phases:(i) derive a base algorithm to best match a candidate-entity to a known model having regard to the initial data;(ii) input select data related to the candidate-entity from a source of data;(iii) produce an output score being a function of the base algorithm;(iv) derive a predicted probability from the output score;(iv) compare the predicted probability with an actual outcome based on actual data derived from the source data at a subsequent period of time relative to the select data;(v) generate a variant of the base algorithm based upon the results of the comparison;(vi) create a new decision-making algorithm based on the variant; and(vii) periodically perform the aforementioned steps using the new decision-making algorithm as the derivative of the base algorithm after the prescribed period of time.The select data is prescribed to characterise a plurality of pre-defined data points associated with the base algorithm selected to provide a qualitative measure of performance to achieve the objective; The output score is derived from applying the select data for each data point and running the base algorithm thereon. The predicted probability is a weighted variable of the data points that is used to predict the likelihood of the objective being achieved.

Methods, computer-readable media, software, and apparatuses may retrieve, from an industry standard setting scoring system and for a vulnerability, a temporal score based on a pre-revision version of a scoring system, and predict, based on a machine learning model and based on the temporal score for the vulnerability, an updated temporal score based on a post-revision version of the scoring system. A mitigating factor score, indicative of a mitigation applied to the vulnerability by an enterprise organization, may be determined. A risk score may be generated for each vulnerability, as a composite of the updated temporal score and the mitigating factor score. The risk scores for vulnerabilities in a collection of vulnerabilities may be aggregated to determine an enterprise risk score for the enterprise organization. In some instances, the enterprise risk score may be displayed via a graphical user interface.

Accordingly, embodiments of the present invention advantageously account for item discrimination in a single-parameter logistic model used for measuring a test-taker's ability and an item's difficulty. Accounting for item discrimination improves the reliability of a test without increasing the number of test items. To account for item discrimination in a single-parameter logistic model, this invention uses the correlation between item response (correct or incorrect) and total test score or other measure of test-taker ability to obtain a Bayesian estimate of the correct-response probability (between zero and one). This correlation is a measure of item discrimination. The numerator in the formula for this correlation contains the difference between the average test score of test-takers who got the item right and the average test score of test-takers who got the item wrong.

The invention provides a machine learning-based football match victory and defeat real-time prediction method, which comprises the following steps of: firstly, according to statistical historical data and real-time event data recorded in a football match, respectively establishing a scoring model to obtain historical features and real-time features; historical features and real-time features are fused and screened in a stacking mode, and then the average number of goals, the average number of shoots, the score ranking and the like are used for expansion to complete feature generation. A football match result is predicted in real time by establishing a graph convolution deep neural network model. Compared with an algorithm which only uses historical data to carry out match prediction, the accuracy is improved.

The invention discloses a score card modeling method. The method comprises the steps of variable determination and screening, intermediate derivative variable logistic regression and model verification. In the variable determination process, a WOE variable and a Recode variable are introduced at the same time, and then the variables are constructed and screened from the two aspects of stability and accuracy; according to the method, variable clustering analysis based on factor analysis and a PCA algorithm is introduced, independent variables are clustered according to principal components of the independent variables, and a plurality of variables are selected from each cluster, so that the explanation degree and coverage of dimensions are reserved to the maximum extent. According to the score card modeling method, the modeling of the score model is completed, human intervention is only needed in several specific steps, the existing modeling time is shortened from one month to three days by adopting the semi-automatic modeling method, and the technical problem that the modeling time is long in the prior art is solved.

The invention discloses a risk-considered benefit distribution method and device for a virtual power plant. The method comprises the following steps: acquiring historical measured data and historicalpredicted data of each participant in the virtual power plant; obtaining a corresponding prediction error and a probability density function according to calculation; fitting a probability density function to obtain a probability distribution curve; integrating the probability distribution curve to obtain an accumulated probability distribution function; obtaining a prediction precision score according to the prediction error and the cumulative probability distribution function; setting a weight for each prediction precision score to enable the sum of all the weights to be 1; according to theprediction precision score and the corresponding weight, calculating to obtain a risk factor; normalizing all the risk factors to obtain corrected risk factors; and correcting the shape value according to the correction risk factor to obtain the distribution income corresponding to each participant. According to the invention, the risk factor of each participant is introduced into the shapley value to ensure the fairness of allocation and the stability of alliance.

