126 results about "Beam search" patented technology

The invention discloses a method and a system for detecting a jailbreak attack of a large language model, which are applied to discovering potential security risks of the large language model, and the method comprises the following steps: constructing a jailbreak attack-malicious query data set of the large language model; an automatic antagonism suffix generation method fusing cluster search and a gradient descent algorithm is provided; the detection system of the prison break attack large language model is constructed by training, testing and evaluating the validity of the resistant suffixes based on a prison break attack data set, and comprises a malicious query data set display unit, a detection validity verification unit, a comparison and analysis unit before and after an attack and a model defense strategy unit. The method is helpful for discovering defects of a large language model alignment mechanism, so that a more powerful large language security model is designed and constructed.

The invention discloses a method for generating an image description text based on a large model, which relates to the technical field of image processing, and comprises the following steps: an image preprocessing step: dividing a target level through semantic segmentation and extracting key visual information by adopting hybrid denoising and self-adaptive normalization; a feature extraction step: fusing the multi-scale visual features and the semantic features, and generating a high-dimensional fusion feature vector through cross-modal alignment; a large model initialization and adaptation step: loading the pre-training model and performing incremental fine tuning, and dynamically adjusting the Prompt template; a text generation step: generating candidate texts through logic constraint and beam search; and a text optimization adjustment step of outputting a final description text based on multi-dimensional evaluation and user preference iterative correction. According to the method, the semantic matching degree, logic coherence and common sense accuracy of image description are improved, multi-element scenes are adapted through dynamic adaptation and iterative optimization, and high-quality text description support is provided for high-precision and diversified scenes.

A method for updating a question-answer mechanism includes obtaining, by a data system implementing the question-response mechanism, an input query, in response to the input query: generating a token sequence associated with the input query by applying a beam search algorithm on generated bound knowledge graphs, applying the input query to a large language model (LLM) of the question-response mechanism to obtain first outputs, applying the generated token sequence to a graph neural network (GNN) to obtain second outputs, applying fusion layers on the first outputs and the second outputs to generate a response associated with the input query and generated token sequence, and performing a remediation using the generated response, wherein the remediation comprises updating the LLM and the GNN based on the generated response.

The invention discloses a robot control system and method based on AI voice, and belongs to the technical field of robot intelligent control. The system comprises a local voice processing module, a dialogue management and context understanding module, an instruction mapping and generating module and a robot communication and control module. The local voice processing module is integrated with a 16 kHz sampling ASR engine, voice-to-text conversion is realized through a 6-layer LSTM network and beam search decoding, and the TTS engine adopts a WaveNet architecture to synthesize voice. A 7 billion parameter large language model is arranged in the dialogue management module, a 32-layer Transform decoder architecture is adopted, 10 rounds of dialogue history is maintained, and intention recognition and entity extraction are achieved through a multi-head self-attention mechanism. And the instruction mapping module performs trajectory planning by adopting quintic polynomial interpolation and quaternion spherical linear interpolation. According to the invention, full-link localization processing is realized, the speech recognition delay is 180 milliseconds, the position error is less than 0.5 mm, and the problems of high delay, high privacy risk and weak complex instruction understanding ability in the prior art are effectively solved.

To provide native hardware support for complex operations such as a beam search operation, an integrated circuit device can be implemented with multiple computational circuit blocks coupled in series to form a pipeline. Each computational circuit block includes an arithmetic logic unit (ALU) circuit having a first numeric input, a second numeric input, a primary result output, and a secondary output. The ALU circuit is programmable to perform an arithmetic operation on the first numeric input and the second numeric input to generate the primary result output, and provide one of the first numeric input or the second numeric input to the secondary output, which is fed back to the ALU circuit.

