154 results about "Sparse vector" patented technology

The invention belongs to the field of natural language processing, particularly relates to a multi-modal document generation method and device based on multi-agent collaboration, and aims to solve the problems that an existing method is low in intention recognition accuracy, limited in retrieval range and not professional enough in content generation. The method comprises the following steps: generating a structured template; analyzing the text input by the user to identify a writing intention, and determining a target template; vectorizing each candidate resource feature to obtain a corresponding sparse vector, a dense vector and a knowledge vector, and performing semantic alignment; extracting context features of the input text, respectively performing multi-path retrieval recall, evaluating and sorting recall results, and screening out target features; and constructing a thinking chain in combination with the knowledge graph, and generating a multi-modal document according to the target template. According to the method, the outline structure can be extracted, the picture / table style can be recognized, the templates adaptive to different document types can be dynamically generated, and full-process automation from user input to document output is achieved.

The invention discloses an enterprise financial document unified management system and method based on distributed storage, and relates to the technical field of enterprise financial document management, and the system comprises a metadata processing module which is connected to a distributed storage node cluster and is used for converting the metadata attribute of an enterprise financial document into a high-dimensional sparse vector. According to the enterprise financial file unified management system and method based on distributed storage, efficient management of multi-dimensional financial file data is realized through dynamic metadata topology reconstruction and a dual-channel index optimization mechanism. A vectorization metadata packaging technology is adopted to convert traditional discrete attributes into high-dimensional association vectors, and quantum annealing path optimization and a three-section aggregation verification strategy are combined, so that the response time of multi-condition combination query is shortened compared with that of a traditional scheme, and the network bandwidth consumption is reduced.

The invention relates to the technical field of mechanical fault diagnosis, and discloses a gearbox fault diagnosis method and system based on lightweight variational Bayesian learning, and the method comprises the steps: collecting a to-be-diagnosed vibration signal of a gearbox, carrying out the preprocessing of the to-be-diagnosed vibration signal, and obtaining an original vibration signal and the fault feature frequency of the original vibration signal; determining amplitude modulation-frequency modulation sparse combination representation of the original vibration signal according to the fault characteristic frequency of the original vibration signal, and constructing a joint probability model according to classification distribution containing sparse vectors and the amplitude modulation-frequency modulation sparse combination representation of the original vibration signal; performing variational Bayesian inference solution on the joint probability model by using the non-overlapping sub-sequence of the original vibration signal and natural gradient optimization to obtain posterior probability estimation of a sparse coefficient; and determining an activation component according to the posterior probability estimation of the sparse coefficient and the sparse precision parameter, and matching the activation component with a pre-established multi-scale amplitude modulation-frequency modulation sparse dictionary to obtain a fault type and a confidence coefficient thereof.

The invention provides a highway intelligent operation and maintenance question-answering system based on a large language model, and belongs to the technical field of natural language processing. The system takes a large language model as a core reasoning engine and combines a domain knowledge base and an RAG technology to realize accurate question and answer of highway operation and maintenance; the method comprises the following steps: based on original knowledge data cutting, generating a title through a large language model, and customizing a knowledge base; receiving query, analyzing an intention by using a large language model, and matching to generate a function; the query is rewritten by using a large language model, dense and sparse vector query is generated, and a double-layer retrieval mechanism is formed; a two-step recall mode is utilized, coarse-grained recall is firstly carried out, then a recall result is subjected to fine-grained optimization through a screening mechanism, and a reasoning text is generated; and finally, inputting the query and reasoning text into the large language model, and generating an optimal answer through single-round and multi-round questions and answers. According to the method, the professionality and reliability of answers are enhanced, and the technical problem that answers are incomplete and inaccurate in an existing question and answer system is solved.

A global navigation satellite system (GNSS) spoofing detection and classification technique is provided. An optimization problem is formulated at the baseband correlator domain by using an optimization algorithm such as the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm, for example. A model of correlator tap outputs of the intended received signal is created to form a dictionary of pre-computed waveform functions (e.g., triangle-like-shaped functions). Sparse signal processing can be leveraged to choose a decomposition of pre-computed waveform functions from the dictionary. The optimal solution of this minimization problem can discriminate the presence of a potential spoofing attack peak by observing the decomposition of two different code-phase values (authentic and spoofed) in a sparse vector output. A threshold can be used to mitigate false alarms. Furthermore, a variation of the minimization problem can be provided that enhances the dictionary to a higher resolution.

