71 results about "Generalized linear model" patented technology

Diffusion weighted images and perfusion weighted images are acquired with an MRI system and used to produce low b, DWI, ADC, CBV, CBF, and MTT images of brain tissues following an ischemic event. These MRI physiological measurements are input along with a spatial location measurement to a generalized linear model that predicts the outcome of tissues surrounding a lesion.

The present invention provides a new and unique method and apparatus for detecting in a short-range communication device, such as a WLAN station (STA), the trend in WLAN signal strength based on one or more characteristics, e.g. received signal strength values and current time of their observation, by fitting a generalized linear model to the values. Based on the detected trend, three things can be inferred: 1) WLAN radio coverage available for STA is strengthening, 2) WLAN radio coverage available for STA is stationary, or 3) WLAN radio coverage available for STA is weakening.

The invention discloses a method for constructing effective connectivity in a brain region based on nuclear magnetic resonance imaging. The method comprises the following steps of: performing generalized linear model analysis on data acquired by preprocessing a functional magnetic resonance imaging (fMRI) image, and extracting a brain activation region; selecting a region of interest according to the brain activation region, and extracting a corresponding time sequence; according to the acquired time sequence, establishing a multivariate autoregression (MVAR) model, and calculating to acquire an effective connectivity matrix; and performing single sample T inspection on the effective connectivity in the effective connectivity matrix by a substitution data method, and eliminating multi-comparison by a false discovery rate (FDR) method to acquire the remarkable effective connectivity in the brain region.

The invention provides a fused HI (Health Index) equivalence lithium ion battery degradation prediction method based on principal component analysis (PCA). GRA (Grey Relational Analysis) is performed on a discharge cut-off voltage as a benchmark HI and the capacity directly characterizing the degradation parameter, and a GLM (Generalized Linear Model) is used to carry out transform modeling. If the relevancy between the transform result and the capacity is larger than 0.7, it can be judged that it is reasonable to use the discharge cut-off voltage as an HI. Then, a variety of indirect HIs are constructed according to the measured parameters of a battery, and the HIs are fused through PCA to get a fused HI. The relation between the fused HI and the discharge cut-off voltage is analyzed using the GLM. If the relevance of the GLM is larger than 0.7 and RMSE is smaller than 0.004, it can be judged that the error is small, the precision of fitting is high, and the fused HI can be used as a substitute parameter of the discharge cut-off voltage. Moreover, GLM can be used to get the relation between the HI and the direct degradation parameter effectively to determine the failure threshold and increase the relevancy between sequences by more than 50%. The constructed fused HI can be used to complete indirect degradation condition prediction and predict the degradation condition of a lithium ion battery based on direct measurement data.

A computerized system and method for estimating levels of obesity in an insured population using claims data. The model uses health risk assessment data comprising age, height, and weight information as well as information about health conditions and health behaviors for a member population. Claims data is used to train a two-stage model on the member population. The first stage comprises a support vector machine, a rule-based module, and a generalized linear model that estimates the probability of obesity. The second stage comprises a regression neural network that operates on the output of the first stage and a subset of the input feature vector. Cost and utilizations in these areas, along with overall health measures as well as demographics and social factors, are inputs to a set of pattern recognition engines that perform regression. The output is the estimated body mass index of the member.

The invention discloses an epilepsy paradoxical discharge locus positioning method and system based on EEG-fMRI. The method includes the steps of synchronously collecting brain electric data and functional magnetic resonance imaging data of a patient in a resting state to be preprocessed, marking epilepsy discharge time points and corresponding duration in the interictal phase, aligning the brainelectric data and the functional magnetic resonance imaging data on terms of time, selecting image frames of the functional magnetic resonance imaging data to extract grey matter part signals so as toconduct layered clustering to establish a epilepsy discharge grey matter template, marking epilepsy discharge related time points, obtaining expected BOLD time signals, conducting correlation analysis through a generalized linear model to find a brain area related to epilepsy discharge activities, and comprehensively obtaining an epilepsy paradoxical discharge locus. Through brain electric-functional magnetic resonance imaging fusion analysis, the epilepsy discharge time points marked by a doctor are expanded, the positioning accuracy of the paradoxical discharge locus is improved, and the method has the advantages of being simple in principle, convenient to implement and stable in result.

