325 results about "Beam search" patented technology

Certain aspects of the present disclosure provide techniques and apparatus for efficiently generating a response to a query input in a generative artificial intelligence model. An example method generally includes generating, based on an input prompt and using a first machine learning model, a set of tokens including one or more subsets of tokens. Each respective subset of the one or more subsets corresponds to a respective portion of a response to the input prompt and includes a fixed number of tokens corresponding to a beam width for a beam search through the set of tokens. The set of tokens is output to a second machine learning model for verification, and information identifying a selected sequence of tokens from the generated set of tokens is received from the second machine learning model. The selected sequence of tokens is output as the response to the input prompt.

The invention provides a personalized teaching method and system based on generative artificial intelligence. The method comprises the following steps: constructing a knowledge graph and a teaching strategy template library of a target course; constructing a feature enhancement matrix; calculating node activation intensity, generating a plurality of candidate learning paths through a beam search algorithm, and screening an optimal learning path in combination with a preset teaching strategy template; using the generative model to generate standardized teaching content adaptive to the teaching strategy; dynamically optimizing a teaching scheme through a neuroplasticity model; and carrying out teaching adjustment based on the optimized teaching scheme. Standardized teaching content is generated by utilizing the generative model, the teaching content quality is improved, the teaching content is closely matched with a learning path and a teaching strategy of a learner, meanwhile, a teaching scheme is optimized through the neuroplasticity model, teaching parameters and cognitive resource allocation are adjusted in time according to real-time feedback and cognitive states of the learner, and the teaching efficiency is improved. Limited cognitive resources are reasonably distributed to key parts of teaching contents, and the teaching effect is improved.

The invention discloses a knowledge graph multi-hop reasoning-based complex manufacturing process intelligent planning decision method, which comprises the following steps of: 1, constructing a manufacturing process knowledge graph to obtain a triple set, generating a query-path pair sequence, and mapping entities and relationships in the query-path pair sequence into low-dimensional vectors through an embedded layer; 2, performing expansion causal convolution calculation on the query-path pair sequence through a shallow feed-forward convolution network module, extracting key process features and potential correlation between processes, and generating a feature vector fusing the potential process correlation; 3, performing rotation position coding and causal mask processing on the feature vector output in the step 2 by adopting a parallel multi-query attention module, and realizing cross-process dependency relationship modeling through a multi-head attention mechanism; 4, generating a global optimal process scheme through beam search; and on the basis of a rule-guided dynamic aggregation iteration strategy, mixing a process path generated by the model with an original training set, and updating parameters.

The invention discloses a non-point source pollution emergency bypass decision control method, particularly relates to the technical field of automation and process control, and aims to solve the problems that existing scheduling cannot align multi-source measurement and evaluate credibility in minutes under a unified time base, and is lack of risk measurement integrated with a tide level phase and a moisture regain path and same-layer gating. The bypass rhythm is easy to misjudge, and the oscillation is difficult to audit. A minute-level risk density is constructed through multi-source alignment and credibility grading under a unified time reference, injected tide level phases and moisture regaining paths, bypass rhythms are constrained by same-layer gating and tide windows, and beam search, boundary arbitration and fixed window closed-loop correction are matched, so that the risk density is improved. Therefore, under the superposition of rainstorm and tide, decision making and execution are stably limited in a non-backflow red line, bearing and equipment limit, misjudgment and oscillation are remarkably reduced, the problems of unavailability and difficult auditing caused by multi-source asynchronization and moisture regain coupling are solved, intelligent decision support is provided for non-point source pollution treatment, and it is ensured that the treatment technology is accurate and efficient.

The invention belongs to a decoding method, and provides a multi-modal brain-computer interface decoding method and a related device for solving the technical problems that an existing non-intrusive brain language decoding method is insufficient in adaptability in global context and weak in generalization performance in a cross-subject scene, and multi-modal neural feature collaborative enhancement is difficult to achieve. And determining the called execution module. The execution module comprises a feature extraction module, a cross-subject standardization module, a multi-mode semantic fusion module, a language recognition module and a semantic consistency module. By obtaining the unified semantic representation and combining the beam search algorithm, the fairness and universality in different language groups are remarkably improved, multiple modes can be supported, and the brain signal decoding precision and robustness are improved. In addition, cross-subject semantic representation generalization can be realized, and individual specificity is effectively reduced.

