28results about How to "Reduce memory consumption" patented technology

The application discloses a low-depth sequencing population genotype filling calculation memory optimization method, which comprises the following steps: carrying out block division on a single nucleotide polymorphism site (SNP), and setting a checkpoint value according to the block division; and carrying out calculation on a forward auxiliary variable and a backward auxiliary variable according to the set checkpoint value, and calculating a haplotype observation conditional probability according to the block division. The application has the advantages of simple principle, effective reduction of calculation memory consumption, improvement of genotype filling efficiency and the like.

This invention discloses an adaptive filtering method based on the multi-kernel Nystrom method, belonging to the field of kernel adaptive filtering. The multi-kernel Nystrom method combines multiple kernel functions, which can include not only Gaussian kernels but also other different kernel functions, mapping the original data to multiple different independent feature spaces. Compared with the traditional single-kernel Nystrom method, the multi-kernel Nystrom method has better filtering accuracy and becomes less sensitive to the choice of kernel parameters. Compared with other multi-kernel adaptive filtering methods, it can significantly reduce computational complexity, time, and memory consumption. Subsequently, this invention introduces the proposed multi-kernel Nystrom method into kernel adaptive filters for the first time. Experiments show that this algorithm can achieve better filtering accuracy than current state-of-the-art kernel adaptive filtering algorithms with relatively low computational complexity.

The RWKV-based novel translation model inference method and system includes the following steps: 1) Collecting novel translations and extracting parallel corpora, and using dynamic MicroBatch concatenation technology for sequence compression; 2) Introducing a lightweight group query attention mechanism on the basis of the RWKV architecture, directly obtaining KV information from the Embedding layer to build the model; 3) Employing a sublinear complexity hybrid parallel training mode, combined with a global scalar scaling FP16 mixed precision strategy for training; 4) Applying a hierarchical distributed heterogeneous architecture, offloading the optimizer to low-performance devices and performing gradient compression transmission; 5) Outputting the translation using a joint decoder and dynamic batch inference technology. This invention, through the above method and system, effectively reduces the computational complexity and memory usage in the long text translation process, improves model training efficiency and inference throughput, and significantly improves the translation efficiency and contextual coherence of ultra-long texts.

The application discloses a power maintenance action prediction method based on target consistency screening and a bidirectional state space, first constructs an observation side video feature sequence, then carries out action semantic token coding, probability weighted mapping and time sequence modeling, obtains action semantic context representation, constructs a joint input vector based on the action semantic context representation, carries out time sequence modeling and distribution construction through a gate recurrent unit, obtains a feature side target distribution, and extracts an observation action representation; subsequently, a future action prediction model based on a bidirectional selective state space is used for prediction, and multiple candidate future action sequences are output; finally, an action side posterior distribution is constructed, a target consistency score is calculated, and the candidate future action sequence with the minimum score is selected as an optimal action prediction result. The application explicitly models statistical correlation and potential target distribution of maintenance actions, introduces a bidirectional selective state space diffusion generation architecture, and realizes accurate and logical prediction of power maintenance actions.

The application discloses a point pair registration method and system based on feature guidance and double constraints, and belongs to the technical field of computer vision and robot grabbing. The method aims to solve the problems of recognition failure and insufficient estimation accuracy of traditional methods when parts are blocked and stacked, and point cloud features are not obvious. The scheme adopts a two-stage optimization strategy. Firstly, in the coarse registration stage, an adaptive sampling PPF algorithm guided by geometric features is proposed. By preferentially sampling points in the edge, curvature and other feature obvious areas, the robustness and efficiency of matching are improved. Secondly, in the fine registration stage, an ICP algorithm with multi-constraint optimization is proposed. The pose result obtained by coarse registration is used as the initial value. Through rotation and angle constraint, a multi-objective optimization function is constructed to guide the fine registration process to converge in the correct direction and avoid falling into local optimum, thereby ensuring the accuracy and physical feasibility of the final pose estimation. The application significantly improves the accuracy of industrial part pose recognition.