The invention discloses a mixture Gaussian probability density weighting based grading model and system. According to the mixture Gaussian probability density weighting based grading system, a distribution function can be fit through mixture Gaussian distribution according to the dispersion degree of grading by experts and the weight can be reasonably arranged according to the probability density of the function. Compared with the prior art, the mixture Gaussian probability density weighting based grading system has the advantages of 1 being capable of fitting an actual distribution function well and guaranteeing the validity of a fitting result in theory; 2 fairly reflecting the level of an evaluation level due to the fact that the grading weight of the experts can be automatically adjusted according to fitting distribution; 3 not damaging the grading data information of the experts due to a grading system. After grading data is obtained, parameter values of the mixture Gaussian distribution can be obtained through an EM algorithm, then weight is given for every subdata through the probability density function, the final score of a player is a weight sum of scores of the experts, and finally a grading model program is embedded to ARM board hardware to achieve interactive operation. The mixture Gaussian probability density weighting based scoring model is widely applied to a game with many judges.

Certain aspects of the present disclosure provide techniques for generalized metric for machine learning model evaluation for unsupervised classification including: for each unsupervised machine learning model of one or more unsupervised machine learning models: generating a first set of synthetic inputs for the model of the one or more unsupervised machine learning models; providing the first set of synthetic inputs to the model trained to output a prediction for each input of the first set of synthetic inputs, wherein the prediction indicates whether the input is of a first class; identifying, based on an output of the model, a second set of synthetic inputs predicted to be of the first class; determining, based on a set of expected normal inputs for the model and the second set of synthetic inputs, an accuracy score for the unsupervised machine learning model; and providing the accuracy score for display.

The invention provides a control method for evaluating the data value of each participant in federated learning. The control method comprises the following steps: a, determining a mathematical expectation V that one participant is added into a joint model in different sequences; b, determining the sum W of contribution scores of all gradients added into the joint model by the participant; c, determining the participant data value based on the mathematical expectation V and the contribution score sum W; and d, repeatedly executing the steps a to c until the data values of all participants are determined. According to the method, two angles influencing the data contribution of each participant, namely the influence of the actual value and marginal effect of the data of each participant on the contribution of each participant, are comprehensively considered, and the data contribution of each participant can be measured more fairly. Meanwhile, the problem that data dimensions or numbers of participants are different is considered, and the probability that the participants enter the joint model in different sequences is continuously updated according to the thought of sampling statistics. The method is simple in process, convenient to use, powerful in function and extremely high in value.

The invention relates to artificial intelligence, and discloses a text abstract generation method and device, computer equipment and a readable storage medium. The method comprises the steps: obtaining to-be-processed text information, and converting the text information into a word vector; inputting the word vector into a pre-trained preset neural network model through a cluster search algorithmto obtain a candidate abstract set of the text information and a log-likelihood probability value of each candidate abstract in the candidate abstract set; obtaining a target redundancy score of eachcandidate abstract in the candidate abstract set; obtaining a reference score of each candidate abstract according to the target redundancy score and the log-likelihood probability value of each candidate abstract; and selecting the abstract of which the reference score is greater than a preset reference score from the candidate abstracts as an abstract corresponding to the text information. In addition, the invention also relates to a blockchain technology, and the reference score can be stored in the blockchain. According to the text abstract method provided by the invention, the redundant words for automatically generating the abstract can be optimized, so that the readability of automatically generating the text abstract is improved.

