51 results about "Candidate key" patented technology

The invention provides a point cloud key point generation method based on a key point detection network and a descriptor network, and the method comprises the steps: carrying out the random sampling of the key point detection network, obtaining candidate key points, carrying out the multi-level clustering with each candidate point as a center, expanding a sensing domain, and aggregating the neighborhood information, aggregated information is sent to a double-head attention mechanism to calculate a neighborhood point weight, and more representative key points are selected; the descriptor network receives the attention feature map of the detection network, carries out fusion learning on global information and local information of each key point, and calculates a descriptor and a direction vector for each key point. Secondary screening of the key points is carried out while the descriptors are optimized by using a joint matching loss function, and the key points with strong characteristic characterization force are obtained. Compared with the prior art, the time consumption is shortened by 90%, the key point repetition rate is 70%, the matching precision is 90% or above, and good robustness is achieved under different Gaussian noises.

A system and method for extracting structured key information from diverse document types using large multimodal models (LMMs) is disclosed. The invention employs a zero-shot analysis to identify candidate keys within an input document, then selects a document schema from a document schema database based on the identified keys. The LMM is prompted with the selected document schema to generate structured key-value pairs, with field constraints enforced by the document schema. Relationships among extracted keys are mapped to a graph representation, enabling robust handling of complex document layouts. The system supports nested structures, tabular data, and alias definitions for fields, and can update document schemas based on ground truth feedback. The resulting structured output is provided in a machine-readable format, enabling reliable and scalable document understanding across varied domains such as invoices, health cards, and driving licenses.

A kind of frame extraction method for road test vehicle image information, comprising: obtaining road test vehicle image information, and marking candidate key frame in road test vehicle image information;According to the marked candidate key frame, based on actual demand, relevant threshold setting and index calculation;According to the set threshold and calculated index, key frame is marked;According to the marked key frame and set threshold and calculated index, similar frame is marked;According to neural network model, no key target frame is marked;According to the marked key frame, similar frame and no key target frame, image information is frame extracted.The present application is highly adaptable to the business scene of automatic driving, and can more effectively solve the data redundancy problem of storage end.Free limited human resources, and then improve the execution efficiency of the entire automatic driving tool chain.And different content section image scene is processed by divide and conquer, and the frame extraction result is more reliable.

In some implementations, a device may obtain a data set that indicates scores for first terms based on association scores between the first terms and second terms. The association scores may be based on occurrences of the first terms in electronic documents that provide transcripts relating to telecommunication interactions, and occurrences of the second terms in electronic records that provide summaries relating to the telecommunication interactions. The device may process an electronic document, that provides a transcript relating to a new telecommunication interaction involving an agent device and a user device, to extract candidate key phrases from the transcript. The device may determine a relevance score for each candidate key phrase based on the score(s), of the data set, associated with the first term(s) in the candidate key phrase. The device may generate tag(s) for the new telecommunication interaction based candidate key phrase(s) associated with highest relevance score(s).

The present disclosure provides an information query method, device, equipment, storage medium and program product, relates to the technical field of data processing and computer, and particularly relates to the technical field of big data and intelligent search. The specific implementation scheme of the information query method is as follows: determining at least one candidate key information matched with the query information according to the query information; determining a first evaluation value for evaluating the candidate key information according to a key information evaluation parameter; determining at least one target key information from the at least one candidate key information according to the first evaluation value; and determining target content information according to the at least one target key information.

The invention proposes a method for constructing an original text reference library based on improved reinforcement learning, and the method comprises the steps: constructing a data set, carrying out the optimization of an online learning large model through combining with an online learning gradient updating strategy, and obtaining an optimized online learning large model; constructing a reference library based on a multi-layer index architecture according to an output result of the optimized LLM large model; for the optimized LLM large model, outputting a result, namely a plurality of candidate key expression statements, inputting the title, the text content and the corresponding candidate key expression statements into an auditing system, and performing manual auditing; inputting a to-be-verified article into the optimized online learning large model, outputting all candidate key expression sentences, and constructing a candidate expression sentence list; a structured phrase set and a sentence pattern structure regular expression formula are obtained from the candidate expression sentence list through a text analysis module, and phrase features are extracted through the structured phrase set; and performing retrieval in the reference library based on the multi-layer index architecture, and outputting a reference position and a similarity score.