This invention discloses a brain-inspired optimization method for sentence structure-based self-attention mechanisms, comprising the following steps: acquiring a large-scale unlabeled corpus, preprocessing the corpus text, and identifying sentence boundaries; constructing structured attention constraints based on sentence boundaries: constructing intra-sentence attention constraints for ordinary characters to allow direct attention computation between characters within a sentence, constructing intra-sentence compression constraints for sentence boundary positions to form sentence-level compressed representations, and constructing cross-sentence constraints to allow both ordinary characters and boundary positions in the current sentence to access the boundary positions of the previous sentence; applying the structured attention constraints to the self-attention modules of each layer of a pre-trained language model for training or fine-tuning based on the corpus; and performing performance testing and brain-likeness analysis on the optimized model. This invention reduces the computational scale and memory consumption of the attention matrix while retaining long-range information integration capabilities, thus lowering computational complexity.

This invention discloses a large-scale power grid simulation method, program product, and storage medium based on algebraic multigrid optimization. The method includes: parsing the netlist of a large-scale power grid to establish a power grid topology matrix; extracting global feature vectors based on the netlist and the power grid topology matrix; compressing the power grid topology matrix to obtain a compressed feature matrix; constructing and training a regression model, which is used to predict the solution time based on the compressed feature matrix, global feature vectors, and coarsening parameters; using the minimum output of the regression model as the objective function, and employing simulated annealing to find the optimal coarsening parameters; establishing a power grid simulation system Ax = b; and solving the power grid simulation system using the algebraic multigrid method with the optimal coarsening parameters to obtain the power grid unknown vectors, thus realizing large-scale power grid simulation. This invention improves solution efficiency and reduces memory consumption while ensuring accuracy.

The application discloses an acquisition method and device for frequent binomial sets, terminal equipment and a storage medium. The number of occurrences of an element of a data set is continuously detected and counted. If the count value of the number of occurrences of the element reaches a preset threshold value, the element is counted into a partial frequent one-item set. The preset threshold value is the minimum support degree set by a user according to actual needs. Item pairs formed by the data set according to the partial frequent one-item set are screened. Hash operation is performed on the screened item pairs. The frequent binomial set is acquired according to the result of the hash operation and the preset threshold value. Thus, the acquisition method for frequent binomial sets provided by the application improves the PCY-Multistage algorithm. Hash operation is only performed on item pairs that meet the partial frequent one-item set, thereby reducing the device memory consumption when calculating the frequent binomial set and improving the user experience.

The application discloses an intelligent garbage classification method and system based on an ARM architecture, and belongs to the technical field of image processing.The application embeds an AlexNet neural network into the ARM architecture, and is used for garbage classification and identification, so that the requirements of high accuracy, short identification time and model updatable are met.Meanwhile, online knowledge distillation technology is used for category incremental learning, so that the performance of the embedded AlexNet network is further improved, and the forgetting problem caused by the incremental learning is avoided.In addition, a plurality of characteristic data are collected through an external multi-sensor, and a self-adaptive weighted fusion algorithm is used, wherein the weight uses a Q-learning algorithm to avoid misjudgment caused by hidden image information and unobvious image characteristics, and the accuracy and generalization ability of garbage classification are improved.

This application relates to a map construction method. Addressing the technical problem that existing environmental occupancy map construction methods, while aiming to improve query and update efficiency and significantly reduce memory usage, struggle to support real-time, large-scale autonomous navigation tasks in dynamic environments, this application provides a boundary representation-based map construction method and related apparatus. The map employs a global-local map framework. The local map includes the occupancy status of surrounding voxels centered on the robot, while the global map includes the voxel occupancy status in the boundary graph. The voxel occupancy status in the boundary graph is determined by using adjacent voxel pairs to represent the boundary surface layer separating free voxels from adjacent voxels, thus obtaining boundary voxels and constructing the boundary graph. A data structure combining two-dimensional hashing and one-dimensional arrays is used to manage the boundary voxels. Under this data structure, the occupancy status of any voxel is determined through the boundary graph. This application effectively improves query and update efficiency and significantly reduces memory usage.

