612 results about "Round complexity" patented technology

The invention relates to the technical field of artificial intelligence and natural language processing, in particular to a key value cache compression and sparse attention calculation method and system for large language model reasoning, and the method comprises the steps: an offline calibration stage; the online reasoning stage comprises the following steps: a pre-filling step; an autoregression generation step: for each newly generated lexical element, projecting a current query vector Q and a key vector K in a key cache to a low-dimensional space to obtain Q'and K '; calculating an approximate attention score based on Q'and K ', and selecting an index I of the first k most relevant lexical elements which are ranked from high to low; and calculating an accurate attention score based on Q and K [I], and calculating with the value vector V [I] to obtain the output of the current lexical element. According to the scheme, the memory and calculation bottleneck of large model reasoning in a scene of long text sequence input are solved, and the method has the advantages of reducing video memory occupation and calculation complexity at the same time.

The invention discloses a process automation execution method based on a large language model, and belongs to the technical field of artificial intelligence and process automation. User intention is analyzed through multi-modal input, and a structured task definition is constructed; the semantic reasoning layer is used for performing task layering, complexity evaluation and sorting optimization; the task execution layer completes subtask scheduling and execution; and the feedback and optimization layer performs performance evaluation and model updating based on execution data to realize closed loop and continuous optimization of the process, so that the technical problems of dynamically analyzing unstructured instructions, automatically optimizing a complex task dependency relationship and adapting to business changes in real time by a process automation tool are solved; according to the method, end-to-end conversion from an unstructured instruction to a structured task is realized, a subtask execution path is dynamically optimized, cross-platform tool calling is supported, the existing system integration cost of an enterprise is reduced, visual display task decomposition logic and prediction and execution time consumption comparison are provided, and the system credibility is enhanced.

The invention discloses a software development task intelligent allocation method and system based on a multi-dimensional capability portrait, and the method comprises the following steps: 1, obtaining multi-source development data generated by a developer in a development process and demand description data of a to-be-allocated software development task, and carrying out the preprocessing, and forming a structured data set; according to the method, a multi-dimensional ability portrait covering technical ability, project experience, collaboration attributes and performance is constructed, a privacy-protected distributed learning mechanism is adopted for dynamic updating, dominant and implicit requirements of tasks are analyzed in combination with natural language processing, complexity and dependency are calculated, and the performance of the performance is improved. A dynamic task feature vector corresponding to a capability feature vector dimension is constructed, and meanwhile, a self-adaptive adjustment mechanism based on real-time data monitoring and online learning is designed to form data closed-loop feedback, so that the problems of one-sided capability evaluation, staticizing task demand analysis and lack of the self-adaptive adjustment mechanism are comprehensively solved; and accurate and intelligent distribution of software development tasks is realized.

The invention relates to a standardized project management and intelligent professional scheduling system, and belongs to the technical field of project management and human resource intelligent scheduling. The project standardization module is disassembled into standardization task steps through a business process association rule mining algorithm, and project complexity and resource tensity are adapted by using a task attribute dynamic weight algorithm; the archive matching module constructs professional archives, extracts feature vectors through a multi-criterion decision-weighted bipartite graph matching algorithm, and generates optimal matching pairs in combination with adaptive weight adjustment and an integer linear programming model; the dynamic scheduling module plans a task execution scheme according to a resource constraint scheduling mechanism, and realizes efficient scheduling in combination with a skill supply and demand prediction algorithm and calendar integration; and the quality optimization module adopts a deliverable anomaly detection algorithm to monitor compliance, feeds back an iterative matching and scheduling strategy through a time sequence prediction optimization algorithm, and perfects a skill map based on a map increment updating algorithm. The system realizes a project full-process closed loop.

The invention discloses an automatic business travel itinerary optimization method, and relates to the technical field of intelligent itinerary planning, and the method comprises the steps: integrating the multi-source heterogeneous data of enterprise policies, personal preferences and real-time traffic through a federated learning framework, and achieving the cross-domain knowledge sharing; the method comprises the following steps: constructing a staged optimization engine by adopting an attention mechanism to dynamically balance cost, time, comfort and sustainability targets: in the first stage, modularly disassembling a travel through sparse constraint linear programming, and quickly generating a Pareto frontier candidate set; in the secondary stage, on the basis of a multi-agent reinforcement learning framework, complex interaction is simulated through a Markov decision process, and strategy iteration is driven through a special reward function for quantifying a comfort index; in order to cope with real-time disturbance, event-driven edge computing nodes are deployed, flight delay and traffic jam emergencies are responded in real time, an incremental topology updating algorithm is triggered, and only affected sub-modules are reconstructed to reduce computing complexity. According to the invention, the bottleneck of dynamic adjustment efficiency and multi-target balance capability is solved.