Embodiments of the invention provide a user behavior similarity computing method and device. The method comprises the steps of collecting a behavior characteristic value corresponding to each of a plurality of behavior characteristics of first class of users and second class of users; screening the plurality of behavior characteristics to obtain a target behavior characteristic set; building a first generalized linear model according to the target behavior characteristic set, computing a first maximum likelihood estimated value of the first generalized linear model by use of an optimization method, and obtaining an estimated parameter corresponding to the first maximum likelihood estimated value; and computing the behavior similarity of a to-be-tested user and the first class of users by use of the estimated parameter and the behavior characteristic value corresponding to each behavior characteristic in a target behavior characteristic set corresponding to the to-be-tested user. According to the method, the similarity of behaviors of different users is analyzed by full use of behavior characteristics of a lot of users, so that the use ratio of collected behavior characteristics of a lot of users is improved.

The invention discloses a genome-wide SNP (single nucleotide polymorphism) site analysis method based on combination of random forest and Relief-F. The method includes: primarily screening SNP sites with a generalized linear model; processing SNP interactive capability with Relief-F; preliminarily putting SNP sites, which are interactive, to the front of a queue; ranking the SNP sites at the rear of the queue with the random forest to recognize edge action of each single SNP site so as to obtain an SNP rank queue; removing the SNP sites at the tail of the queue; performing processing again with the Relief-F and the random forest; allowing iteration to obtain a ranking result of the SNP sites. The method has the advantages that the action of each single SNP site and interaction of the SNP sites are comprehensively considered, genome-wide SNP data can be processed so as to find those related to complex diseases, and the method is significant to the research on pathogenesis of the complex diseases, prediction on risk of diseases, development of biological drugs and the like.

The present invention provides a microalloyed steel mechanical property prediction method based on globally additive model, including the following steps: determining some influencing factors of the microalloyed steel mechanical property prediction model; calculating the components and contents of carbonitride precipitation in a microalloyed steel rolling process; expressing the microalloyed steel mechanical property prediction model as an additive form of several submodels according to generalized additive model; estimating the microalloyed steel mechanical property prediction model; and verifying reliability of the submodels. The microalloyed steel property prediction models obtained in the foregoing solution have advantages such as high prediction precision and a wide adaptation range, and may be used for design of new products and steel grade component optimization, so as to reduce the quantity of physical tests, shorten the product research and development cycle, and reduce costs.

The invention discloses an analytical method for minimum discernable change of a stereopicture of human eyes. In the method, regression analysis is carried out on subjective evaluation data of an observer by a Thurstone model under the frame of a generalized liner model, under the condition that image quality of a left viewpoint is not changed, the critical value of decreased quality of a right viewpoint image is measured when the human eyes can sense the image quality change of the stereopicture, thereby determining the maximum change range of quality capable of being decreased of the right viewpoint image relative to the left viewpoint image during asymmetric stereoscopic video coding, and ensuring that the purpose of improving the code efficiency by decreasing the image quality of the right viewpoint, and enabling the observer can not sense the fact that the image quality of the right viewpoint is decreased by using stereoscopic vision masking effect, thus ensuring the whole quality of the stereopicture.

The invention relates to systems and methods for prioritizing video slices of H.264 video bitstream comprising: a memory storage and a processing unit coupled to the memory storage, wherein the processing unit operates to execute a low complexity scheme to predict the expected cumulative mean squared error (CMSE) contributed by the loss of a slice of H.264 video bitstream, wherein the processing unit operates to execute a series of actions comprising assigning each slice a predicted value according to the low complexity scheme; extracting video parameters during encoding process, said video parameters; and using a generalized linear model to model CMSE as a linear combination of the video parameters, wherein the video parameters are derived from analytical estimations by using a Generalized Linear Model (GLM) over a video database, encompassing videos of different characteristics such as high and low motion, camera panning, zooming and still videos, further comprising wherein the GLM is constructed in a training phase as follows: determining the distribution of the computed CMSE to be a Normal distribution with the Identity link function; sequentially adding covariates using the forward selection technique where by the best model is evaluated at each stage using the Akaike's Information Criterion (AIC); the training phase of the model generates regression coefficients; the final model is validated through the testing phase by predicting the CMSE for different video sequences, not in the training database; and by using the regression coefficients, the expected CMSE values are predicted for each slice.