The invention relates to the technical field of side channel analysis, and particularly provides a side channel analysis method and system based on bit decoupling and a medium, and the method comprises the steps: obtaining power consumption data during encryption, calculating an intermediate value, and carrying out the multi-label decoupling; extracting multi-scale features from the preprocessed energy traces, performing weighted fusion, and outputting global shared features; mapping the global shared features into initial bit features through linear projection, calculating a dependency relationship between the bit features by using an attention mechanism, and generating interactive bit features; the fusion proportion is dynamically adjusted through gating parameters, fusion features are input into a multi-label classifier, the posterior probability of each bit is obtained, and a loss function is adopted for training; and inputting a to-be-measured energy trace into the trained neural network model, outputting a prediction probability, and selecting a target key by calculating the confidence coefficient of the candidate key through traversal. According to the method, the quantity of required energy traces is greatly reduced, and meanwhile, the attack success rate in a low signal-to-noise ratio environment is remarkably improved.

The present application relates to the technical field of vehicle identity authentication, and discloses a vehicle distributed identity authentication method and system based on biological characteristics, which comprises the following steps: a user completes initial biological characteristic registration on a source vehicle, and biological characteristics are bound with random keys by a fuzzy binding algorithm to generate a binding factor and a biological characteristic hash value and store them in association; when the user uses a target vehicle, the target vehicle locates the source vehicle according to the user identity and requests verification. The source vehicle sends the binding factor and the biological characteristic hash value to the target vehicle; the target vehicle collects the current biological characteristics of the user, recovers the candidate key by decoding operation using the binding factor, and generates a temporary biological characteristic hash value; and the authentication is completed by comparing the temporary biological characteristic hash value with the received biological characteristic hash value. The method uses a decentralized hash transmission and fuzzy binding verification mechanism to realize efficient, safe and non-repeated cross-vehicle identity authentication on the premise of protecting the privacy of biological characteristics.

The application discloses a key management method and a storage device, and belongs to the field of storage security. The method comprises the following steps: deriving different types of key fragments from an authentication key according to different preset classification strategies, and storing the different types of key fragments in different storage partitions corresponding to the preset classification strategies; receiving a data access request, wherein the data access request comprises data index information; reading and verifying the legitimacy of candidate key fragments matched with the data index information from the different storage partitions based on different preset classification strategies, and determining the candidate key fragments passing the verification as valid key fragments; and aggregating the different types of valid key fragments to generate the authentication key, and obtaining target access data of the data access request through the authentication key. The application can effectively improve the security of key storage and the flexibility of key management.

This invention provides a hierarchical tree-structured attention routing method, apparatus, and device for ultra-long contexts. The key-value (KV) cache is partitioned into devices based on token positions and equipped with local Q-index subtrees. Each device searches its local subtree from the root, layer by layer, to obtain a local candidate key position index set. Simultaneously, the query vector is input into the Q-routing tree to select paths layer by layer, obtaining the activated attention expert set, memory expert set, and FFN expert set. Then, each device's attention expert and memory expert performs calculations only within the range corresponding to the local candidate key position index set and obtains local outputs. Finally, a Q-communication tree aligned with the device's physical topology performs a step-by-step tree-structured reduction of each device's local outputs and merges them with the local FFN outputs to obtain the layer output. Thus, the three stages of KV retrieval, expert routing, and cross-device reduction all converge to the logarithmic level through the tree structure, thereby achieving collaborative dimensionality reduction.

The application provides a video reasoning segmentation method and device based on reinforcement learning visual fine-tuning, and relates to the technical field of image processing. The method comprises the following steps: obtaining candidate key frames and video description text based on input text and a video to be segmented; inputting the input text, the candidate key frames and the video description text into a first-stage model to obtain an evaluation result of the candidate key frames; based on the evaluation result, retaining or updating the candidate key frames to obtain selected key frames; inputting the selected key frames, the video description text and the input text into a second-stage model to obtain positioning information of a target object in the selected key frames; and determining a segmentation result of the target object in multiple frames of the video to be segmented based on the positioning information of the target object in the selected key frames. The first-stage model and the second-stage model are obtained by training a pre-trained multi-modal large language model using reinforcement learning. The application can improve the accuracy of the video reasoning segmentation result.

The invention relates to the technical field of security evaluation of encryption algorithms, and provides a virtual shuffling security analysis method and system for bit slice misuse, and the method comprises the steps: creating a counter array which comprises a plurality of elements, and the index of each element corresponds to the value of a candidate key; for virtual shuffling, collecting a plurality of groups of tracks, screening out effective tracks for each track group, and if the number of the effective tracks is smaller than a threshold value, discarding the track group; performing LDA operation on the reserved track group to obtain a recommended key set, and for each candidate key in the recommended key set, adding one to the value of the index position of the corresponding counter array; and after all the track groups are processed, traversing the counter array, finding out the key with the highest count as a correct key, counting the count of the correct key, and obtaining the actual recommendation frequency of the correct key. And the risk of misuse of bit slice virtual shuffling in IF-ELSE implementation can be efficiently identified.