The invention discloses a distribution network fault identification method and a related device. The method comprises the following steps of constructing an input matrix according to a switching action of an input sample, setting a label for the fault type of the output sample according to the input sample, and constructing an output matrix according to the label of the output sample, acquiring a trained neural network model by performing forward transmission and back propagation training on the neural network model through an input matrix, wherein a loss function in the training process is a log-likelihood function, and a back propagation algorithm is a mixed algorithm combining root-mean-square propagation and learning rate attenuation, taking the input matrix and the output matrix as sample data, calculating the mean value and the standard deviation of current values of all the sample data, and after normalization preprocessing on the current data of the sample data through a z-score standardization method, inputting the current data into the trained neural network model for processing, and outputting a fault type corresponding to the current data. The problems of poor distribution network fault identification speed and poor accuracy in the prior art are solved.

A method, computing platform, and computer program product are provided. A computing platform receives, for a clinical trial, study design information including a set of parameters related to a success of a clinical trial and factors related to a relevance of the parameters. The computing platform provides multiple sets of varying parameter values. At least one trained machine learning model is applied to the study design information including the sets of parameter values to predict multiple success scores including a predicted overall success score. The computing platform outputs the provided set of values for the set of parameters that are associated with the at least one trained machine learning model predicting a best overall success score for the clinical trial.

The invention discloses a single equipment quality evaluation method based on fault probability and equipment state. According to the invention, the definite equipment quality characterization parameter Q with relatively complete theoretical support and the equipment quality evaluation mathematical model for calculation are provided, and the mathematical model is not only based on the mathematical statistical analysis theory, but also combined with the equipment state quantity observation value, so the calculation result is more accurate than that of the existing score deduction method and the like. According to the method, the running time of the equipment and the state quantity change process of the equipment are considered at the same time, and the method is more accurate than the method for evaluating the quality of the equipment by singly using the state quantity of the equipment for state evaluation.

According to an aspect, there is provided a computer-implemented method for processing a data set, the data set comprising respective data subsets for a plurality of subjects, each data subset comprising a plurality of data entries, each entry comprising respective parameter values for each of a plurality of parameters at a respective time point, wherein for a first data subset relating to a first subject in the plurality of subjects, one or more parameter values for at least a first parameter in the plurality of parameters is missing from the first data subset, the method comprising, for a first missing parameter value in a first data entry in the first data subset (a) determining completeness scores for the first parameter, wherein each completeness score indicates a level of completeness of the data entries in the first data subset for the first parameter and a respective one of the other parameters in the plurality of parameters; (b) determining correlation scores for the first parameter, wherein each correlation score indicates a level of correlation between the parameter values in the data set for the first parameter and the parameter values in the data set for a respective one of the other parameters in the plurality of parameters; (c) determining a subset of the plurality of parameters to use to form regression trees based on the determined completeness scores and the determined correlation scores; (d) forming a plurality of regression trees, wherein each regression tree relates to a respective parameter combination of the first parameter and one or more of the other parameters in the determined subset, and each regression tree is trained to predict a parameter value for the first parameter based on input parameter values for the one or more other parameters in the parameter combination, wherein each regression tree is trained using training data comprising parameter values for the parameters in the respective parameter combination, wherein the training data includes the parameter values in any data entry in the first data subset for which a parameter value is present for all of the parameters in the respective parameter combination; (e) using each regression tree to predict a parameter value for the first parameter based on parameter values in the first data entry for the one or more other parameters in the parameter combination; and (0 combining the predicted parameter values to estimate the first missing parameter value. A corresponding apparatus and computer program product are also provided.

The present invention relates to the field of artificial intelligence, machine learning and Big Data, and provides a device (100) and method (300) for anomaly detection, performed in a stream of events. To this end, the device (100) for anomaly detection on an input stream (101) of events (102) is provided, wherein each event (102) comprises a value (103). The device (100) comprises a modeling unit (104), configured to represent each event 102 from the input stream (101) of events (102) into at least one histogram (105) over a predefined tree data structure (106), to obtain a model 107 of the input stream (101); a change detection and adaptation module (108), configured to detect a long term change (109) of the distribution of the events (102) in the input stream 101 based on the model (107) of the input stream (101); and a scoring unit (110), configured to calculate an anomaly score (111), representing the probability of an anomaly in the input stream (101) of events (102), based on the model (107) of the input stream (101), and the detected long term change (109).