The invention belongs to the technical field of dialogue generation, and relates to a digital intelligent carrier pigeon WeChat intelligent dialogue generation method based on deep learning. Dialogue data are obtained, and semantic anomaly detection is carried out through a regular expression and a BERT model in combination with a public corpus; organizing dialogue data in a mode of constructing a tree structure and introducing a three-dimensional timestamp; for emotion information of emoticons, text and visual features of expressions are combined through a hybrid encoder, a dynamic attention mechanism is adopted, weighting and memory length adjustment are carried out according to importance of historical statements, after user portrait features are introduced, user features and dialogue codes are fused, and therefore, the emotion information of the expressions is obtained. In the generation stage, through a Transform-based model and a Beam Search algorithm, a diversity penalty term is introduced in the generation process, and generation parameters are dynamically adjusted according to a user portrait, so that the diversity and security of generated dialogues are ensured, the precision and coherence of a dialogue generation model are improved, and the technical problems of context coherence and personalized generation are effectively solved.

Methods, systems, and devices for fast beam selection via user equipment (UE) receive (Rx) beam selection reporting are described. A UE may transmit a beam selection report conveying a selected Rx beam, position, and orientation to a network entity. The network entity may build a database with the conveyed information. The network entity may “tag” each position for which a UE has transmitted a beam selection report and, if another UE gets close to a “tagged” position, the network entity may indicate the associated position, orientation, selected Rx beam, and a corresponding transmit (Tx) beam to that UE. The UE may use the information to select Tx and Rx beam search windows and window centers. Compared to other methods, the process described herein may result in a smaller search window, decreased complexity of the beam selection process, decreased power consumption, decreased latency, and more accurate tracking of UE dynamics.

The invention relates to a lyric generation method and system based on a large language model and thinking chain reinforcement learning. The method comprises the steps of designing a thinking chain path for lyric creation example data, and performing parameter adjustment on a large language model; performing multi-round reasoning on each paragraph through the adjusted large language model to obtain each thinking path; scoring each thinking path based on a process reward model; and selecting a thinking path with the highest score as a positive sample by using a Monte Carlo tree search algorithm, selecting other paths as negative samples by using a beam search algorithm, and training a lyric generation model. Model fine tuning is carried out by designing a thinking chain path, and the ability of the model to generate a lyric thinking path can be enhanced; each paragraph is subjected to multi-round reasoning through a large language model, various different thinking paths are generated, the individuation, logicality and emotional expression of lyric generation can be enhanced, and therefore a music creation text closer to actual requirements is provided.

The invention discloses a Chinese text grammar error correction method combining syntactic information and a pre-training language model, and the Chinese grammar error correction capability can be effectively improved by fusing the syntactic information into the pre-training language model Chine BART. The method comprises the following steps: importing a data set and converting the data set into a target format; syntactic information is generated by utilizing HanLP, part-of-speech tags and dependency relationships are converted into vectors, and the vectors are fused with default word embedding; in the encoding stage, a syntax-enhanced multi-head self-attention mechanism and a residual network are used; in the decoding stage, a self-attention mechanism and an encoder-decoder attention mechanism are combined with syntactic information for generation, and a decoder is gradually optimized in combination with a residual network; a plurality of candidate sentences are generated by using Beam Search, and the optimal candidate sentence is selected according to syntactic consistency and semantic rationality. According to the method, the syntactic information and the pre-training language model are combined, so that the model learns semantic information of sentences and pays attention to the syntactic structure, and the Chinese grammar error correction capability of the model can be improved.

The invention provides a method and device for determining a code patch, equipment and a storage medium, and the method comprises the steps: generating a test case for a set fragment in a to-be-tested file by executing a first large language model; running the test case for the to-be-tested file; if the error description of the test failure is obtained, inputting the error description and the problem code block into a second large language model to obtain at least one code patch for the problem code block; wherein the at least one code patch is obtained by the second large language model according to at least one code repair strategy through a cluster search mode; the at least one code patch is in one-to-one correspondence with the at least one code repair strategy; a final patch is determined from the at least one code patch and the set segment is replaced. According to the method, the test case is automatically generated through the first large language model, the patch is automatically generated for the problem code block through the second large language model, and the automation degree and efficiency of the test are improved.

The invention discloses an intelligent task planning method and device based on environment perception, equipment and a medium, and the method comprises the steps: fusing abstract environment information into a large language model, and carrying out the fine adjustment of the large language model, so as to enhance the environment perception capability of the large language model; inputting a task target and an environment state table into the fine-tuned large language model, and generating a multi-step action sequence with the highest probability through a step-by-step beam search strategy; according to the generated action sequence, the environment data state table is updated by simulating interaction between the robot agent and other objects, and therefore interaction tracks corresponding to the steps are generated. According to the method, the model is adjusted into an environment perception large language model fine tuning structure, so that higher accuracy and practicability are brought to intelligent task planning; the large language model can explore a wider search space by utilizing a step-by-step beam search strategy, so that more comprehensive search is beneficial to generating a more reasonable and effective plan.