Provided is a data structure information generation device for generating, by using a large-scale language model for processing according to a prompt, data structure information in which pieces of position information about keys and values included in a table portion of data described in a table format is designated in a text format. The device comprises: an input reception unit for receiving an input of the table portion of the data described in the table format; a proposal unit for proposing, to an operator, candidates of the keys and the values analyzed from the table portion; an instruction reception unit for receiving, from the operator, an instruction for correcting or confirming the candidates of the keys and the values; a range designation reception unit for receiving, from the operator, designation of a range of the table portion in the data described in the table format; a data structure information generation unit for generating the data structure information on the basis of the confirmed keys and values and the designation of the range of the table portion; and a presentation unit for presenting the generated data structure information to the operator.

The application provides a data query method and device, computer equipment and a storage medium. In the method, a data holder splices second holder key data obtained by performing privacy intersection calculation on data items corresponding to candidate key values, and then encrypts the second holder key data by using first holder key data obtained by performing privacy intersection calculation, and sends the encrypted data to a data query party. The data query party calculates first query party key data by using key reference data obtained by performing privacy intersection calculation and a query key value, and decrypts the encrypted data by using the first query party key data, and then obtains corresponding second query party key data based on the first query party key data, so as to determine to-be-queried data items from the decrypted data by using the second query party key data. Only encrypted data is transmitted between the data holding equipment and the data query equipment, and no specific plaintext data is involved, so that the data security of both parties is protected.

The invention discloses a contract information identification method and a related product. The method comprises the following steps: acquiring a contract text and a metadata template; the metadata template comprises a plurality of texts to be retrieved; if the text word number of the contract text is greater than a preset threshold value, constructing a temporary vector knowledge base based on the contract text; performing semantic filling matching on the plurality of to-be-retrieved texts in the metadata template to obtain a filled metadata template; matching the filled metadata template with a temporary vector knowledge base to obtain a candidate key set; the candidate key set comprises a matching result of each to-be-retrieved text; the matching result at least comprises a candidate value; and verifying the candidate key set to obtain a contract information identification result. According to the method, while the capability of processing long documents is guaranteed, the extraction precision and robustness of key information in complex and non-standardized contract terms are remarkably improved, and then the accuracy of contract information identification is improved.

Embodiments of the present application relate to a positioning method and device of a mobile device, an electronic device and a storage medium. A key frame is acquired by a vision camera installed on the mobile device; mapping information and a feature descriptor of the key frame are determined; a bag-of-words model and a key frame image database are obtained according to the feature descriptor of the key frame; a current image frame is acquired, and the current image frame is converted into a current frame bag-of-words vector based on the bag-of-words model; a candidate key frame with the highest matching score with the current image frame is determined from the key frame image database according to the current frame bag-of-words vector, and a target key frame is determined according to the candidate key frame; a target node index corresponding to the target key frame is determined according to the mapping information, and pose information of the mobile device is calculated according to the target node index, so as to position the mobile device; that is, the prior pose of the mobile device is provided by the bag-of-words model, and the positioning accuracy is improved.

The invention relates to the technical field of vehicle identity authentication, and discloses a vehicle distributed identity authentication method and system based on biological characteristics, and the method comprises the steps: a user completes initial biological characteristic registration on a source vehicle; the biological characteristics and the random secret key are bound through a fuzzy binding algorithm to generate binding factors and biological characteristic hash values, and the binding factors and the biological characteristic hash values are stored in an associated mode; and when the user uses the target vehicle, the target vehicle locates the source vehicle according to the user identity identifier and requests verification. The source vehicle sends the binding factor and the biological characteristic hash value to the target vehicle; the target vehicle collects the current biological characteristics of the user, the candidate key is recovered through decoding operation by using the binding factor, and a temporary biological characteristic hash value is generated; and the authentication is completed by comparing the temporary Hash value with the received Hash value of the biological characteristic. According to the method, a decentralized Hash transfer and fuzzy binding verification mechanism is utilized, and efficient and safe cross-vehicle identity authentication without repeated input is realized on the premise of protecting the privacy of biological characteristics.