The invention relates to the technical field of information retrieval, and provides a multi-dimensional heterogeneous knowledge recall enhanced hybrid retrieval method and system.The method comprises the steps that texts recalled through keyword retrieval, sparse vector retrieval and dense vector retrieval are screened through a reciprocal sorting fusion algorithm, and a text type candidate knowledge list is obtained; based on user query, generating and checking a query statement through a large language model, and retrieving an entity-relationship-attribute triple from the knowledge graph library; based on user query, generating enhanced knowledge through a knowledge graph enhanced retrieval method fusing keyword retrieval, vector retrieval and community retrieval; and carrying out format alignment and duplicate removal on the text type candidate knowledge list, the triple result and the enhanced knowledge to form a multi-modal candidate pool, and carrying out reordering score calculation and ordering on each piece of recall knowledge in the multi-modal candidate pool through a reordering model and a business rule to obtain a final retrieval result. And the coverage blind area of single retrieval on heterogeneous knowledge is solved.

The invention discloses a data scarcity industrial vertical field-oriented knowledge-guided retrieval enhancement generation method, belongs to the technical field of natural language processing and industrial intelligence crossing, and can improve the retrieval accuracy and generation reliability of a large language model in an industrial scene. According to the method, a'sparse vector + dense vector 'mixed knowledge base is constructed, and general knowledge and industrial field texts are fused; after the model generates an initial text, judging whether external retrieval is needed or not through multi-dimensional evaluation token confidence; extracting attention weights for the low-confidence tokens, and screening key tokens to generate a retrieval query; dynamically adjusting the weight of a retriever based on a BGE-M3 model, and optimizing a retrieval result through reordering; the retrieval knowledge is converted into a context with an index, and a prompt template is constructed to generate a correction value iteration calibration text; and finally, optimizing the text format, and generating a structured response meeting industrial requirements. The method solves the problems of lack of professional knowledge of large language models in the industrial field, shallow retrieval and generation fusion, lack of knowledge calibration mechanisms and the like.

The invention provides a retrieval method and system suitable for different data densities, and the method comprises the steps: distributing each data point to a corresponding hash bucket based on a locality sensitive hash algorithm, and forming an initial inverted index based on an identifier of the hash bucket and the data points; processing data points in each hash bucket based on a hierarchical navigable small world insertion algorithm to form an HNSW graph, and forming an inverted index containing an HNSW graph index; obtaining a candidate bucket set of the query point; obtaining an HNSW graph corresponding to each candidate bucket in the candidate bucket set based on the constructed inverted index; retrieving from the HNSW graph corresponding to the candidate bucket based on a hierarchical navigable small world search algorithm to obtain preliminary retrieval results, and determining a global retrieval result according to the preliminary retrieval results and returning the global retrieval result. According to the method, efficient retrieval of dense vectors and sparse vectors can be carried out uniformly, the retrieval efficiency and precision are improved, the universality is high, and the complexity and development cost of the system are reduced.

The embodiment of the invention discloses a multi-modal document retrieval method and device, electronic equipment and a storage medium. The similar problems that in an existing retrieval enhancement system, multi-modal data (pictures and tables) in documents are difficult to process, the content stored in a knowledge base is incomplete, the correlation between retrieval results and problems is low, non-professionals are difficult to construct the knowledge base and a question and answer system by themselves, and the like can be solved. The method comprises the steps of obtaining a to-be-processed multi-modal document; preprocessing the to-be-processed multi-modal document according to a preset regular expression to obtain a target text list; according to the target text list, a first knowledge base and a second knowledge base are created, and an index relation exists between the first knowledge base and the second knowledge base; according to the dense vector and the sparse vector corresponding to the second knowledge base, the to-be-retrieved content is retrieved, a target retrieval result is obtained, and the target retrieval result is determined according to sub-segment results obtained through retrieval of the dense vector and the sparse vector.

