239 results about "Approximate computing" patented technology

The invention provides an online log analysis method, which comprises the following steps of: acquiring a log file from each node of a cluster system on line; defining an event sequence and an event correlation rule according to the log file, counting unitary event sequences, searching out a frequent unitary event sequence set, and obtaining a binary frequent event sequence candidate set C (2); performing statistics on support degree counting and posterior degree counting of the candidate binary frequent event sequence in the C (2), and calculating the degree of confidence and posterior degree; generating a binary frequent event sequence set F (2) and a binary rule set R (2); repeating the process until F(k) and R(k) are obtained, performing join operation on k-element frequent event sequences in the F(k), and generating an F(k+1)-element frequent event sequence set; performing approximate calculation on attribute values of elements which are not in R(k+1) but in F(k+1), until all frequent event sequences are generated; finding the event correlation rule according to the frequent event sequences; and predicting events in the log according to the event correlation rule on line.

The invention relates to a soft tissue deformation simulation method based on smooth particle hydrodynamics, belonging to the technical field of graphic processing. In the method, a smooth particle hydrodynamics method is selected, and a viscoelastic mechanics model is used for reflecting the biomechanical characteristics of soft tissue. The method comprises the following steps of: constructing a series of equations related to soft tissue deformation simulation according to the viscoelastic model; selecting a proper support domain search strategy and a smooth kernel function, approximately calculating each related item of the equation by adopting the particle approximation method, and calculating the variation values of the density, the position, the velocity and the like of each particle along with time through the display integration method; and dynamically outputting the status of each time step size of the particle model to a screen, rendering texture irradiation, and displaying the real-time deformation process of soft tissues and organs under stressing conditions. The method does not need troublesome grid computing, thereby increasing the accuracy and the real-time performance of soft tissue deformation simulation.

The invention discloses a pedestrian detection method. Multiple times of convolution and pooling are performed on an input image through the pedestrian detection method based on a convolutional neuralnetwork; the features of the original image are extracted so as to obtain the corresponding feature graph of the original image; the corresponding feature graph after zooming of the original image isapproximately calculated through the image feature pyramid rules; a candidate window is generated through a region proposal network RPN; a candidate proposal window is further selected and summarizedaccording to the pedestrian size distribution in the candidate window; the corresponding weight of different scales of pedestrian targets on different scales of images is trained by using the training data having the tag; and the classifier network is trained. The summarized candidate window is solved, and the confidence obtained through the classifier and the set threshold are compared and finalpedestrian detection judgment is performed. Heavy calculation amount of obtaining the feature graph through image zooming calculation can be avoided by application of the image feature pyramid, and detection is performed on different feature graphs by using the weighing mode of different weights so that misjudgment and leak detection caused by single feature graph detection can be effectively avoided.

A system and method for improved Time Of Arrival (TOA) distance measurements between nodes of a wireless ad-hoc network. Specifically, the present invention is a system and method of distance estimation using square-root raised-cosine pulse shaping and chip matched filters on direct sequence spreading waveforms, the multiplication of which produces raised-cosine filtered pulse responses. The responses are used to identify a time when a function is at a maximum, corresponding to the actual signal reception time. The system and method produces a raised-cosine filtered pulse response and an auto-correlation function based on a received signal. A peak value of the auto-correlation function is calculated based on a quadratic approximation, which is corrected using a signal sampling phase offset detected between the raised-cosine filtered pulse response and the calculated peak value. The calculated peak value is then corrected to represent an actual reception time for received signals.

The invention provides a relevance vector machine-based multi-class data classifying method, which mainly solves the problem that the traditional multi-class data classifying method cannot integrally solve classifying face parameters and needs proximate calculation. The relevance vector machine-based multi-class data classifying method comprises a realizing process comprising the following steps of: partitioning a plurality of multi-class data sets and carrying out a normalizing pretreatment; determining a kernel function type and kernel parameters; setting basic parameters; calculating the classifying face parameters; calculating lower bounds of logarithms and solving variant values of the lower bounds of the logarithms and adding 1 to an iterative number; if the variant values of the lower bounds of the logarithms are converged or the iterative number reaches iterating times, finishing updating the classifying face parameters, and otherwise, continuing to updating; and obtaining a prediction probability matrix according to the updated classifying face parameters, wherein column numbers corresponding to a maximum value of each row of the matrix compose classifying classes for testing the data sets, and samples which have the prediction probability less than a false-alarm probability and the detection probability corresponding to a false-alarm probability value set in a curve are rejected. The relevance vector machine-based multi-class data classifying method has the advantages of obtaining classification which is comparable to that of an SVM (Support Vector Machine) by using less relevant vectors and rejecting performance and can be used for target recognition.