The invention discloses a medical text condition generation method under rule constraints, and belongs to the technical field of natural language processing and medical information. According to the method, a medical rule knowledge graph is constructed, graph neural network coding is adopted, and medical business rules are converted into constraint representation; designing a rule-perceived attention mechanism, and fusing medical rule constraints in a generation process; constructing a constraint-aware Transform decoder, and ensuring the generation normalization by adopting an improved beam search algorithm; and establishing a multi-level quality control system, and performing three-dimensional evaluation from local consistency, global consistency and rule conformity to generate a result. According to the method, the technical problem that existing medical text generation cannot meet complex rule constraints while guaranteeing fluency is solved, the rule coincidence rate reaches 95% or above, the quality score is improved by 18-25% compared with an existing method, and the method is excellent in performance in application of multiple medical departments.

A method for performing time constrained queries for event sequence visualization may include receiving an event sequence dataset including event sequences. Each event sequence includes at least two sequentially occurring events. The method further includes generating a forward index having event pattern sets that include pattern occurrence data for each event sequence positioned according to a fixed positional order. The method includes generating a time constraint search structure, such as a B−tree, B+tree, or table, that includes keys and pointer values. The method also includes inserting, into the search structure, a sequence identifier corresponding to each event sequence based on the fixed positional order and time interval data for the at least two sequentially occurring events of each event sequence. The method may also include searching the search structure to identify key values satisfying a received time interval constraint, and outputting the sequence identifiers corresponding to the identified key values.

The invention discloses a text-to-instruction system based on fuzzy pinyin matching and parameter robust analysis. The text-to-instruction system comprises an input preprocessing module, a text-to-instruction conversion module and a text-to-instruction conversion module, the fuzzy pinyin matching module is used for uniformly transferring the standardized text and a pre-stored instruction template into a silent pinyin sequence, and constructing a weighted editing distance matrix based on a preset pinyin character replacement cost; the parameter alignment search module is used for executing beam search on the weighted editing distance matrix and outputting a plurality of candidate matching paths with the minimum cost and a corresponding parameter fragment set; the parameter robust distribution module is used for performing legality and consistency checking on parameters in the candidate matching paths, completing parameter standardization and performing error correction in combination with a parameter candidate dictionary; the instruction output module is used for encoding the template identifier and the parameter key value pair obtained through robust distribution into a structured instruction and outputting the structured instruction; according to the method, text-to-instruction conversion can be completed in real time.

The invention discloses a monomolecular structure design and generation method based on a large language model, and relates to the crossing field of monomolecular electronics and artificial intelligence. In order to solve the problems of low single molecule design efficiency and insufficient data, SingMolT5 is obtained through field fine tuning on the basis of MolT5, and generation from a natural language to a molecular SMILES is realized. According to the method, data are collected from literatures in the single molecule field and an existing disclosed molecule library, a fine tuning data set containing 329 high-quality instructions is constructed, and the fine tuning data set comprises three basic data types including a molecular skeleton, an anchoring group and molecules. The model is initialized by a MolT5 check point, and a cross entropy loss function is used in the fine tuning process; reasoning, generating and using a beam search strategy; the follow-up evaluation indexes comprise BLEU, Levenshtein, MACCS, RDK, Morga and effectiveness. The method provides an innovative technical path for'low experience and data dependent type single molecule design 'in the field of single molecule electronics, and is particularly suitable for aided design of a single molecule device core function unit and a single molecule sensing probe molecule, namely, a target molecular structure can be quickly converted through a natural language, the design threshold of researchers is reduced, and the design efficiency is improved. And a traceable technical basis can be provided for screening and iteration of a monomolecular structure based on a high-quality data set and rigorous evaluation logic.