This invention relates to a method and system for identifying and repairing holes in geometric models for fluid simulation, belonging to the field of fluid simulation technology. It solves the problems of unfriendly operation and low repair efficiency in existing technologies. The method includes: acquiring the geometric model of the fluid simulation and mapping the geometric model to obtain model data; receiving a hole identification command, identifying holes based on the model data, acquiring a set of hole data, and displaying the number of holes, a positioning button, and a repair button in the hole repair attribute panel; receiving a command for any positioning button, displaying the hole in the geometric model in the view area based on the hole corresponding to the positioning button; receiving a command for any repair button, repairing the corresponding hole based on the repair button, and updating the model data. This achieves rapid identification and repair of holes in the geometric model.

The application provides a global data reconstruction method and device based on position mapping and a storage medium, and relates to the technical field of data processing. The method comprises the following steps: for each source data in at least two source data, a mapping relationship between global data and the source data is respectively constructed, and the mapping relationship is stored as index data independent of the source data, wherein the index data corresponding to different source data is different; in response to a reconstruction instruction, the positions of target data segments corresponding to each data point in the global data in the source data are determined based on the index data; and the target data segments are extracted from the source data for fusion to generate the global data. The application realizes the decoupling of the mapping logic and the source data, avoids the residence of full data in the memory, significantly reduces the memory consumption, and improves the calculation efficiency and data reusability.

A kind of multi-outlet neural network design method and device for edge scene dynamic resource, its method includes: (1) given pre-training neural network, generate effective branch library, and screening is carried out;(2) further reduce memory occupation to the branch in branch library by self-merging, the accuracy of branch is quickly recovered by retraining through few-sample knowledge distillation after self-merging, and form branch candidate library;(3) based on priority, the optimal branch combination that satisfies memory and time requirement is found out to branch candidate library by depth-first search;(4) after selected branch combination is received by on-device scheduler, the branch of pre-training neural network is updated in real time by SBRAN component.The multi-outlet neural network with heterogeneous and dynamic characteristics is converted from traditional single-outlet neural network, and neural network can more flexibly adapt to the change of memory resource, so as to improve running efficiency and accuracy on edge device.

The invention discloses a picture generation method and device, a storage medium and equipment. The method comprises the steps of obtaining sparse voxel illumination data corresponding to a highest detail level model in a virtual scene; determining a cascade coordinate space corresponding to the virtual scene according to the spatial position and the observation depth of the virtual camera, and mapping the world coordinates of each detail hierarchy model into the cascade coordinate space; determining a boundary region of the adjacent detail hierarchy models in the cascade coordinate space; obtaining original voxel data of the current detail hierarchy model corresponding to the boundary region; the sparse voxel illumination data is loaded in the boundary region, and the sparse voxel illumination data and the original voxel data are mixed to generate a rendering picture of the virtual scene, so that smooth visual transition of the illumination effect between adjacent detail hierarchy models is realized, and the memory consumption is effectively reduced.

This application provides a method, apparatus, device, storage medium, and program product for generating SQL test cases, relating to the field of database technology. The method includes: generating a root path based on a breadth-first search strategy; splitting the root path into multiple intermediate paths based on an intermediate path generation algorithm; and converting the intermediate paths into SQL test cases through a path state recording mechanism, which includes a breakpoint-based continuation generation function. This method aims to reduce memory consumption, thereby improving the stability and efficiency of SQL test case generation.

This invention discloses a deep learning-based dual-modal image segmentation method and related apparatus for rectal cancer lesions. This invention achieves collaborative segmentation of T2WI and DWI modal medical images without explicit registration. Specifically, it utilizes T2WI encoding branches, DWI encoding branches, and a bidirectional cross-modal feature fusion module to achieve adaptive association and soft registration between T2WI and DWI modalities, ensuring high robustness of the model in complex images and effectively identifying lesions with multiple morphologies and blurred boundaries. Simultaneously, the decoding module performs weighted fusion of semantic features at different levels, improving the segmentation accuracy of small lesions and boundary regions, resulting in more consistent predictions and sharper boundaries. The combined sliding window and result fusion significantly reduces memory consumption and improves the continuity of segmentation results while maintaining prediction consistency.