A system and method for computing anisotropic texture mapping parameters by using approximation techniques reduces the complexity of the calculations needed to perform high quality anisotropic texture filtering. Anisotropic texture mapping parameters that are approximated may be computed using dedicated processing units within a graphics processor, thereby improving anisotropic texture mapping performance. Specifically, the major axis and minor axis of anisotropy are determined and their respective lengths are calculated using approximations. Other anisotropic texture mapping parameters, such as a level of detail for selecting a particular level are computed based on the calculated lengths of the major and minor axes.

The invention relates to a loss allocation suspicion analysis-based anti-electricity stealing analysis method. The loss allocation suspicion analysis-based anti-electricity stealing analysis method is put forward based on analysis on different line loss and power consumption anomaly and by means of suspected power consumption calculation. With the loss allocation suspicion analysis-based anti-electricity stealing analysis method adopted, the electricity stealing behaviors of users are ultimately reflected by the anomaly of measured power consumption of the users. The objective of the invention is to approximately calculate power consumption of the users in a transformer area which are not measured actually so as to determine suspected power users. According to the loss allocation suspicion analysis-based anti-electricity stealing analysis method of the invention, line loss calculation is divided into statistical line loss calculation, theoretical line loss calculation and management line loss calculation; and calculation methods of different lines are put forward; loss-counted power consumption abnormal conditions are calculated in different aspects; suspected power consumption is analyzed; suspected power consumption is compared and distinguished according to a plurality of power consumption abnormal conditions, so that the suspected power consumption of the users can be analyzed quantitatively; a specific power consumption suspicion degree method is put forward; the suspicion degrees of the users are calculated; thresholds are screened based on the suspicion degrees; required target users are filtered out; and an investigation list is formed.

The invention discloses a joint path selection and power distribution method for energy collection nodes in a wireless sensor network, and belongs to the field of cooperative communication technologies. The method comprises the steps of analyzing a system scene, describing problems; establishing a system mathematic model; and then finding the optimal solution by using an optimization method. The method aims at the special application scene and is derived from the actual application, and being different from the traditional independent sensor node or gateway resource distribution, the method comprehensively considers the joint power distribution and path selection of sensor nodes and gateways, and maximizes the handling capacity performance of the communication nodes by using the gateways as relay stations. According to the method of the invention, the solution of an optimization problem is processed by using convex optimization to convert a target function of the optimization problem without approximate calculation, so that computation complexity is greatly reduced while accuracy of the problem is not influenced, and delay generated by system overheads is reduced; a Lagrangian multiplier method is used in an optimizing process, and thus an optimizing speed is rapid; a subgradient method and an incremental step length are used in an iterative process, so that optimizing is more accurate.

The invention discloses an approximate floating-point multiplier for a neural network processor and a floating-point multiplication. When the approximate floating-point multiplier executes fractional part multiplying operation on an operand, part bits are intercepted from all high bits of a fractional part of the operand according to designated precision, and 1 is supplemented to the front and the back of the intercepted part bits to obtain two new fractional parts; multiplying operation is performed on the two new fractional parts to obtain an approximate fractional part of a product; and zero is supplemented to a low bit of the normalized approximate fractional part so that the bits of the approximate fractional part are consistent with the bits of the fractional part of the operand, and therefore the fractional part of the product is obtained. According to the approximate floating-point multiplier, an approximate calculation mode is adopted, different bits of the fractional part are intercepted according to a precision demand for corresponding multiplying operation, energy loss of multiplying operation is lowered, multiplying operation speed is increased, and therefore the performance of a neural network processing system is more efficient.

The invention discloses a quick source signal reconstruction method achieving blind sound source separation of two microphones, belongs to the field of speech signal processing, and particularly relates to a quick source signal reconstruction method achieving sound source separation of the two microphones on the situation that the number of sound sources and the surrounding environments are unknown. On the situation that temporal envelopes of reconstruction source signals are not influenced, a small additional phase is added to each frequency component of reconstruction signals, contribution of each source signal to a time-frequency point is approximately calculated to simulate the proportion of each source signal in the time-frequency point, and accordingly the source signals are restored. A traditional solving process of an underdetermined system of equations is avoided, calculation steps are simplified, and the aim of quickness is achieved. Accordingly, compared with an existing algorithm, the method has the advantages that calculation amount is little, and the signal-noise ratio is high on the situation that the number of the sound sources is increased.