The invention relates to a unified modeling method employing a generalized measuring transverse correlation mode. The unified modeling method is characterized by comprising the following steps of grid equipment modeling, secondary equipment modeling, primary equipment and secondary equipment correlation establishing, asset modeling, other-data abstract modeling, generalized measuring model establishing and meta-model defining, wherein secondary equipment modeling is conducted according to the IEC61850, and a PSR is used as the center of a generalized measuring model. The method is also applicable to unified modeling of an automated system of all related information in a power system. The method increases the software development efficiency and accuracy, shortens the development cycle and realizes power grid model unification, a complete data platform is provided, the flexible metadata mechanism is adopted, a multi-layer system framework for services is used, and therefore convenience is brought to electric power information resource sharing.

A Physical Fault Analysis (PFA) outcome prediction tool utilizes previously-generated evaluation data and associated PFA outcome data to generate a Bayesian Generalized Linear Model (BGLM), and then utilizes the BGLM to generate a PFA outcome prediction for newly-submitted evaluation data that operably characterizes measured operating characteristics of an IC chip that is being developed. The BGLM generation methodology by utilizing a Generalized Linear Model (GLM) in a Bayesian framework to form a hierarchical model representing the evaluation data and associated PFA outcome data as a linear combination. The PFA outcome prediction includes a credible interval of a posterior distribution that effectively represents a cross-sectional portion of the BGLM corresponding to the newly-submitted evaluation data. The previously-generated evaluation data and associated PFA outcome data are stored in a training data library, which is updated to include newly-submitted evaluation data and associated PFA outcome after each PFA is performed.

The invention relates to a construction method for evaluating river ecological conditions through aquatic organism community multi-parameter indexes. Human interference gradient identification is carried out based on physicochemical environment factor investigation and principal component analysis, parameter screening is carried out by combining a generalized linear model and AIC information, parameter weights are determined by considering the weights of parameters for human interference gradient sensitivity, and construction of multi-parameter index evaluation river ecological conditions is realized. Aiming at the current situation that the standard method for constructing the aquatic organism community multi-parameter indexes is lacked in China at present, the invention provides the standardized construction method which has universality, is easy to operate and can be popularized, results obtained by adopting the standardized construction method have comparability, and scientific data support can be provided for managers to make relevant decisions.

The present invention discloses a hidden Markov model calculation method for describing a driving behavior of an electric vehicle. The method comprises: modeling a driving condition of a single electric vehicle by using a hidden Markov chain, and describing a change of the driving condition of the electric vehicle within a period of time. According to the method, the change of the usage of the electric vehicle can be flexibly captured, and according to needs, can be described using a travel start time, a travel ending time, a travel start location and a travel ending location, etc. Because of a large quantity of parameters to be estimated, a B-spline method is used to convert a model into a generalized linear model, and spline nodes are selected reasonably, so as to reduce the quantity of parameters to be estimated.

The invention provides a crowd abnormal event detection method based on a hybrid tracking and generalized linear model. The crowd abnormal event detection method based on the hybrid tracking and generalized linear model includes the following steps of tracking pedestrian trajectories, extracting feature points of tracking paths, and conducting classification of at least one direction and activityof foreground objects using neural networks based on the generalized linear model. The crowd abnormal event detection method based on the hybrid tracking and generalized linear model conducts modelling using the neural networks with the hybrid tracking model and a genetic algorithm combined, and then an established feature model is used for representing group event information in the scene more intuitively. Furthermore, selecting a appropriate feature model has an important influence on the performance of final group abnormal event detection.