The invention relates to the technical field of industrial robot operation and maintenance, and provides an industrial robot operation and maintenance management method and device based on an improved knowledge graph. The method comprises the following steps: establishing a dynamically updated operation and maintenance knowledge graph according to operation and maintenance management requirements of the industrial robot; obtaining a sparse vector of each entity in the operation and maintenance knowledge graph by using a similarity coding rule, and calculating the entity similarity between every two entities of the same category; sampling the operation and maintenance knowledge graph by different sampling strategies according to entity similarity to obtain a training sample, and training to obtain an operation and maintenance strategy prediction model; establishing an anomaly detection model; and determining a sensing end of the operation and maintenance knowledge graph according to the anomaly detection model, and determining a decision end of the operation and maintenance knowledge graph according to the operation and maintenance strategy prediction model so as to perform real-time anomaly detection and operation and maintenance strategy prediction on the operation monitoring signal of the industrial robot. According to the embodiment of the invention, the efficiency, precision and flexibility of operation maintenance management of the industrial robot can be considered.

The invention discloses a digital coding metamaterial forward-looking video imaging method and system based on sparse low-rank decomposition, and mainly solves the problems of motion blur and poor imaging stability in a non-sparse observation scene of a moving target in the prior art. According to the implementation scheme, echo data of N frames of observation scenes are collected; decomposing a scattering coefficient vector of an observation scene corresponding to each frame of echo into a low-rank vector of a background area corresponding to the observation scene and a sparse vector of a dynamic change area, and expressing a separation problem of the two change areas as a joint low-rank and sparse optimization problem; performing convex relaxation processing on the optimization problem, and converting the optimization problem subjected to the convex relaxation processing into an optimization problem in an unconstrained form to obtain a sparse low-rank decomposition-based metamaterial forward-looking video imaging model; and solving the constructed imaging model, and realizing multi-frame image reconstruction and video synthesis based on a solving result. According to the method, the imaging resolution and reconstruction stability of the non-sparse scene containing the moving target are improved, and the method can be used for intelligent security check and medical detection.

Disclosed embodiments include a computational memory system. The computational memory system includes at least one computational memory chip including one or more processor subunits and one or more memory banks formed on a common substrate. The at least one computational memory chip is configured to store one or more portions of an embedding table in the one or more memory banks, the embedding table including one or more feature vectors. The one or more processor subunits are configured to receive a sparse vector indicator from a host external to the at least one computational memory chip and, based on the received sparse vector indicator and the one or more portions of the embedding table, generate one or more vector sums.

Conventional Orthogonal frequency division multiplexing (OFDM) is unable to retain its orthogonality and suffers loss in performance in high Doppler circumstances. The present disclosure converts a signal received in delay-time domain into delay-Doppler domain and extracts a guard band region from the received converted signal. A 2-dimensional fast Fourier transform is performed on guard band region extracted from received converted signal. A 2-dimensional fast Fourier transform of the received converted signal is divided with the 2-dimensional fast Fourier transform of transmitted signal to extract phase information. A dictionary is created using a pre-defined set of values of delay and Doppler. A sparse recovery problem is formed for received converted signal using an orthogonal matching pursuit algorithm. One or more parameters are estimated by identifying one or more locations pertaining to the one or more columns corresponding to L significant non-zero locations comprised in sparse vector of sparse recovery problem.

The embodiment of the invention provides a semantic retrieval method and device based on reordering, equipment and a storage medium. According to the technical scheme provided by the embodiment of the invention, the method comprises the following steps: determining a first matching score corresponding to a dense vector of multiple pieces of structured data and a query text in a vector library, and a second matching score corresponding to a sparse vector of multiple pieces of structured data and the query text; carrying out fusion processing on the first matching score and the second matching score to obtain a fusion score corresponding to the multiple pieces of structured data, screening out a first quantity of candidate data from the multiple pieces of structured data according to the fusion score, and carrying out semantic processing on the candidate data to obtain semantic data corresponding to each piece of candidate data, according to the method, the candidate data of the first quantity are reordered according to the query text and the semantic data, and the retrieval results of the second quantity are screened out from the candidate data of the first quantity according to the reordering result, so that the balance between the retrieval efficiency and the precision is realized, and the semantic retrieval effect is improved.