The invention relates to a multi-modal automatic abstracting method based on deep learning, and belongs to the technical field of computer natural language processing. The method comprises the following steps: 1, extracting text features by utilizing BERT, and extracting image features by utilizing GoogleNet; 2, the extracted image-text features are connected in series to form a series vector; 3, fusing the text features and the image features; 4, inputting the fused text features into a decoder of the text abstract, calculating by adopting the cross attention and self-attention with masks of the decoder to obtain autoregression results of word vectors of the text abstract, and selecting the autoregression result with the maximum probability in an autoregression sequence by utilizing beam search to output the text abstract; 5, inputting the fused image features into a Softmax layer, and performing image abstract output on a sample image with the maximum probability in the samples; compared with the prior art, the method has the advantages that the text abstract is generated and the picture most related to the abstract is selected by adopting deep learning, so that the text abstract and the related picture can be accurately generated, and the practicability is very high.

The invention relates to the technical field of information retrieval, and provides a dialogue type generation retrieval method and system based on large model document index perception. The dialogue type generation retrieval method based on large model document index perception comprises the steps that according to user query input, a trained large language model is adopted, and an information sorting list related to the user query input is output; on the basis of a corpus, a cross attention layer is adopted, key information of a context is extracted, a proposition is obtained, and a document identifier is obtained; in the first-stage training, generating loss is introduced to generate information related to the current query, and a document identifier is generated; in the second-stage training, a ranking list of document identifiers retrieved by comprehensive loss optimization is introduced; and outputting a paragraph ranking list corresponding to the document identifier through a beam search decoding strategy. Through innovative document identifier design and training strategies, more effective context understanding and denoising are realized, and the dialogue type retrieval accuracy and user experience are improved.

The invention discloses an open domain Text-to-SQL (Structured Query Language) multi-hop table retrieval method and system, a storage medium and equipment, and belongs to the technical field of Text-to-SQL. The method aims at solving the problems that according to an existing table retrieval method, an unrelated table is retrieved due to the fact that an entity is similar to an entity in a plurality of unrelated tables, and the retrieved table is farther away from a related field due to the fact that the entity is possibly not matched with the related field. According to the method, multi-hop retrieval is adopted, an original user question retrieval table is adopted in the first hop, and a tabular question retrieval table rewritten by LLM is adopted in the later hop steps; a plurality of search results are kept in each hop by adopting beam search, the probability of correlation between a table and a question is calculated after each hop of search stage is ended, and a plurality of groups of most possible search results are selected and input to a rewriting stage together with an original user question; all the tables are sorted, and the tables are selected according to the tables for retrieval.

The invention discloses a high-dimensional vector approximate nearest neighbor retrieval method for persistent storage. The method comprises an index construction stage and a search stage. In the index construction stage, a clustering algorithm is adopted to carry out space division on a data set, a center replacement strategy is introduced to select a representative vector for establishing an entry point diagram, and meanwhile, a second-layer neighbor graph is constructed on the original data set; in the search stage, the query vector is quickly positioned to the closest entry point by using the result of the entry point diagram, so that the search path is shortened. According to the method, the optimal product quantization coding strategy is adopted, so that the SSD data loading overhead is reduced; a multi-beam search strategy is adopted to effectively utilize the page-by-page reading characteristic of a file system, so that the time overhead is reduced, and the data loading amount is increased; and an iterative search strategy is adopted to collect query vector search features, the search process is self-adapted, and the search efficiency is improved. Aiming at mixed architecture challenges of large-scale data, the invention provides an efficient search solution for optimizing approximate nearest neighbor search.

A technique for solving combinatorial problems, such as vehicle routing for multiple vehicles integrates evolutionary algorithms and reinforcement learning. A genetic algorithm maintains a set of solutions for the problem and improves the solutions using mutation (modify a solution) and crossover (combine two solutions). The best solution is selected from the improved set of solutions. A system that integrates evolutionary algorithms, such as a genetic algorithm, and reinforcement learning comprises two components. A first component is a beam search technique for generating solutions using a reinforcement learning model. A second component augments a genetic algorithm using learning-based solutions that are generated by the reinforcement learning model. The learning-based solutions improve the diversity of the set which, in turn, improves the quality of the solutions computed by the genetic algorithm.