The invention discloses a prediction method of an urban road signalized intersection motor and non-motor traffic conflict number. Expectation value of a motor and non-motor conflict is analyzed by utilizing a plurality of motor and non-motor traffic conflicts and conflict flow of the motor and non-motor traffic conflicts as data and a generalized linear model, a traffic conflict model is built to the motor and non-motor traffic conflict, and urban road signalized intersection motor and non-motor traffic safety is evaluated. According to the prediction method of the urban road signalized intersection motor and non-motor traffic conflict number, traffic flow parameters of traffic conflicts of motor vehicles and non-motor vehicles are obtained through a traffic flow detecting device, the traffic conflict model is built, the urban road signalized intersection motor and non-motor traffic conflict is predicted, and safety evaluation is conducted to the non-motor vehicles of the urban road signalized intersection according to the predicted motor and non-motor traffic conflict number, and the defects of imperfection of a current traffic accident data base are overcome. The prediction method is more accurate and scientific compared with manual observation conflicts of the prior art, capable of promoting the traffic conflict technology to be applied in engineers, and has practical engineering applying value in the aspects of city traffic safety management and evaluation.

The present invention relates to the field of distributed systems, and discloses a deadlock detection method suitable for generalized models, comprising the following steps: (A) a regular node stores output side information; (B) the regular node detects and solves a local deadlock and transmits reduction information to an initial node; (C) the initial node performs reduction and checks overall deadlocks. The present invention is advantageous in effectively reducing the computation pressure of the initial node and solving to a certain degree the problem of common local deadlocks.

The invention discloses a traffic accident frequency prediction method based on traffic analysis cells. According to the method, first, historical accident data of all the traffic analysis cells within a research range is collected; second, a geographically weighted regression model is adopted to set a relation between accident frequency of the traffic analysis cells and explanatory variables andset a matrix form of parameters in front of the explanatory variables and a weight function in the parameter estimation process; third, model parameter estimation is performed; and finally accident frequency prediction is performed according to a valid accident prediction model. Through the method, the problems that in the past, differences in geographical units at a county level and a city levelare large, it is assumed that all the geographical units are mutually independent through a generalized linear model, consequently, spatial correlation and heterogeneity of accident data cannot be explained, and prediction accuracy is influenced are solved, different influences of different spatial position explanatory variables on accident frequency can be described, and guidance is provided fortraffic safety control.

The invention provides a brain functional region positioning method based on local smoothing regressions. The brain functional region positioning method based on the local smoothing regressions comprises the following steps of pretreating data and deciding a design matrix X; taking a voxel vi as the center of a sphere and r as a semidiameter for the establishment of a spherical selected region and extracting the time sequence of all the voxels in the spherical selected region; according to the time sequence of all the voxels in the spherical selected region and the design matrix, forming an objective function and optimizing the objective function; calculating a condition specificity effect of the voxel vi; turning to the next voxel vi+1 and repeating steps from S2 to S4 till the execution of the steps on each voxel of a whole brain; and setting a threshold value for a whole brain perception mapping so as to obtain a brain functional region positioning map relevant with stimulus conditions. All the generalized linear models based on the regressions of the single voxels and based on Gaussian smoothing filtering can be regarded as special cases of the invention; the brain functional region positioning method based on the local smoothing regressions can be integrated into a framework used for a searchlight method; after the obtainment of regression coefficients, mahalanobis distances between various predictor coefficients are calculated; and through the adjustment on hyper-parameters alpha and beta, smoothing effects of various degrees are obtained.

The present disclosure provides a new scalable coordinate descent (SCD) algorithm and associated system for generalized linear models whose convergence behavior is always the same, regardless of how much SCD is scaled out and regardless of the computing environment. This makes SCD highly robust and enables it to scale to massive datasets on low-cost commodity servers. According to one aspect, by using a natural partitioning of parameters into blocks, updates can be performed in parallel a block at a time without compromising convergence. Experimental results on a real advertising dataset are used to demonstrate SCD's cost effectiveness and scalability.

The invention discloses a user consumption behavior prediction model training method and device, equipment and a storage medium, and relates to the technical field of machine learning model training.The method includes: obtaining training set and a test set from a database; obtaining a prediction model initialization weight vector, an inverse connection function and a learning rate parameter; forany training data, normalizing the weight vector and constructing a first random variable, sampling the first random variable for multiple times to obtain a first inner product estimated value, and updating the weight vector according to the inverse connection function, the first inner product estimated value, the label information of the training data and the learning rate parameter; and testingeach weight vector by adopting test data, and obtaining a prediction model of the trained user consumption behavior according to the weight vector with the minimum risk of the prediction model. In the generalized linear model training process, inner product estimation is approximated by sampling random variables multiple times, the model training efficiency is improved, the model accuracy is ensured, and the generalized linear model can effectively predict user consumption behaviors.