The invention discloses a channel routing optimization system based on a graph neural network, and the system comprises the following modules: a dynamic network graph construction module which is used for collecting network topology and link state data, and constructing a dynamic network graph model; the node embedding generation module is used for inputting the data into an improved MTGNN model and generating a high-quality node embedding vector by introducing a contrast learning mechanism; the candidate path discovery module adopts a greedy neural network algorithm based on beam search, carries out parallel extension and retains high-score candidate paths according to node embedding vector scores, and generates a candidate path set; the optimal path selection module is used for calculating the comprehensive utility value of each path, sorting the comprehensive utility values and outputting an optimal path; and the routing configuration and optimization module is used for generating and issuing a routing configuration instruction according to the optimal path. According to the invention, the accuracy and efficiency of routing decision and the utilization rate of network resources are improved, and intelligent path optimization in a complex dynamic network environment is realized.

The invention provides a context adaptive speech recognition method integrating entity replication and vector retrieval, and belongs to the technical field of artificial intelligence. The method comprises the following steps: converting input voice into acoustic feature representation; converting the external entity dictionary into vector representation; screening out candidate entities through an index technology, and determining to generate output or copy entities from a standard word list according to a matching degree calculated by an attention mechanism; jointly optimizing the model by comprehensively optimizing the objective function, wherein acoustic modeling, language modeling, entity copying accuracy and entity distinguishing capability are included; continuously updating the index during training; in reasoning, in combination with beam search and confidence threshold mechanisms, selection is made between generation of outputs from a standard word list and copying of entities. In the aspect of recognition performance, a dynamic copy-generation fusion mechanism and a contrast learning strategy are adopted in the method, the accuracy of named entity recognition is effectively improved, particularly, the method is excellent in performance when proper nouns such as personal names and place names are processed, and meanwhile the distinguishing capacity for entities similar in pronunciation is enhanced.

In order to reduce load due to a beam search of a base station and a communication device, one aspect provides a base station characterized by comprising: a reception unit for receiving, from another base station, a signal including information about a reception beam that the other base station uses when receiving a signal from a first communication device; and a control unit that determines a candidate for a beam to be used for communication between the other base station and the second communication device according to information about a transmission beam used when transmitting a signal to the first communication device and the information about the reception beam.

This invention provides a near-field millimeter-wave codebook generation method and related components. The near-field millimeter-wave codebook generation method is applied to a near-field hierarchical communication beamcodebook system. The method includes: obtaining the level number of the codebook at the current level in the near-field hierarchical communication beamcodebook system; determining the codebook level of the current level according to the relationship between the level number and a preset level, wherein the codebook level includes an upper-level codebook and a lower-level codebook; and obtaining a near-field millimeter-wave communication codebook based on the codebook level of the current level. The near-field millimeter-wave codebook generation method provided by this invention considers the characteristics of the spherical wavefront in the near-field region, avoids performance degradation, and achieves coverage of the entire Fresnel region. It designs a lower-level codebook covering polar coordinates and an upper-level codebook designed to reduce beam search overhead.

The invention discloses a front door adjustment-based harmful modal cause detection depolarization method, which comprises the following steps of: constructing a structural causal model, creating a back door path between a modal cause and a harmful label by using unobserved confounding factors, and eliminating harmful modal cause detection through front door adjustment; the front door adjustment then blocks the back door path by introducing reasoning as an intermediary between the memetic and the tag. The method comprises the following steps: decomposing the causal effect of a medal factor on a harmful tag into the influence of the medal factor on an intermediary, quantifying by adopting a multi-modal large-scale language model with different beam searches to calculate the probability of each reasoning sequence and the influence of the intermediary on the tag, and performing effective approximation through a normalized weighted geometric averaging method; finally, post-interpretation is synthesized from the inference sequence and the predictive tag using a large language model. According to the scheme, the most advanced performance is achieved while information interpretation is provided.

The application provides a knowledge processing method and device for power inspection defect detection, and the method comprises the following steps: obtaining an optimal path to connect a starting point entity and an ending point entity in the power equipment knowledge graph to form a Markov decision process according to the starting point entity and the ending point entity; training the Markov decision process to obtain a trained Markov decision process; an agent in the trained Markov decision process selects output edges in turn by using a random beam search strategy, and traverses to a new entity; when the agent searches for an optimal path by using the random beam search strategy, the path automatically forms a predicted explainable source. The application not only makes the power equipment knowledge graph knowledge completion explainable, but also improves the ability to understand complex environments and infer high-level causal relationships.