The invention discloses a multi-language speech recognition method. A multi-language speech recognition model comprises an encoder, a language classification model, a CTC module and a decoder. The method comprises the following steps: acquiring acoustic features of an audio to be identified, and inputting the acoustic features to the encoder to obtain an encoder output result containing acoustic feature local information and acoustic feature global information; a language classification model carries out preset processing on the encoder output result to obtain a language classification result; the CTC module is combined with a weighted finite state conversion machine, and an encoder output result is mapped to a streaming nbest identification result through a prefix beam search algorithm; and the decoder re-scores the output result of the encoder according to the streaming nbest identification result. The multi-language speech recognition model only shares an encoder at the bottom layer, and the language classification model, the CTC module and the decoder at the top layer are independent. The problem that homophones of different languages are confused in speech recognition is reduced through shared bottom layer parameters and a re-scoring process, and the language classification accuracy is improved.

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 document retrieval query rewriting method and system based on a large language model, and the method comprises the steps: collecting a query rewriting data set, carrying out the correlation filtering and retrieval increment, and retaining a data sample highly related to a query rewriting task; an auxiliary task data set highly related to the query rewriting task is collected, and a multi-task SFT data set is constructed; constructing a self-supervised fine tuning model based on GPT-2, taking query as input, taking query rewriting as output, and training the self-supervised fine tuning model by using a multi-task SFT data set; a self-supervised fine tuning model is regarded as an intelligent agent, and target alignment is carried out based on reinforcement learning, so that the self-supervised fine tuning model better conforms to the user intention when generating query rewriting; a plurality of candidate rewrites are generated for each query using a beam search algorithm, a self-supervised trim model of target alignment is entered, and a set of related documents is retrieved. According to the method, the retrieval relevance is improved, and the user satisfaction is improved.

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.

The invention discloses a limited autoregressive decoding method and system based on incremental analysis and a medium, and the method comprises the steps: for a given Text-to-SQL task, formatting and coding an input natural language query problem and a corresponding database mode; generating an SQL query by adopting a large language model; the score is adjusted according to the log-softmax score predicted by the model and the currently assumed token, and in the step of generating each token of the SQL, the candidate tokens are subjected to incremental analysis to check whether the candidate tokens conform to grammar and semantic rules of the SQL; performing validity check on the generated SQL query through different preset check models; and endowing negative infinity scores to the tokens which do not pass the check, excluding the negative infinity scores in beam search, and outputting, evaluating and optimizing the effective SQL query sequence which passes the check. The execution precision of the language model can be effectively improved.

The invention discloses a network attack path deduction method and device based on a large language model, equipment and a medium, and relates to the technical field of network security, and the method comprises the steps: generating each initial deduction result corresponding to a current abnormal network activity through employing a target large language model and a beam search algorithm; verifying each initial network attack deduction path by using a target knowledge graph to obtain a target confidence degree corresponding to each initial network attack deduction path; and determining a target network attack path with the highest target confidence from the initial network attack deduction paths, and determining an initial deduction result corresponding to the target network attack path as a target deduction result so as to display the target deduction result by using a time sequence animation. The initial network attack deduction path is generated by using the large language model, and the target network attack deduction path is determined based on the knowledge graph, so that the problem that complex and changeable attack techniques cannot be adapted due to rule engine stiffness is solved.

The invention provides a text error correction method and device, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence, and the method comprises the steps: inputting a target text into a plurality of error correction components to obtain a to-be-corrected result set, and aiming at any to-be-corrected word of the to-be-corrected result set, performing error correction on the to-be-corrected word; modifying the target text by using all the error correction recommendations of any word to be corrected to generate a candidate text set, so as to screen out a reserved text from the candidate text set; and continuously carrying out error correction on another word to be subjected to error correction in each reserved text until all words to be subjected to error correction in the result set to be subjected to error correction are traversed, and determining a recommended text from the finally obtained reserved texts. According to the text error correction method based on beam search multi-model integration provided by the invention, the error correction capability can be improved to the greatest extent by integrating the advantages of a plurality of error correction components.

The invention belongs to the field of unmanned aerial vehicle mobile search, and particularly relates to an optimal search method under an unmanned aerial vehicle SAR (synthetic aperture radar) grid scene, which comprises the following steps: constructing an SAR unmanned aerial vehicle search system model; constructing a search target function of maximizing the residual resources and the number of searched people according to the recognized people number constraint, the search region area constraint and the unmanned aerial vehicle resource constraint; solving the residual resource minimization and search number maximization search objective function by using a deep reinforcement learning method to obtain an optimal search path; the system optimizes the deflection angle and the moving distance of the unmanned aerial vehicle in the search model in real time according to the optimal search path scheme; and detecting the finally generated path area by using a target detection algorithm to identify trapped persons, and optimizing the search efficiency of the search model. According to the method, the optimal search path of the unmanned aerial vehicle under the condition of minimum resource consumption in the search scene can be reflected, meanwhile, human eye recognition is replaced by the recognition algorithm, manpower is liberated, and the search efficiency is improved.