The invention discloses a method and device for predicting a target variable in the technical field of computers. The specific embodiment of the method comprises the steps of obtaining feature data and a model identifier; determining a prediction model which corresponds to the model identifier and is obtained by training a machine learning model; and inputting the feature data into the predictionmodel to predict a target variable. A model trained by machine learning is adopted to predict a target variable, so that the problem of inaccurate link function selection caused by irregular distribution of the target variable and poor mathematical distribution fitting when a generalized linear model is adopted is avoided. The target variable is directly modeled, the prediction model obtained through machine learning is more flexible in characteristic space division compared with a traditional generalized linear model, and the predicted target variable is more accurate.

Diffusion weighted images and perfusion weighted images are acquired with an MRI system and used to produce low b, DWI, ADC, CBV, CBF, and MTT images of brain tissues following an ischemic event. These MRI physiological measurements are input along with a spatial location measurement to a generalized linear model that predicts the outcome of tissues surrounding a lesion.

Use of Generalized Context Trees to assign a unique state from a finite set to any string is provided. The method optionally refines the generalized context tree into a refined generalized context tree having a finite state machine (FSM) property. Refining occurs whenever the generalized context tree does not have the finite state machine property. Alternately, a method for constructing a representation of a source usable within an FSM is provided, comprising evaluating a node comprising a suffix tail and verifying the suffix tail is included in the representation, and inserting at least one node to the representation when the suffix tail is not in the representation.

A quantification system is configured for debiasing media creative efficiency. In some embodiments, the quantification system leverages a weighted generalized linear model (GLM) to determine the individual impacts of media creatives beyond network effects. To prepare input data for fitting the weighted GLM, the quantification system analyzes spot airing data, creates a specific data structure for storing observations (e.g., network-media creative combinations) that can be provided to the weight GLM as input, and computes additional input data points needed by the weighted GLM (e.g., network spend, media creative efficiency per network-media creative combination, etc.). The weighted GLM is then fitted to obtain coefficients representing the individual impacts of the media creatives. The quantification system utilizes the computed impacts to adjust the previously computed media creative efficiency for each media creative. In this way, relative performance of media creatives can be objectively quantified across networks without needing digital evidence.

The invention discloses a land utilization change driving force screening method and device based on LASSO regression. The method comprises the steps that explanatory variable and response variable data in a research region is acquired; a LASSO regression model is constructed, and LASSO regression is executed; a residual sum of squares and multi-colinearity indexes in each step of LASSO regressionare calculated, and driving factors needing to be eliminated are determined according to calculation results; remaining driving factors are used as explanatory variables of the LASSO regression modelto execute LASSO regression again; and the importance of the driving factors is determined according to coefficients of response variables in the LASSO regression executed again. Through the screening method and device, a refined model can be constructed, variable screening and complexity adjustment can be performed during fitting of a generalized linear model, and accurate screening can be performed according to numerous dependent variables in multi-colinearity. The screening method and device can be widely applied to the environment modeling field.

The invention discloses a machine learning preoperative positioning method based on a generalized linear model. The machine learning preoperative positioning method comprises the following steps thatobtaining a structure image and a functional magnetic resonance image; preprocessing the structure image and the functional magnetic resonance image; segmenting and registering the structure image andthe functional magnetic resonance image, and extracting a motion network by using a double regression method; constructing a generalized linear prediction model, and performing generalized linear prediction model fitting on brain partitions; according to the method, the motion activation graph of the motion area prediction individual can be more accurately identified on the basis of resting statefunctional magnetic resonance, and passive task activation can be effectively predicted by using a generalized linear prediction model trained by active task activation with actual task functional magnetic resonance image activation as a reference; the generalized linear prediction model has important clinical application value for patients who cannot achieve satisfactory task performance, including old people, children and tumor patients.