The application relates to a bundle search processing method and device, computer equipment and a storage medium. The method comprises the following steps: performing bundle search on a word table based on a current to-be-matched object and a current bundle width to obtain a current search result, wherein the current search result comprises a plurality of target words which are matched successfully with the current to-be-matched object and the number of which matches the current bundle width; performing width attenuation on the current bundle width according to a bundle width attenuation mode matched by the number of executed bundle searches to obtain a bundle width for next bundle search; taking each target word as a to-be-matched object for next bundle search, and iteratively performing bundle search until the bundle width obtained by attenuation is equal to a bundle width threshold value, and taking the search result at this time as a target to-be-matched object, and performing bundle search on the word table based on the bundle width threshold value to obtain a target search result. By adopting the method, the search efficiency can be improved while ensuring the search effect in an ultra-long sequence scene.

The invention provides a method and a system for safely and quickly testing a text graph model based on adversarial attacks, which belong to the field of adversarial attacks, and comprise the following steps: S1, when the text graph model deploys a cue word filtering mechanism, quickly generating an adversarial text through a beam search method by adopting a text modal attack mode, guiding the text graph model to generate an unsafe image which is semantically consistent with the sensitive content; s2, when a security review mechanism for the generated image is deployed in the text graph model, an image modal attack mode is adopted, an embedded confrontation method based on a potential space is designed, so that the generated image avoids a threshold value which is judged to be unsafe, and a momentum iteration fast gradient method is adopted to accelerate the attack speed; and S3, if the two defense mechanisms are deployed in the text graph model at the same time, the two attack modes are jointly used for attack. According to the method, the coverage range and the efficiency of the security test method for the text generated image model are improved.

The application relates to an outdoor geographical semantic retrieval enhanced navigation and search method and system, which comprises the following steps: receiving a natural language instruction, and extracting navigation target semantic elements and search target semantic elements; under the constraint of a structured index, performing beam search, and outputting the latitude and longitude and boundary information of the entity; decomposing the natural language instruction into a subtask sequence, and under the constraint of a predefined executable action and condition space, generating a structured behavior tree; calling a routing engine, generating a global topology path and a waypoint sequence based on the nearest reachable boundary point of the entity; based on the scheduling of the structured behavior tree, performing real-time navigation along the waypoint sequence; after reaching the vicinity of the target, searching for the target related to the search target semantic elements and performing clustering positioning under the constraint of a polygon boundary, obtaining a final target point and converging navigation. Compared with the prior art, the application has the advantages of high positioning accuracy and wide application scenarios.

This base station device is provided with: a first beam search instruction unit that instructs execution of first beam search processing in which a beam identification signal containing an identical beam identifier is simultaneously transmitted from a plurality of antennas to a wireless station, for each beam identifier, using a transmission beam linked to the relevant beam identifier; a second beam search instruction unit that instructs execution of second beam search processing in which, on the basis of a report signal that includes selected beam identifiers indicating a plurality of beam identifiers selected on the basis of the quality of reception by the wireless station of the beam identification signals for each beam identifier, obtained through the first beam search processing, a beam identification signal that includes the selected beam identifiers is individually transmitted, for each selected beam identifier, at least from among the plurality of the antennas; and a transmission / reception unit that, on the basis of results of the first beam search processing and results of the second beam search processing, determines a combination of an antenna and a transmission beam to be used for wireless communication with the wireless station.　

The invention provides an electric energy transaction intelligent contract management method and system based on a block chain, and relates to the technical field of intelligent management, and the method comprises the steps: carrying out the credit evaluation of a transaction queue through a Monte Carlo tree search algorithm, and selecting a preferable electricity selling node; receiving a power purchase demand and constructing a to-be-matched sequence; performing intelligent contract matching by adopting a beam search algorithm, and introducing a time decay penalty factor to process a continuous non-transaction scheme; and finally, generating a block chain smart contract. According to the invention, intelligent matching and credit management of electric energy transactions are realized, and the transaction efficiency and reliability are improved.

