12 results about "Generalized inverse matrix" patented technology

The present application relates to the field of artificial intelligence. Provided in the present application are a method and system for implementing a self-attention mechanism having a linear complexity, and a device and a medium. The method comprises: performing positional encoding on sequence data to be processed, and performing feature mapping to apply the positional encoding to a query matrix and a key matrix; on the basis of the query matrix and the key matrix, performing calculation to obtain a low-rank query matrix, a low-rank key matrix and a low-rank generalized inverse matrix; augmenting a value matrix to obtain an augmented matrix, and on the basis of the augmented matrix, the low-rank key matrix, the low-rank generalized inverse matrix and the low-rank query matrix, obtaining a sequence self-attention matrix; and transposing the sequence self-attention matrix, using the transposed matrix as sequence data to repeat the foregoing steps, using the newly obtained sequence self-attention matrix as a feature self-attention matrix, and on the basis of the sequence self-attention matrix and the feature self-attention matrix, obtaining a self-attention matrix having a linear complexity. The present application implements linear self-attention, better maintains the model performance and has relatively good expansibility.

Provided is an information processing method, etc. that assists a user in interpreting behavior of a generated machine learning model. In the information processing method, a computer executes processing of recording a plurality of sets of an explanatory data vector xn input to an existing machine learning model (21) and an objective data vector yn output from the machine learning model (21) in association with each other, calculating an interpretation matrix A_dagger which is a vector product of an explanatory matrix X in which a plurality of sets of the explanatory data vector xn is arranged and a generalized inverse matrix of an objective matrix Y in which the objective data vector yn is arranged in an order corresponding to the explanatory data vector X, and outputting a chart (41, 42, and 43) related to the interpretation matrix A_dagger.

The present application belongs to the technical field of signal processing. Disclosed are a signal processing method and apparatus, and a device. The method comprises: using an encoding matrix to modulate driving signals sent in parallel by a plurality of TX electrodes, wherein the encoding matrix comprises a full-column-rank matrix of size M×N or a full-row-rank matrix of size N×M, M and N are both positive integers, and M＞N; and using a decoding matrix to decode sensed signals, wherein the decoding matrix is a generalized inverse matrix of the encoding matrix, the sensed signals are obtained by means of sensing by a plurality of RX electrodes, there is coupling capacitance between any RX electrode among the plurality of RX electrodes and each TX electrode among the plurality of TX electrodes, and the coupling capacitance is determined on the basis of the signals decoded by means of the decoding matrix. In the method, when a touch control panel uses a CDM driving mode, driving signals are modulated using a non-square encoding matrix, thereby increasing SNRs of sensed signals generated when the touch control panel uses the CDM driving mode.

The invention relates to the technical field of cable-stayed bridge cable force control, in particular to a cable-stayed bridge cable force and line shape cooperative control method and system based on finite element analysis. The actually measured vertical displacement deviation of each measuring point is compared with a preset displacement deviation threshold value, and the bridge linear urgency degree is obtained; based on the actually measured vertical displacement deviations of all the measuring points, cable force adjustment amounts of all cables in the finite element model of the cable-stayed bridge are solved by adopting a generalized inverse matrix so as to determine a comprehensive cable force adjustment scale, and an initial allowable discrete value is determined in combination with the bridge linear urgency degree and a preset design cable force average value of all the cables; constraint conditions in the cable-stayed bridge cable force and linear cooperative control process are optimized. According to the invention, the cable force uniformity constraint threshold is adaptively adjusted according to the actual risk and demand of the bridge, and the rationality and feasibility of the cable force and line shape cooperative control strategy are improved.

The application relates to the technical field of cable-stayed bridge cable force control, in particular to a cable-stayed bridge cable force and linear collaborative control method and system based on finite element analysis, which comprises the following steps: by analyzing the cumulative degree of the vertical displacement of each measuring point when a unit cable force is applied to all cables, comparing the measured vertical displacement deviation of each measuring point with a preset displacement deviation threshold, and determining the bridge linear urgency; based on the measured vertical displacement deviation of all measuring points, using a generalized inverse matrix to solve the cable force adjustment amount of all cables in the finite element model of the cable-stayed bridge, determining the comprehensive cable force adjustment scale, combining the bridge linear urgency and the preset design cable force average value of all cables, determining the initial allowable dispersion value, and optimizing the constraint condition in the cable force and linear collaborative control process of the cable-stayed bridge. According to the application, the cable force uniformity constraint threshold is adaptively adjusted according to the actual risk and demand of the bridge, and the rationality and implementability of the cable force and linear collaborative control strategy are improved.

The invention discloses a centrifugal pump cavitation characterization and recognition method based on load spectrum inversion. The method comprises the following steps that S01, a transfer function of a centrifugal pump system is obtained; s02, calculating a generalized inverse matrix; s03, calculating an optimal regularization coefficient; s04, calculating a load spectrum to be inverted; s05, calculating a one-third octave result; and S06, constructing an MMS index, screening a cavitation sensitive frequency band, and comparing a prior characteristic frequency band to realize cavitation identification. By means of the method, the target of recognizing the cavitation phenomenon of the centrifugal pump can be achieved in the early cavitation stage of the centrifugal pump from the perspective of the excitation source, and the method can be used for the field of centrifugal pump fault diagnosis and the like.