The invention discloses a wafer scheduling method and semiconductor process equipment. The method comprises the following steps: determining a first candidate action sequence according to an objective function value of each action sequence obtained by cluster search; determining a second candidate action sequence which reaches the process chamber fastest in all action sequences corresponding to the same new wafer; in response to the fact that the number of actions in the candidate action sequence reaches an action number threshold value, determining a target action sequence according to the target function value; in response to the fact that a first action of the target action sequence meets a preposition constraint, the first action of the target action sequence is executed in a simulation mode, and the preposition constraint is that new wafers executing different tasks are taken out from the loading and unloading positions according to the sequence of reaching the process chamber; and in response to completion of simulation scheduling of all wafers, outputting an action scheduling sequence including sequential simulation execution. According to the invention, the wafer can be processed as soon as possible. The development period is short, the maintenance cost is low, and the result is more accurate and reliable.

A computer-implemented method for optimizing a video encoder. The method includes the step of defining a plurality of CUs of the video encoder that correspond to a plurality of stages. The plurality of stages includes at least a first stage, and a second stage immediately after the first stage. The method further includes the steps of providing a decision space of encoding parameters of the video encoder, and defining, for the second stage, a subspace being a subset of the decision space. The subspace contains b1 optimal decision paths from the first stage to the second stage, wherein b1 is defined as a beam size for the first stage. The b1 optimal decision paths are the b1 decision paths that have lowest accumulated cost among all decision paths from the first stage to the second stage.

The invention discloses an optimization method and device for a natural language processing model, and relates to the technical field of information security, and the method comprises the steps: reducing the text confusion degree through employing an iterative beam search mode, and guaranteeing the text quality at the same time, i.e., carrying out the quantitative evaluation and screening of readability and usefulness in a processing process, and improving the processing efficiency. And performing model fine tuning by using the optimized and transformed mixed data to reduce the influence of harmful texts in the third-party data on the model security alignment, thereby improving the security performance and the alignment degree of the large model, and solving the problem that the security alignment of the large language model cannot be effectively coped with in the related technology, and the security alignment efficiency is improved. The technical problem that possible harmful information damage in the third-party data is difficult to fundamentally reduce is solved, and the technical effects that the safety performance and the alignment degree of the large language model are improved, it is ensured that the large language model can give safe and reliable response when facing the data safety problem, and harmful information is prevented from being provided are achieved.

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 an intelligent fast beam training method for low-altitude millimeter wave communication, and belongs to the technical field of wireless communication. The method comprises the following steps: establishing a channel and transmission model of the low-altitude millimeter wave cellular-free large-scale MIMO system; in the channel estimation stage, a pilot signal is used for estimating a channel and preliminarily positioning the position of a user; in the beam training stage, firstly, customized three-dimensional beam codebook design is carried out based on obtained positioning information; and then, an enhanced multi-agent depth Q network algorithm is adopted to efficiently carry out beam search in the designed customized codebook, so that accurate three-dimensional beam alignment is realized. According to the method, the customized codebook and the intelligent learning algorithm are combined, the beam forming process is effectively optimized, the complexity and overhead of beam training are remarkably reduced, the convergence speed is increased, the system performance and reliability of downlink communication are improved, and the method is suitable for a high-maneuverability low-altitude unmanned aerial vehicle communication scene.

The invention provides an optimization method for beam search for non-completely independent break points. The optimization method comprises the following steps: S1, judging completely independent break points and non-completely independent break points; s2, if a non-completely independent segmentation point exists, namely, an unconsumed ancestor node exists, the node needs to be reserved for the remaining sub-graphs, network memory peak information is obtained, at the moment, a scheduling scheme of nodes in front of the non-completely independent segmentation point is determined, the beam width is 0, and only a limited number of nodes without completely consuming the memory enter the remaining sub-graphs for recalculation; according to the non-completely-independent segmentation points, the whole network graph is segmented into independent sub-graphs, but the non-completely-independent segmentation points need to be reserved for the remaining sub-graphs, that is, the last independent sub-graph has N schemes, the optimal scheme is selected, and serial numbers from the current moment 1 to the moment t are recorded; independently analyzing serial numbers at the (t + 1) th moment and after the (t + 1) th moment, reserving serial numbers of completely independent segmentation points, and respectively performing beam search on the remaining sub-graphs to sort; and S3, finally, sorting and splicing the sub-graphs to form sorting information of the whole network graph.