An apparatus, system, and method are disclosed for performing fast approximate computation of statistics on query expressions in order to improve query optimization within a database management system by accurately and quickly estimating the sizes of intermediate query results. This is accomplished by analyzing a query for join instruction and identifying a fact table and a dimension table within those join instructions. Then, frequency statistics corresponding to distinct values of within the fact table are retrieved from a catalog table. Those frequency statistics are used in combination with a full scan of the dimension table accurately and quickly estimate frequency statistics for an expected join between the fact table and dimension table. The estimated frequency statistics corresponding to the expected join may then be used in such operations as query optimization.

The invention discloses a large-scale MIMO downlink user scheduling method assisted by machine learning, which comprises the following steps: S1, a base station obtains a characteristic mode energy coupling matrix in a characteristic direction through an uplink detection signal sent by a user; S2, the base station calculates the sum rate under various user and beam combinations by using the eigenmode energy coupling matrix and the machine learning method; S3: the greedy algorithm is used to realize the user scheduling with the sum rate maximization criterion, and the optimal user beam pairingcombination is obtained. The invention obtains the statistical channel information through the uplink detection signal, and adopts the sum rate maximization criterion to carry out the user scheduling.Under the condition that the base station only has statistical channel information, the user scheduling complexity under large-scale antennas can be greatly reduced by the targeted feature extractionand the design of neural network, and the performance is close to the optimal, so the method has good applicability and robustness.

The invention discloses a fault diagnosis system and a fault diagnosis method based on sequence and consequence analysis of an event tree. The system comprises a data input module, an event tree resolution module, a fault tree preprocessing module, a fault tree analysis module and a diagram display module, wherein the data input module is used for acquiring reliability data and structure data of a given system, and storing the data in modes of a fault tree model and an event tree model; the event tree resolution module is used for processing the given event tree model and the fault tree model, processing a successful branch of the event tree with a respective substitution method, and constructing a traditional sequence fault tree model and a consequence fault tree model; the fault tree preprocessing module is used for preprocessing the given fault tree model; the fault tree analysis module is used for performing fault mode analysis, probability computation, importance computation, sensitivity computation and uncertainty analysis on the given fault tree model; and the diagram display module is used for displaying a fault analysis result of the given system in a diagram form. The system and the method can accurately construct sequence and consequence fault trees, and perform various approximate computation and accurate computation on the successful branch of each function event of the event tree, so that the reliability is improved.

The invention provides an air combat behavior modeling method based on fitting reinforcement learning, and realizes intelligent decision of tactical action in virtual air combat simulation. The method includes the steps: sampling the track of the aircraft combat process; fitting utility functions in a state space and approximately calculating the utility functions by Behrman iteration and least square fitting; making operational decisions, making action decisions by the aid of the fitted utility functions in the forecasting process, and determining finally executed action according to forecasted execution results. The fitting efficiency and the acquisition efficiency of the utility functions can be effectively improved, and optimal action strategies can be more rapidly acquired as compared with a traditional method.

A stable and accurate reactor physics thermotechnical coupling calculating method can simultaneously obtain statuses of a neutronics field, a fluid field and a heat transfer field; all variables are simultaneously updated in an iteration process, so the method has better value stability when compared with a conventional coupling method; the method uses a simultaneous time-space neutron diffusion equation and a thermotechnical equation discrete equation to obtain a novel nonlinear equation group with larger scale; using a Newton iterative method to resolve values of the novel nonlinear equation group, wherein the Newton iterative method needs to calculate a Jacobian matrix of the equation group in each iteration step, so physics thermotechnical coupling calculation is difficult. The novel method builds a value relation between the diffusion equation and thermotechnical equation, and explicit calculation can provide the Jacobian matrix, so problems caused by approximation calculation f the Jacobian matrix can be prevented; in addition, the novel method inherits Newton method characteristics, and hopes to provide faster convergence speed.