The application discloses a multi-modal three-dimensional target detection method based on structural feature and semantic feature fusion, and belongs to the technical field of target detection.The application solves the problems of low precision, poor real-time performance and insufficient robustness of the existing method.The application constructs an explicit matching fusion mechanism of low-layer structure guided initialization, high-layer semantic optimization and time sequence expansion, projects 3D boundary boxes of a laser radar and a camera to multi-view image planes, calculates geometric similarity and category consistency constraints of 2D projection regions, and constructs an explicit matching graph in combination with low-layer structure information and high-layer semantic features.The application guides weighted aggregation of cross-modal target features with high confidence through a sparse matching graph, aligns historical frame targets to a current frame coordinate system to construct a space-time matching graph, aggregates historical information through expansion to a time sequence dimension, and realizes efficient and accurate three-dimensional target detection of the laser radar and the visual multi-view camera.The application method can be applied to multi-modal three-dimensional target detection.

This invention discloses an efficient method for constructing a high-precision geological attribute model for transparent geology, including the steps of constructing a representation method for the geological attribute model; the steps of constructing access operations for the geological attribute model; the steps of constructing an octree operation for the geological attribute model; the steps of constructing the octree structure of the geological attribute model; and the completion of the geological attribute model construction. This invention also discloses a system for implementing the aforementioned efficient method for constructing a high-precision geological attribute model for transparent geology. This invention not only achieves the construction of the geological attribute model, but also boasts higher reliability, better accuracy, and higher efficiency. The construction process consumes less memory, requires less storage space for the constructed model, and facilitates relatively easy dynamic updates.

The application discloses an image classification method and device based on a time-reversible pulse Transformer. The application proposes a time-reversible pulse Transformer neural network based on a multi-scale calculation architecture, which can output four image features of different resolutions for downstream visual tasks, and can directly use the last layer feature for image classification. Meanwhile, the reversible pulse Transformer main network adopts a time-reversible structure at a macro level, and pulse features of adjacent time steps can be calculated from each other, thereby saving the memory consumption of intermediate variables and reducing the training display memory. Compared with a similar structure of a non-reversible pulse Transformer, the training memory is obviously reduced, and the accuracy on the image classification task is significantly improved.

The present application relates to the field of artificial intelligence and data processing technology, specifically to a temperature control sensor data completion method based on a generative adversarial network, comprising obtaining a time series data matrix of a temperature control sensor containing missing data; constructing an adversarial model containing a generator and a discriminator; inputting the data matrix into a forward topological hard embedding layer, calculating the time and space partial differential physical residual and embedding the neuron transmission equation to output a latent feature matrix; inputting the latent matrix into an active feature selection module, eliminating redundant vectors and generating a synthetic sequence matrix; inputting the synthetic sequence and the real sequence into the discriminator to obtain the confidence, calculating the composite loss of the adversarial, reconstruction and residual constraints; using a gradient penalty mechanism to iterate the network parameters until convergence, and finally outputting the target temperature time series to complete the matrix. The present application effectively solves the problem of missing sensor data by introducing partial differential physical residual constraints and active feature selection, improving the adversarial network interpolation accuracy and the reliability of the time series sequence.

The application relates to the technical field of natural language processing, and discloses an NLP model inference acceleration method based on elastic loading, which comprises the following steps: performing horizontal cutting on the hierarchical direction of a natural language model according to a horizontal cutting angle; extracting an attention matrix and a full connection layer in the natural language model, and performing vertical cutting on the attention matrix and the full connection layer according to a vertical cutting angle; calculating the hierarchical importance of a cutting level according to a target scene, and calculating the importance of a cutting slice according to the target scene; generating a submodel of the natural language model according to the hierarchical importance and the importance of the cutting slice; and interleaving and superimposing model loading tasks and inference calculation tasks in the submodel to perform the tasks. In addition, the application also relates to the blockchain technology, and medical scene data can be stored in the nodes of the blockchain. The application further provides an NLP model inference acceleration device, equipment and storage medium based on elastic loading. The application can improve the model inference efficiency and has great application prospects in the medical field.

The present application relates to magnifying glass technical field, specifically to a kind of magnifying glass special effect implementation method under Wayland environment, comprising the following steps: step S1: synthesizer creates structure for magnifying glass, and based on structure initialization default state of magnifying glass;Default state includes magnifying mode;Step S2: the core processing function of magnifying glass is registered to synthesizer;Step S3: when user activates magnifying glass, it is judged that the magnifying mode selected by user, and initialization is carried out based on the magnifying mode selected by user;Step S4: synthesizer carries out rendering preprocessing based on the magnifying mode selected by user;Step S5: after rendering preprocessing is completed, synthesizer executes rendering process, to complete magnifying glass special effect.The scheme of the present application makes that magnifying glass is always with the focus of user as center, improves the reliability of user barrier-free use.