The invention discloses a large-scale space splicing molded surface adjusting method based on a Maxwell kinematics supporting adjusting mechanism, and belongs to the technical field of large-scale space splicing molded surface adjusting.The method comprises the steps that S1, firstly, a specific implementation instrument for adjustment is determined, and the molded surface precision requirement of a splicing structure of the specific implementation instrument is determined; s2, according to the molded surface precision requirement, the molded surface precision requirement of the splicing structure is quantitatively expressed as a movable platform pose change parameter; s3, obtaining the variable quantity d of the total length in the branch chains according to the pose change parameters of the movable platform; s4, calculating the pre-tightening force according to d, solving the constructed constraint equation by using a generalized inverse matrix, and calculating the elastic deformation of each ball head according to a Hertz contact theory; s5, determining the driving length of the linear driving pair according to the elastic deformation; s6, the motor is controlled according to the driving length, and adjustment work is executed; according to the large space splicing molded surface adjusting method based on the Maxwell kinematics supporting adjusting mechanism, the positioning accuracy of a precise instrument can be improved, and the control difficulty is reduced.

The invention discloses a signal processing method, device and equipment, and belongs to the technical field of signal processing. The method comprises the steps that driving signals sent by a plurality of TX electrodes in parallel are modulated through a coding matrix, the coding matrix comprises a column full-rank matrix with the size being M * N or a row full-rank matrix with the size being N * M, both M and N are positive integers, and M is larger than N; the sensing signal is decoded through a decoding matrix, the decoding matrix is a generalized inverse matrix of a coding matrix, the sensing signal is obtained through sensing of a plurality of RX electrodes, a coupling capacitor exists between any RX electrode in the plurality of RX electrodes and each TX electrode in the plurality of TX electrodes, and the coupling capacitor is determined by a signal decoded by the decoding matrix. According to the method, the driving signal is modulated by using the non-square matrix type coding matrix when the touch panel adopts the CDM driving mode, so that the SNR (Signal to Noise Ratio) of an induction signal generated when the touch panel adopts the CDM driving mode is improved.

The application discloses a mechanical arm control method based on deep reinforcement learning, which comprises the following steps: a servo controller is designed, a mechanical arm visual servo system is built, an adaptive servo gain DDPG model is trained, a DDPG training environment is built, DDPG training parameters are set and DDPG model training is completed, expected images when the mechanical arm grabs a target object are set and acquired, and expected features are extracted, current target images are acquired, current features are extracted, an extended Kalman filtering method is used to estimate a depth value, the depth value is substituted into calculation of an image Jacobian matrix generalized inverse matrix, meanwhile, expected features and current features are operated to obtain image feature error e, according to a current state of the mechanical arm visual servo system, an adaptive servo gain algorithm based on deep deterministic policy gradient DDPG which is trained is used to determine a servo gain value, and the servo controller is used to perform servo control on the mechanical arm. The application has the characteristics of stronger stability, higher robustness and faster convergence speed.

The invention relates to a cigarette formula obtaining method and device, computer equipment and a readable storage medium. The method comprises the steps of obtaining key chemical components associated with a standard formula corresponding to a target cigarette, generating a generalized inverse matrix corresponding to each single-material cigarette based on the inventory and the key chemical components of each single-material cigarette, generating a candidate formula of the target cigarette according to each generalized inverse matrix and the standard formula, and outputting the candidate formula of the target cigarette according to the candidate formula. And finally, adjusting the candidate formula by utilizing the standard formula and the inventory of each single cigarette to obtain a target formula of the target cigarette. Key chemical components associated with a formula serve as entry points of tobacco material selection, generalized inverse matrixes corresponding to the single tobacco materials are generated on the basis of the inventory and the key chemical components of the single tobacco materials, candidate formulas are generated on the basis of a standard formula and the generalized inverse matrixes, the candidate formulas are adjusted by referring to the inventory of the standard formula and related ingredients, and the tobacco material selection accuracy is improved. Therefore, the target formula of the target cigarette is obtained, and the acquisition accuracy of the target formula is ensured.

The application discloses a method for obtaining human eye wavefront aberration and a human eye aberration measuring device, and relates to the technical field of optical instruments. The method for obtaining human eye wavefront aberration is applied to a Hartmann sensor and comprises the following steps: calculating an actual slope matrix according to a wavefront light spot diagram obtained by the Hartmann sensor; multiplying a generalized inverse matrix of a standard slope matrix with the actual slope matrix to obtain a Zernike coefficient matrix; subtracting the product of the actual slope matrix, the standard slope matrix and the Zernike coefficient matrix to obtain a residual error matrix; and updating the Zernike coefficient matrix according to the residual error matrix until the residual error matrix of the updated Zernike coefficient matrix is less than or equal to a preset value. The standard slope matrix is a matrix formed by the slope of a wavefront with all Zernike polynomial terms superimposed. The technical scheme provided by the application can improve the accuracy of the obtained human eye wavefront aberration.

The application provides a kind of image semantic segmentation method and system based on generalized convolution, and the information of feature layer is restored at pixel level by down-sampling convolution layer and deep feature layer, which is based on the mathematical basis of generalized inverse matrix, and can realize pixel-level recovery of information by transforming convolution matrix. On the one hand, generalized convolution can restore the size of feature map, and can construct a new up-sampling method for image semantic segmentation field. On the other hand, the up-sampling method of accurate matrix can make the up-sampling and down-sampling of the network symmetrical, so that only the parameters of the prediction part need to be fine-tuned when deploying.