The invention discloses a social network public opinion propagation key user mining method and system, and relates to the technical field of network public opinion analysis, and the method comprises the steps: obtaining historical social network public opinion propagation data and a preset propagation topological graph; according to historical social network public opinion propagation data, performing key user mining on a preset propagation topological graph to obtain a candidate key user set; calculating a comment area atmosphere torsion score and post-torsion propagation income of each candidate key user in the candidate key user set; and respectively judging whether each candidate key user is a key user of a preset propagation topological graph or not according to the comment area atmosphere twisting score and the post-twisting propagation income of each candidate key user. According to the method, a key user group which has a twisting negative public opinion atmosphere and has a relatively good inhibition effect on further propagation of negative public opinions can be mined.

The application provides a hierarchical tree attention routing method, device and equipment for an ultra-long context, cuts a KV cache according to token positions to each equipment and matches with a local Q-index sub-tree, so that each equipment searches from a root to a layer in the local sub-tree to obtain a local candidate Key position index set, simultaneously inputs a query vector into a Q-routing tree to select a path layer by layer to obtain an activated attention specialist set, a memory specialist set and an FFN specialist set, then the attention specialist and the memory specialist of each equipment perform calculation only in a range corresponding to the local candidate Key position index set and obtain a local output, finally a Q-communication tree aligned with a physical topology of the equipment performs tree-like reduction on the local output of each equipment stage by stage to obtain a layer output after being fused with a local FFN output; thus, three links of KV retrieval, specialist routing and cross-equipment reduction are converged to a logarithmic level by means of a tree structure, and collaborative dimension reduction is realized.

The embodiment of the invention provides a key point detection method and device, a storage medium and a program product, and the method comprises the steps: determining at least one reference point corresponding to a target image in response to a key point detection request for the target image, and generating a heat map corresponding to the at least one reference point; for the heat map corresponding to any reference point, extracting a feature map from the heat map by using a key point detection model, and extracting a target feature from the feature map, the target feature being formed by fusing features of multiple dimensions; determining a plurality of candidate key points corresponding to the heat map based on a feature map extraction result and a feature extraction result; and searching a target key point corresponding to the reference point from the plurality of candidate key points, and determining a position coordinate of the target key point in the target image. According to the scheme of the embodiment of the invention, the accuracy of key point detection in the image is improved.

The invention provides a multi-hop reasoning method and device based on shielding knowledge activation, and relates to the technical field of artificial intelligence. Determining a corresponding candidate key phrase set from the initial query information by utilizing a natural language processing library, and detecting shielded key phrases in the candidate key phrase set; on the basis of retrieval conditions, a candidate document set is retrieved from a retrieval database through a dense retriever, and the retrieval conditions comprise the current round of questions corresponding to the current round of query and the shielded key phrases; determining a target document from the candidate document set, and on the basis of the shielded key phrases, the current round of questions and the target document, generating a next round of questions corresponding to the next round of query by using a large language model; and determining a final answer based on the initial query information, the next round of question and the single-hop termination condition. According to the method, the target document is locked, and the document is utilized to complement the shielded key missing information to generate the next round of problem, so that the accuracy of retrieval enhancement generation in a complex scene is improved.

The application provides a smart glasses key design method and device, an electronic device and a storage medium, and belongs to the technical field of optimization design. The method comprises the following steps: performing ergonomics simulation on each candidate key position based on finger operation stability and hand action posture data to obtain reachability scores of the candidate key positions; performing ergonomics simulation on each candidate key position based on historical mis-touch frequencies to obtain calibrated mis-touch probabilities of the candidate key positions; calculating comprehensive operation risk scores of each candidate key position in each type of use scenario based on the calibrated mis-touch probabilities, the reachability scores and mis-touch cost levels; determining optimal key positions corresponding to each type of use scenario based on the comprehensive operation risk scores, generating a use scenario and key position relationship mapping table, and completing the layout design of the smart glasses keys according to the use scenario and key position relationship mapping table. The application can improve the operation convenience and user experience of the smart glasses keys in different use scenarios.

The application relates to a key region identification method based on a four-dimensional characterization index. The method comprises the following steps: based on a sample set, calculating an information surface entropy, a maximum gradient consistency, a local-global similarity divergence and a local-global extreme point ratio to form a four-dimensional characterization index vector of a key region; fusing the four-dimensional characterization index vector based on a Pareto dominance relationship, dividing candidate key regions into different front levels through non-dominated sorting, calculating the crowding distance of each front region, and screening the key region in the test sample space in combination with the front level and the crowding distance. The method can improve the utilization efficiency of a new type of aero-engine digital twin test resources.