The invention discloses an electric power equipment safety long text knowledge retrieval method based on sparse and dense joint enhancement, which comprises the following steps of: encoding an electric power equipment safety long text and retrieval query by utilizing a pre-training language model for performing supervision and fine tuning on an electric power equipment safety field label data set, and generating a semantic word vector; a weighted sparse vector fusing semantic information and a dense semantic vector enhanced through nonlinear transformation are generated through fine-grained semantic enhancement branches; respectively calculating sparse similarity based on a common item weight product and dense similarity adopting a double-distance weighted fusion strategy combining a cosine distance and an Euclidean distance; and carrying out weighted summation on the two similarity scores through the learnable weight to obtain a final similarity, and realizing accurate sorting retrieval of the long text knowledge. The method gets rid of dependence on global features of sentence vectors, can accurately capture local key information in a long text, and has high retrieval precision and semantic robustness.

This invention belongs to the field of wireless communications technology and discloses a large-scale multiple access method based on adaptive matching pursuit. First, some active users in a large-scale user terminal send spread-spectrum signals to a base station to perform uplink access transmission. Next, the base station receives the uplink signals of multiple users and models them as block-sparse vectors, using these vectors to reconstruct a new uplink receive signal equation. The base station then uses an adaptive matching pursuit algorithm based on the block-sparse model to detect active users and reconstruct the user-sent data. This invention is suitable for scenarios where large-scale terminals have sporadic burst uplink transmissions. It can achieve unauthorized random access for large-scale terminals while reducing user terminal overhead, thereby improving the uplink throughput of cellular networks.

The invention belongs to the technical field of robot control, and particularly discloses a wire harness assembly robot control method and system based on machine learning, and the method comprises the steps: firstly collecting the joint torque, the position error of an end effector, the posture change and other multi-dimensional sensor data in real time when a robot wire harness is plugged; intercepting and constructing a multi-channel contact event tensor by a sliding time window; inputting the effective contact event into a pre-training sparse auto-encoder, encoding to obtain a sparse vector, reconstructing a reconstructed tensor, and decomposing a sparse component of the effective contact event and a residual component of interference noise; identifying an effective contact event by judging whether the norm of the sparse vector L1 exceeds a first threshold value or not; and finally, calling a predefined fine tuning control strategy to execute an action response. Weak contact event recognition precision and action response accuracy are improved, assembly quality and safety are guaranteed, and the method is suitable for efficient control of the wire harness assembly robot.

The invention relates to the technical field of wireless communication, and discloses a robust semantic communication method, device and equipment based on sparse vector coding, and the method comprises the steps: extracting continuous semantic features of original data through a semantic encoder model; discretizing the continuous semantic features into index bits through a vector quantization method by using a learnable feature dictionary; the index bit is mapped into a sparse vector by adopting sparse vector coding, and the sparse vector is sent after being expanded by a codebook; and a receiving end identifies the index through a multipath matching pursuit algorithm and reconstructs the original data. According to the method, a semantic feature extraction model, a vector quantization technology and sparse vector coding are fused, and gradient approximation, joint loss function optimization and a transmission parameter dynamic adjustment mechanism are designed, so that the problems of poor digital transmission compatibility and weak channel adaptability of traditional semantic communication are effectively solved; and the semantic transmission reliability and the system resource utilization efficiency under the dynamic channel condition are improved.

The invention discloses an X-ray diffraction spectrum sparse reconstruction method and system based on a dynamic dictionary, and the method comprises the steps: generating an initial dictionary, and carrying out the dynamic optimization and updating of the dictionary, and obtaining a sparse dictionary; selecting an X-ray diffraction scanning angle subset based on the sparse dictionary, constructing an observation matrix, encoding an observation process, and obtaining compressed and sampled X-ray diffraction projection data; and sparse vectors are obtained from the obtained X-ray diffraction projection data after compressed sampling, and a reconstruction spectrum is obtained. And the accuracy of sparse representation is improved. The invention provides a high-throughput XRD (X-Ray Diffraction) material spectrogram analysis method combining dynamic dictionary learning, a compressed sensing theory and deep learning, and aims to improve the analysis efficiency and precision of crystal structure recognition and phase composition in new material research and development.