The invention provides a novel hybrid user interest similarity calculation method. A scoring matrix of used items is established by a user, when it is found that the scoring matrix of the user is empty, the similarity of characteristic attributes of the users is calculated, and similar users are searched for forecasting scores. When the number of articles commonly scored between a target user andother users is relatively small, the similarity degree of the articles is calculated, and the interest similarity of the users is indirectly calculated, wherein user interest similarity calculation ismainly divided into the three parts that the distance value of user scores is directly calculated, the contribution values of a set of scores are worked out and whether the set of scores are singularvalues or not in a whole scoring system is judged; finally, through the three user interest similarity calculation methods, smooth transition from similarity calculation according to the user attributes to similarity calculation according to user scoring information under the cold start state is achieved through a sigmoid function. The prediction scores of non-scored items of the target user arecalculated according to the user interest similarity, and N items with the highest predicted scores are selected to be recommended. By means of the method, the problems of cold start and data sparsitycan be effectively relieved, and the accuracy of forecast recommendation can be effectively improved.

The invention provides an end-to-end speech recognition modeling method based on multi-level identification, and the method comprises the steps: decoding inference which employs a post-inference algorithm; the post-inference algorithm comprises the steps that a model corresponding to a fine-grained text sequence generates a posterior probability output sequence, the output sequence can uniquely correspond to a coarse-grained sub-sequence calculation model to generate a log-likelihood value of the coarse-grained sub-sequence, and the log-likelihood value serves as cross validation of an existing prediction output sequence; and according to the likelihood probability scores obtained through calculation in the two steps, an existing decoding path is cut, and it is guaranteed that the search path is controlled within the beam width range.

The invention relates to a scoring method and device, computer equipment and a storage medium. The method comprises the steps of obtaining answer information of a target user for a target question in response to an answer instruction corresponding to the target question, and obtaining evaluation parameters of multiple evaluation dimensions corresponding to the target question, so that the evaluation parameters of the multiple evaluation dimensions can be pre-configured, and the time delay of scoring processing is reduced; scoring the answer information according to the evaluation parameters of the multiple evaluation dimensions, and obtaining score feedback information, wherein the score feedback information comprises score information and score meaning information corresponding to scores; and outputting the score information and the score meaning information corresponding to the score, so that the high-accuracy score and the score meaning can be provided for the user in a short time, and the user can learn the score and the meaning and make improvement.

An object-linking interface for a computer system dynamically links a selected target object within an host application with a second user application. The operating environment includes a plurality of means testers, each of which is associated with a specific user application, and an object mapper. The means tester is passed a copy of the target object and tests the suitability of the associated application for processing the target object and generates a corresponding likelihood score. The object mapper queries each means tester to obtain the likelihood score and identifies, based upon said likelihood score, one or more applications suitable for processing the target object.

The invention provides a method and device for training a relation extraction model, equipment and a storage medium. The method described herein includes the steps of determining a probability distribution of a set of rules for describing logic of a connection between a pair of target entities under a given triple condition based on the given triple consisting of the pair of target entities in a document and a target relationship associated with the pair of target entities. Based on the probability distribution of the set of rules under a given triad condition, a probability distribution of a score under the given triad condition is determined, the score indicating whether the target relationship in the document is valid for the target entity pair. A trained relationship extraction model is obtained by maximizing a likelihood function of a parameter of a probability distribution of the scores under given triad conditions based on the flag values corresponding to the scores. According to the fact of the present disclosure, by utilizing rules, long-range dependencies of relationships can be readily captured and better interpretability can be provided.

The invention discloses a method and a device for distinguishing commodity relations, and relates to the technical field of computers. A specific embodiment of the method comprises the steps of screening a set with directed edges based on a target loss function; establishing an objective function for distinguishing a complementary relationship or a substitution relationship by introducing a mapping vector, and then establishing a joint objective function through category constraint and multi-step path constraint; inputting a training set in the set into the joint target function, and performing optimization solution on the training set to obtain a joint target model through training and obtain a possibility score through calculation; based on the likelihood scores, distinguishing relationships between the commodities. According to the embodiment, the technical problems of data sparsity and no consideration of interdependence between commodities can be solved.