A method and system deliver media content. Multiple first requests for media content are received and first key phrases are extracted into a dictionary. Second media content requests and associated context are collected for media content served to a latent interest audience. The context is used to obtain candidate key phrases. For each of the contexts, the candidate key phrases that match the first key phrases are determined. A subset of the matching candidate key phrases are assigned to the campaign. The subset of matching candidate key phrases is expanded using semantic similarity and stored in a second table. A new media content request that does not include an explicit user identification is received. Based on the new media content request, new request key phrases are determined and compared to the expanded subset. Campaigns with intersections are selected and utilized to deliver media content.

In some implementations, a device may obtain a data set that indicates scores for first terms based on association scores between the first terms and second terms. The association scores may be based on occurrences of the first terms in electronic documents that provide transcripts relating to telecommunication interactions, and occurrences of the second terms in electronic records that provide summaries relating to the telecommunication interactions. The device may process an electronic document, that provides a transcript relating to a new telecommunication interaction involving an agent device and a user device, to extract candidate key phrases from the transcript. The device may determine a relevance score for each candidate key phrase based on the score(s), of the data set, associated with the first term(s) in the candidate key phrase. The device may generate tag(s) for the new telecommunication interaction based candidate key phrase(s) associated with highest relevance score(s).

In some implementations, a key performance indicator (KPI) monitoring controller device may receive, from a user interface device, an indication that a first key-set topic associated with a network device is to be monitored. The KPI monitoring controller device may determine a first key-set threshold, which may be a quantity of key-set instances that are permitted to be monitored for the network device, and which is selected from multiple candidate key-set thresholds based on a subscription status associated with the network device. The KPI monitoring controller device may determine a first set of key-set instances to be monitored by a KPI monitoring collector device, with a quantity of key-set instances associated with the first set of key-set instances being less than or equal to the first key-set threshold. The KPI monitoring controller device may transmit, to the KPI monitoring collector device, an indication of the first set of key-set instances.

The invention discloses a deep learning side channel attack method, system and device based on a Boosting integration method and a medium, and belongs to the technical field of cryptography security assessment, and the method comprises the steps: obtaining a side channel signal sequence and a real key tag; performing wavelet denoising on the side channel signal sequence to obtain a denoised signal sequence; extracting feature vectors according to the de-noised signal sequence to construct a training sample set; constructing a plurality of weak learners adopting a multi-layer sensor structure; giving initial sample weights to samples in the training sample set, and training a plurality of weak learners in sequence in multiple rounds of iteration; carrying out weighted fusion on the output of the plurality of trained weak learners to obtain integrated prediction probability distribution; and calculating accumulated scores of the candidate keys according to the integrated prediction probability distribution, sorting the accumulated scores, and obtaining a position index of a correct key as a guess entropy. According to the method, key leakage features are reserved through wavelet denoising, difficult case learning is reinforced through dynamic sample weighting, and robustness and generalization ability are improved through multi-model fusion.

A method and system deliver media content. Multiple first requests for media content are received and first key phrases are extracted into a dictionary. Second media content requests and associated context are collected for media content served to a latent interest audience. The context is used to obtain candidate key phrases. For each of the contexts, the candidate key phrases that match the first key phrases are determined. A subset of the matching candidate key phrases are assigned to the campaign. The subset of matching candidate key phrases is expanded using semantic similarity and stored in a second table. A new media content request that does not include an explicit user identification is received. Based on the new media content request, new request key phrases are determined and compared to the expanded subset. Campaigns with intersections are selected and utilized to deliver media content.

This invention discloses a key node detection method based on Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). This method primarily aims to optimize and improve the accuracy of key node detection in complex networks, enabling it to find an optimal set of non-dominated solutions suitable for various application scenarios. First, for the key node detection problem in complex networks, this invention designs a non-uniform attack cost, using PWC in non-cascade scenarios and a cascade failure simulation framework based on node load and capacity constraints in cascade scenarios, designing attack effects and constructing a dual-objective optimization function. Second, a bisection method is used to obtain the minimum number of nodes required for total failure, compressing the search space and improving the evolution speed. Prior knowledge heuristics and random generation of the initial population are used to improve population quality. Next, based on the GA algorithm framework, the crossover, mutation, local search, and selection strategies are improved to adapt to the current discretized space, achieving comprehensive exploration of the candidate key node set within the search space. Finally, based on the PSO framework, the update strategies for particle velocity and position are improved to adapt to the discretized space, achieving directed guided local optimization and improving solution quality. Finally, based on the Knee Point technique and non-dominated sorting technique, the Pareto front is selected to obtain the optimal set of non-dominated solutions, thereby obtaining the key node set with the minimum attack cost and the maximum attack effect.