The application discloses a self-correction task planning method based on an implicit scene graph and speech feedback closed loop, and aims at the planning logic breaking problem caused by the lack of structured reasoning ability of a visual language-action model in a long-range task. The self-correction task planning method realizes the noise reduction and decoupling of perception by constructing an implicit semantic scene graph, preforms a trajectory in a thinking space by using multi-agent collaboration and heuristic beam search, and introduces a speech reflection mechanism to convert logical fallacies into natural language constraints, so that the strategy self-correction is realized without parameter updating. Meanwhile, the application also constructs a double-layer evaluation system integrating semantic manifold understanding and time sequence logic verification, and solves the problem that existing indexes are difficult to quantify the logical rationality of the planning itself.

Embodiments of the present disclosure relate to beam failure detection. According to embodiments of the present disclosure, after a serving cell is activated from a deactivated state, a terminal device monitors a beam failure on a serving beam of the serving cell. If a beam failure instance indication is received within a short time after the activation, the terminal device determines that a beam failure is detected on the serving beam. In this way, beam failure recovery can be triggered faster to perform candidate beam search.

The application belongs to the technical field of 6G communication, and particularly relates to a hierarchical codebook design method based on a cascaded channel. The application defines a cascaded channel direction by combining an incident direction and a reflection direction, and one cascaded channel direction may correspond to multiple groups of incident-reflection directions; the cascaded channel direction defined in this way realizes a hierarchical codebook; the hierarchical codebook can be used for beam search in a single-user static scene, and can also be used in an analog beamforming scene of an active array to solve hierarchical codebook design based on a constant modulus constraint; unlike a traditional hierarchical codebook design which needs to know the array size in advance, the hierarchical codebook proposed in the application can be applied to any size of array. The application has more universality and application flexibility, and effectively solves the problem of too long beam search time of an intelligent reflecting surface. The codebook can be used on any size of array, and when the number of arrays increases, the codebook can be directly used without changing the codebook, and the larger the array, the better the performance of the codebook.

A method for performing time constrained queries for event sequence visualization may include receiving an event sequence dataset including event sequences. Each event sequence includes at least two sequentially occurring events. The method further includes generating a forward index having event pattern sets that include pattern occurrence data for each event sequence positioned according to a fixed positional order. The method includes generating a time constraint search structure, such as a B−tree, B+tree, or table, that includes keys and pointer values. The method also includes inserting, into the search structure, a sequence identifier corresponding to each event sequence based on the fixed positional order and time interval data for the at least two sequentially occurring events of each event sequence. The method may also include searching the search structure to identify key values satisfying a received time interval constraint, and outputting the sequence identifiers corresponding to the identified key values.

This application discloses a method, apparatus, storage medium, and device for solving plane geometry problems. The method includes: first, acquiring the target plane geometry problem to be solved and its corresponding target image; then, using a preset first deep learning model to predict entity relationships in the problem text to obtain a first neural symbol representation corresponding to the problem text; and using a preset second deep learning model to transform the target image to obtain its corresponding second neural symbol representation; finally, inputting the first and second neural symbol representations into a symbolic reasoning system, and using beam search to perform reasoning to obtain the solution. It is evident that because this application first uses a deep learning model to convert the problem text and the target image of the target plane geometry problem into neural symbol representations, and then uses beam search to perform reasoning through a symbolic reasoning system, the accuracy of the solution to the plane geometry problem can be improved.