Generally, a method and apparatus for making a prediction-based market including unconventional prediction options to market participants includes determining a prediction framework that includes a plurality of conditional scenarios. The method and apparatus includes calculating realization odds for each of the conditional scenarios using an approximation calculation technique and via an interface, receiving a plurality of predictions associated with selected conditional scenarios, each prediction having an associated value and building the prediction-based market using the predictor. The method and apparatus further includes updating realization odds for each of the conditional scenarios in the prediction framework using the approximation calculation technique and settling the predictions based at least on the updated realization odds.

The invention discloses an urban rail transit ATS energy-saving time table compilation method. By analyzing energy-saving operation modes of a plurality of trains, decomposing operation stages and approximately calculating the available energy-saving operation overlapping time, train operation energy consumption is provided for energy consumption calculation; and in addition, the genetic algorithm is introduced, the aim of minimizing the total energy consumption is transformed into the aim of optimizing the optimal arriving time and departure time of the multiple trains, and the multiple trains are subjected to subsection calculation in an off-line calculation mode. An ATS energy-saving time table generated through the method can achieve the purpose of reducing the total operation energy consumption of the multiple trains.

The invention provides a thermistor temperature detecting method based on DSP, wherein a relational expression between T and lgR-T is obtained by utilizing the temperature of thermistor R-T characteristic detection power equipment and fitting a lgR-T curve in DSP. The method comprises the following steps of: converting the computation for lnR into the approximate operation of addition, subtraction, multiplication and division, obtaining a computing formula by adopting a high-item truncation approximate computing method, then determining a segmenting mode by adopting an error comparison method, and finally obtaining the lnR value of each segment by adopting a segmentation algorithm to realize accurate temperature computation. The temperature detecting precision is greatly improved by adopting the method; the method is simple to realize and has high operational speed; the hardware structure is simple, and the cost is saved.

Under the present invention, a computer work gradient matrix (hereinafter “matrix”) is developed by applying a set of background loading levels and a set of elemental work increments to a computer system, and then measuring the resource consumption of the computer system based thereon. Task work path(s) are also constructed by applying specific tasks to the computer system at a certain background loading level, and measuring the resource consumption of the computer system over time. Once the matrix and task work path(s) have been provided, they can be used to approximate the resource consumption of the computer system in response to a task applied at any background loading level.

The invention belongs to the related technical field of structural topology optimization, and discloses a substructure interpolation model modeling method for topology optimization design of multi-level lattice structures. The method includes the following steps: selecting configuration forms and characteristic geometric parameters of substructures; defining the density of the substructures as design variables, and establishing a relationship between the design variables and the characteristic geometric parameters; changing values of the design variables, and generating a series of substructures with different density values; performing finite element discretization on the series of substructures; obtaining a series of super-element stiffness matrices; extracting the characteristic reduction basis of the series of super-element stiffness matrices and a characteristic reduction basis mapping coefficient of each super-element stiffness matrix; constructing a relationship between the characteristic reduction basis mapping coefficient and the design variables; and establishing an approximate calculation formula of the super-element stiffness matrices. According to the scheme of the invention, an approximate calculation model of the super-element stiffness matrices of the substructures is established, the model is used for the topology optimization design of scale-associated multi-level lattice structures, the connectivity of the structures can be guaranteed, and the design scheme can be directly used for manufacturing.

In training data, a similarity matrix is generated for each of types of data corresponding to different kernels, and graph Laplacians are formed individually from the similarity matrices. An entire graph Laplacian is defined as linear combination of the individual graph Laplacians with coupling constants. Observation variables and latent variables associated therewith are assumed to form normal distributions, and the coupling constants are assumed to form a gamma distribution. Then, on the basis of a variational Bayesian method, a variance of the observation variables and the coupling constants can be figured out with a reasonable computational cost. Once the variance of the observation variables and the coupling constants are figured out, a predictive distribution for any input data can be figured out by means of a Laplace approximation.

The invention provides a dual-medium hybrid fading communication system performance analysis method based on an AF (Amplify and Forward) protocol, and belongs to the technical field of dual-medium cooperative communications. The method comprises the following steps: establishing a Moment Generating Function (MGF) model that satisfies the signal-to-noise ratio of a wireless link in the Nakagami distribution; establishing an MGF model that satisfies the signal-to-noise ratio of a power line link in the logarithmic normal distribution; according to the two MGF models respectively established in the above steps, determining an MGF model of the total signal-to-noise ratio of a system under Maximal Ratio Combining (MRC); and determining the bit error rate and outage probability of the system according to the system total signal-to-noise ratio MGF model. According to the dual-medium hybrid fading communication system performance analysis method based on the AF protocol provided by the invention, the system performance calculation complexity is simplified, optimal distribution parameters of system performance can be obtained, and a theoretical expression of the bit error rate and the outage probability of the system can be calculated approximately.