The application discloses a hyperspectral image reconstruction method and system based on mask logic coupling and a deep unfolding state space model, and belongs to the technical field of hyperspectral image processing. The method acquires a two-dimensional compressed observation image and a corresponding physical mask output by a single-frame snapshot compression imaging system, and inputs a weight-shared deep unfolding iterative network after dark current correction, noise suppression and normalization processing. In each iteration stage, the spectral-spatial joint features of the physical mask are extracted and a gating signal is generated, the gating signal is used for modulating the hidden state transition process of the state space model, and coupling and fusion of the physical mask priori and the deep features are realized. Meanwhile, a physical consistency mapping, a dynamic step updating and a recursive residual redistribution mechanism are introduced to constrain and compensate the reconstruction process. Finally, a reconstructed hyperspectral data cube is output.

The present application relates to an asymmetric mask distillation method for small mask autoencoder pre-training, comprising: performing asymmetric masking on video inputs of a student encoder and a teacher encoder to obtain a plurality of tubular tokens as feature inputs of the teacher encoder and the student encoder, wherein a masking rate of the video inputs of the teacher encoder is less than a masking rate of the video inputs of the student encoder; and performing feature alignment between the student encoder and the teacher encoder.

The invention discloses an electrocardiosignal analysis method based on symbol coding and large language model fine tuning, and belongs to the field of medical information engineering. The method comprises the following steps: recording a user question oriented to electrocardiogram analysis according to 12-lead continuous electrocardiogram data collected by a sample set, and recording a corresponding electrocardiogram characteristic description text as label data; preprocessing the electrocardio data; extracting key points of the electrocardio data signals; determining a bidirectional mapping rule of the voltage value and a lower case alphabet; determining a bidirectional mapping rule of the time interval and the capital letter table; encoding the electrocardiosignal; performing fine tuning training on the large language model; obtaining a weight value and a feature description text; resampling and reconstructing the electrocardiosignal by using linear interpolation and wave crest alignment; and annotating the attention degree of the electrocardiosignal model. According to the method, a test is carried out on a PTB-XL data set, the Bleu-4 index of a text analysis result generated on the same test set reaches 43.37, the Rouge-L index reaches 53.98, good performance is achieved, and the occupation of a video memory is reduced to 17.5 GB.

The application relates to a classification method, device and equipment for pathological images and a storage medium. The method comprises the following steps: processing a full-view digital pathological image to be detected to obtain a down-sampling image block; constructing a distillation and adversarial network, training the distillation and adversarial network, and extracting a classification sub-network from the trained distillation and adversarial network; inputting the down-sampling image block into the classification sub-network for classification to obtain an initial classification result; and obtaining a classification result corresponding to the full-view digital pathological image to be detected based on the initial classification result. According to the method, small data amount down-sampling image blocks can be input into a light classification sub-network to quickly and efficiently classify corresponding pathological images.

A neural network method, system and device for generating a three-dimensional hologram, the method of which is based on an improved Holo-U2Net network to construct a two-layer nested U-shaped structure, the outer layer U-shaped structure comprising 11 residual U-shaped modules RSU, and the inner layer U-shaped structure residual U-shaped module RSU, for extracting multi-scale features and high-frequency features; the outer layer U-shaped structure allows the network to be deeper and to fuse the feature maps output from the bottleneck layer residual U-shaped module RSU and each residual U-shaped module RSU in the encoding stage; the network extracts multi-scale features within the block and aggregates multi-level features between the blocks, having stronger fitting capability and the ability to generate high-quality holograms; a lightweight Fourier convolution module is implemented in the residual U-shaped module RSU, comprising convolution, batch normalization, a linear rectifier function, and the addition of feature information obtained in the Fourier domain, further improving the ability of the Holo-U2Net network to generate high-quality holograms; the system and device can generate high-quality three-dimensional holograms based on the neural network method for generating a three-dimensional hologram, and have the advantages of high image quality, short time consumption and low memory consumption.