The invention discloses a data processing method and device, a storage medium and a computer system. The method comprises an initial class grouping stage, a class grouping optimization stage and a class arrangement stage. In the initial class grouping stage, initial class grouping is carried out based on a first data set and a second data set, in the class grouping optimization stage, on the basis of an initial class grouping result and under the constraint of a synchronous class condition, adaptive cluster search is used for class grouping optimization to obtain a class grouping optimization result, and in the class arrangement stage, class arrangement and class grouping optimization is carried out based on the class grouping optimization result. According to the invention, the performance and efficiency of course arrangement processing can be effectively improved.

The application provides a video intelligent synthesis method for multi-modal content conversion, comprising: acquiring video demand data, performing semantic analysis on the video demand data, generating a video intelligent synthesis data set, performing multi-path material retrieval based on the video intelligent synthesis data set, pruning in combination with a preset dynamic pruning strategy, generating a video synthesis material candidate set and a key frame demand set, generating an intelligent synthesis video segment based on the key frame demand set, performing quality screening on the intelligent synthesis video segment, generating a candidate video synthesis segment set, performing video intelligent synthesis based on beam search guidance according to the candidate video synthesis segment set and the video synthesis material candidate set, generating a candidate intelligent synthesis video set, performing quality evaluation on the candidate intelligent synthesis video set based on a preset video quality evaluation mechanism, and performing secondary synthesis according to the evaluation result to generate an optimal synthesis video, thereby improving the efficiency, accuracy and reliability of automatic video production.

The invention discloses a three-dimensional indoor positioning method based on visual sense and 5G sensing fusion, and relates to the technical field of indoor positioning and space perception. The method comprises the following steps: modeling an indoor reflecting surface to obtain a virtual anchor point; a user azimuth vector detected, identified and tracked by a camera target is projected to a scene reflection surface to construct a candidate reflection path and identify a direct incidence path, and the candidate reflection path and the direct incidence path are mapped to a beam codebook of a base station, so that a beam search range is narrowed, and a higher angle resolution is obtained; the user terminal establishes a communication link through the reference signal, and the base station acquires channel state information; in the channel parameter processing stage, angle and time delay estimation is carried out on channel parameters; an estimation result is matched with a direct or virtual anchor point path, and pseudo paths which do not conform to scene constraints are eliminated, so that a direct path and an effective reflection path are distinguished; the position of the user is jointly solved by using multi-angle constraints formed by the direct path and the reflection path, and continuous estimation of the three-dimensional position of the user is realized by combining a filter.

Systems and methods for control signaling for beam searching latency reduction are disclosed herein. A g Node B (gNB) may determine that a first Synchronization Signal Block (SSB) and a second SSB are to be spatially correlated and may select a first transmit (Tx) beam to transmit the first SSB and a second Tx beam to transmit the second SSB accordingly. The gNB may also transmit a correlation message including spatial correlation information to help a UE determine the spatial correlation. The UE may measure the first SSB on a first subset of a plurality of receive (Rx) beams and measure the second SSB on a second subset of the plurality of Rx beams, and select an Rx beam for one or both. In some embodiments, channel state information reference signals (CSI-RS) that are quasi co-located (QCLed) with a given SSB may be measured in place of the SSB.

This invention discloses a method, system, device, medium, and program product for generating interpretability across the entire AI decision-making process. The method includes the following steps: constructing a domain ontology knowledge graph for the AI ​​model, including a data layer, a concept layer, and a decision layer; mapping the input features of the AI ​​model to data layer nodes, and mapping abstract concepts in the business logic of the AI ​​model to concept layer nodes; calculating the contribution vector of the input features to the output result, selecting data layer nodes corresponding to input features with contribution values ​​exceeding a preset threshold as the starting node set, and selecting the decision layer nodes corresponding to the output result as the target nodes; constructing an active subgraph based on the starting node set and the target node's centered neighbor nodes; performing critical path mining in the active subgraph based on a hybrid energy scoring mechanism, using a beam search algorithm to find the optimal explanation path, and generating natural language explanation text. This invention significantly improves the understandability and credibility of the explanation.