The invention discloses an antenna design method based on a neural network, comprising the steps of constructing an antenna initial model; an RBF neural network and PSO algorithm parameters are initialized; several groups of antenna design parameters are selected and input into the initial antenna model to obtain the corresponding antenna model response; the fitness function value of RBF neural network parameters and the optimal value of algorithm are calculated; the optimal parameters of RBF neural network are obtained; the RBF neural network model is tested and optimized; the optimized RBF neural network model is used as the proxy model to simulate the response of antenna design parameters, and the antenna design is completed. The invention can effectively improve the prediction accuracyand the convergence speed of the neural network, The optimal neural network is used as a proxy model to fit the electromagnetic simulation data of antenna design parameters, which can replace the time-consuming electromagnetic simulation to achieve the instantaneous approximate calculation from antenna structural parameters to electromagnetic response, reduce the number of electromagnetic simulation, reduce the computational cost and improve the efficiency of antenna design.

The invention provides a subgraph based link similarity quick approximate calculation system and a method thereof. The method comprises the following steps: inputting a graph structure and a node pair with similarity to be calculated; continuously extending certain nodes around the node pair into a subgraph; calculating the similarity of all node pairs in the subgraph through iteration so as to obtain the similarity among specific node pairs; stopping the iteration process when the results of two adjacent iterations meet the requirement of a convergence condition, or the approximate degree of the results meet the requirement of a user. The system and the method can quickly and approximately calculate the similarity of specific node pairs.

The invention discloses a spatial mesh antenna electrical property dynamic response analysis method based on a second-order approximate calculation formula. The method includes the steps that antenna geometric parameters, material parameters and electrical parameters are input; an ideal antenna far zone electric field is calculated; an antenna structure finite element model is established; node, unit and shape function information is extracted; unit electrical property first-order and second-order coefficient matrixes are calculated; total electrical property first-order and second-order coefficient matrixes are assembled; antenna structure mode analysis is carried out; a mode matrix is output; an electrical property dynamic response first-order derivative column vector and a second-order Hessian array are calculated; dynamic loads are applied; mode coordinates are calculated; the change amount of the antenna far zone electric field is subjected to approximate calculation; dynamic response of the far zone electric field is calculated; whether electrical properties meet requirements or not is judged; if yes, an antenna structural design scheme is output; or else antenna parameters are updated. The electrical property dynamic response calculation time can be effectively shortened on the premise of guaranteeing calculation precision, and electromechanical integrated optimization design of an antenna structure is achieved.

An approximate computation unit computes an approximate of a determinant of a Hessian matrix relating to a parameter of an observation model represented as a linear combination of parameters determined by each layer 1 latent variable of factorial hidden Markov models. A variational probability computation unit computes a variational probability of a latent variable using the approximate of the determinant. A latent state removal unit removes a latent state based on a variational distribution. A parameter optimization unit optimizes the parameter for a criterion value that is defined as a lower bound of an approximate obtained by Laplace-approximating a marginal log-likelihood function with respect to an estimator for a complete variable, and computes the criterion value. A convergence determination unit determines whether or not the criterion value has converged.

The invention discloses a multi-level hierarchical DSmT rapid approximate reasoning fusion method. Aiming at the condition that only a monad focal element has the belief assignment in an ultra power set space, a binary tree and trinomial tree grouping technique is applied to perform the rigid grouping to realize the mapping of different granular focal elements. In the meantime, the belief assignment summation is performed to each grouping focal element corresponding to each information source, two information sources of a coarsing ultra power set space are fused by applying the DSmT rule and the proportion distributing conflict rule, a fusion result is stored to be used as a connecting weight between a father node and a child node, and then the normalized treatment is performed to each grouping focal element belief assignment. The multi-level hierarchical frequency is determined through setting the depth of the tree. The invention has the advantages of simple computation, high computational efficiency, good robustness of the approximate computing result and the like, and solves the problem that the DSmT evidential reasoning has the computational bottleneck along with the increase of discrimination frame focal elements.