The invention provides a large language model modal expansion method and device based on parameter fusion and decoupling, and relates to the technical field of large language models. The method comprises the following steps: performing fine adjustment on a pre-training language model to obtain a plurality of multi-modal large language models; task vector extraction is carried out on each multi-modal large language model; sparsification is carried out on the original task vector by adopting a sparsification strategy to obtain sparse vectors, and fusion is carried out on the sparse vectors to obtain a fusion task vector; constructing model parameters according to the fusion task vector; constructing a mode-exclusive binary mask for each multi-mode large language model according to the fusion task vector; and constructing a fusion model according to the model parameters and the binary mask. The invention provides a multi-modal language model extension method with non-training fusion, modal decoupling, performance retention and continuous extension capabilities, which is suitable for efficiently integrating a plurality of MLLMs, reconstructing an original model structure, coping with application scenes such as continuous integration of new tasks and the like.

A computer-implemented method for detecting attribute value pairs from corpus data using a computer comprising a processor and a computer readable medium comprising instructions executable by the processor to at least: receive the corpus data comprising at least one pair; detect a layout template of the at least one pair; measure the merit of the layout template by determining at least one of (a) relative magnitudes of content probabilities based on a probability of the contents of an attribute cell and a probability of a corresponding value cell, (b) the validity of a name-value pair, or (c) the pointwise mutual information of a frequency matrix M corresponding to a sparse vector capturing context information of a word; and output detected attribute value pairs.

The invention discloses an article recommendation method based on RAG and knowledge graph guidance. The method specifically comprises the following steps: preprocessing and warehousing a structured abstract based on a large language model, adopting an offline processing mode of first abstracting and then warehousing, and utilizing the large language model LLM to obtain a structured abstract document through a two-stage prompt project; performing deep normalization on user input by utilizing the knowledge graph, converting all non-standard input into standardized entity names, and complementing the association relationship of the non-standard input and the standardized entity names; acquiring an article by adopting a sparse vector and dense vector mixed recall strategy; independent reordering and scoring links driven by a Qwen3 32B large model are introduced, a context-aware paired comparison task is designed for the large language model LLM, and a final recommendation result is ensured. The method has the advantages that it is ensured that recommended content is highly consistent with user requirements; professional recommendation services far beyond general schemes are provided; and the quality of recommendation results is greatly improved.

The invention relates to a bilateral sparse representation-based image compressed sensing method, which comprises the following steps of: obtaining an original image, and segmenting the original image to obtain a plurality of image blocks; performing bidirectional Haar wavelet transform on the image blocks to obtain a sparse matrix, vectorizing the sparse matrix, and obtaining a one-dimensional sparse vector; and carrying out compression and image reconstruction on the one-dimensional sparse vector to obtain a final reconstructed image. According to the method, the precision of the compressed sensing algorithm in the image reconstruction process can be greatly improved, and meanwhile, the calculation processing efficiency is remarkably improved.

The invention provides a time-space structured sparse optical coding deepwater heterogeneous network data synchronization method and system, and the method comprises the steps: carrying out the sparse decomposition of the time sequence clock deviation data of each current underwater node at a transmitting end device of a deepwater heterogeneous network based on a trained over-complete dictionary, and obtaining an original sparse vector; converting the dynamic compression observation data corresponding to the original sparse vector into an optical signal, and transmitting the optical signal to a receiving end device of the deepwater heterogeneous network by a transmitting end device through an umbilical cable; demodulating the optical signal into an observation vector through receiving end equipment, inputting the observation vector into the trained deep expansion network model, and outputting a reconstructed sparse vector by the deep expansion network model; and recovering the reconstructed sparse vector into a clock skew signal through the over-complete dictionary, and generating a clock synchronization calibration instruction to realize data synchronization of the corresponding underwater node. According to the invention, the problems of high bandwidth occupation, low signal ratio and poor reconstruction time efficiency existing in the underwater network time synchronization technology can be solved.

A vector similarity determination method and a vector search method. The vector similarity determination method comprises: acquiring a first dense-sparse vector and a second dense-sparse vector (S101); calculating a first similarity and a second similarity on the basis of the first dense-sparse vector and the second dense-sparse vector (S102); and determining the similarity between the first dense-sparse vector and the second dense-sparse vector on the basis of the first similarity and the second similarity (S103).

The invention discloses an indoor multi-target visible light positioning method based on compressed sensing, and relates to the technical field of visible light communication.The method comprises the steps that visible light signals are converted into sparse vectors, and autocorrelation signal vectors in the multiple visible light signals are aggregated to obtain the total power of the signals; when the compressed sensing algorithm CS measures indoor visible light transmission, an observation matrix is formed, measurement power is given to the observation matrix based on the total power, and a plurality of sparse vectors are recovered by using the measurement power; aggregating cross-correlation measurement vectors in the plurality of visible light signals, reconstructing sparse vectors to obtain a plurality of positioning sparse vectors, identifying positions of non-zero elements in the positioning sparse vectors, and determining positions of a plurality of targets; according to the process, sparsity and cross correlation in multi-target visible light signals are mined, a multi-target positioning problem in a complex environment is converted into a sparse recovery problem by using a compressed sensing (CS) method, and multi-target signals are distinguished and identified by using sparse weak signals, so that high-precision multi-target positioning is realized.

The invention provides a radar channel and communication channel estimation method and device of a communication and inductance integrated system, and belongs to the technical field of wireless communication. The method comprises the following steps: determining a joint sparse vector of a radar channel and a communication channel based on a sparse vector of the radar channel and a sparse vector of the communication channel; determining a channel dynamic sparsity probability expression based on the joint sparse vector and a space-time Markov model; solving the channel dynamic sparsity probability expression based on a three-layer Bayesian reasoning framework to obtain a channel dynamic sparsity probability; substituting the dynamic sparsity probability of the channel into a maximum likelihood estimation problem of the environmental perception parameters, and solving the maximum likelihood estimation problem to obtain the environmental perception parameters; and estimating a radar channel and a communication channel based on the environment perception parameters. According to the radar channel and communication channel estimation method and device of the communication and inductance integrated system, the estimation precision of the radar channel and the communication channel can be effectively improved.

The invention discloses a compressed sensing method and system based on a guaranteed-dimension semi-tensor product, a medium and equipment. The method comprises the following steps: acquiring a vector expression form of an original image; converting the obtained vector into a sparse vector representation through discrete cosine transform; generating a Gaussian random matrix, and performing dimensionality-preserving semi-tensor product operation through all-one vectors with weights to construct a measurement matrix; sampling the sparse vector by using the measurement matrix to obtain a compressed transmission vector; and the Gaussian random matrix and the transmission vector are sent to a receiving end, the receiving end reconstructs and recovers the original image based on an l1 norm optimization algorithm, and compressed sensing transmission is completed. According to the method, the measurement matrix is improved based on the semi-tensor product of the guaranteed dimension, and the measurement matrix construction method can reduce the size of the generated Gaussian random matrix, so that the memory is saved, the speed in the transmission process is higher, and the bandwidth cost is saved.

The invention relates to a verification code processing method. The method comprises the following steps: in response to a verification code acquisition request sent by a client, generating a first verification code for the verification code acquisition request; converting the first verification code into a first sparse vector based on a preset mapping rule; performing encryption processing on the first sparse vector to obtain an encrypted vector; and sending the encryption vector to the client to enable the client to reconstruct a first sparse vector based on the encryption vector, and converting the first sparse vector into a first verification code based on a preset mapping rule. By adopting the method, the security of verification code transmission and verification can be improved.

Certain aspects of the present disclosure provide techniques and apparatus for wireless channel estimation using machine learning. A sensing matrix is processed using a set of one or more layers of a machine learning model, based on a learned sparsifying dictionary, to generate a set of associated sparse vector representations. A channel estimation is determined based on output of a final layer of the set of one or more layers of the machine learning model.