The invention relates to the field of nonlinear system control, and provides a preset time reinforcement learning method and system of a continuous nonlinear system, and an electronic device, and the method comprises the steps: constructing a zero-sum game framework based on a kinetic model and a performance index of the nonlinear system; determining a value function and a Hamiltonian function based on a zero-sum game framework; applying a preset neural network model to carry out approximation on the value function, and determining an approximation error; based on the Hamiltonian function and the approximation error, an approximate optimal control strategy and a worst interference strategy are determined; constructing a Lyapunov function based on the value function and the weight error of the value function; and based on a Lyapunov function, an approximate optimal control strategy and a worst interference strategy, a reinforcement learning result is verified. The method and the device are used for overcoming the defects that convergence time cannot be dynamically adjusted, parameter complexity is high and robustness is insufficient in the prior art, and the scheme of the invention can meet dual requirements of a continuous nonlinear system on dynamic convergence and anti-interference performance.

The invention belongs to the technical field of communication, and particularly relates to a multiplication active code decoding method based on an SO-ORBGRAND decoder. The SO-ORBGRAND decoder outputs soft information by calculating a correct posterior probability of a conjecture code word sequence, and high-performance decoding is realized through soft information exchange and optimization processing between the row decoder and the column decoder. In the iteration process, a method for judging the validity of the code word through the check matrix is designed by utilizing the polarization characteristic of the multiplication active code, so that the iteration is supported to jump out in advance to reduce the complexity. The invention provides a multiplication active code decoding algorithm based on an SO-ORBGRAND decoder, and compared with various non-system multiplication active code decoding algorithms and the performance of traditional long-code polarization codes, the method has remarkable advantages in error correction performance; a method for judging the validity of the code word through the check matrix by utilizing the polarization characteristic design of the multiplication positive code is provided, the complexity is reduced while the judgment success rate is improved, and the implementation overhead is reduced.

The invention belongs to the technical field of low-code development platforms, and particularly relates to a low-code platform model construction and dynamic execution method and system based on a hybrid DSL (Digital Subscriber Line), and the method comprises the steps: describing static model information in an application by using a structure DSL in a declarative grammar; defining a behavior DSL in a mounting mode under a corresponding element of the structure DSL; analyzing the structure DSL into a meta-model object tree in the platform, and converting the behavior DSL into an executable expression tree / intermediate representation; jointly driving the meta-model object tree and the expression tree / intermediate representation by utilizing a template engine, and automatically generating at least one application code in a front-end page, a back-end interface and a database script; a common structure DSL and a behavior DSL are combined and packaged into a scene template, and the scene template is stored in a template library so as to be reused by different service scenes, and modularization and templating of service modeling are realized. The complexity of the system is reduced, and the development efficiency and quality are improved.

The invention discloses a virtual space multi-person interaction synchronous control method oriented to an end-cloud collaborative architecture. The invention relates to a computer graphics and three-dimensional modeling technology, and discloses a cross-format lightweight and geometric consistency maintenance method based on a three-dimensional model. Through format-independent geometric representation and a self-adaptive lightweight strategy, efficient compression and precision maintenance of three-dimensional model cross-format conversion are realized. The method specifically comprises the following steps: performing format analysis and geometric feature extraction on an input model, and establishing a unified internal representation; adaptively selecting a multi-level LOD lightweight strategy based on the complexity of the model; the accuracy of key information is ensured through geometric feature keeping and topology consistency detection; the geometric consistency is dynamically maintained by combining error monitoring and an iterative correction mechanism; and generating a target format lightweight model and carrying out quality verification. According to the method, adaptive precision control, multi-level consistency maintenance and format irrelevant processing are combined, the model size and conversion errors are effectively reduced, and cross-platform compatibility and geometric fidelity are improved. The method can be widely applied to the fields of industrial design, game development, virtual reality and the like.

The invention discloses an intelligent foundation treatment scheme generation system and method based on a rule engine and a large language model, and belongs to the technical field of civil engineering. According to the method, the accuracy of a rule engine, the professionality of RAG and the flexibility of a large language model are combined, a hierarchical processing mechanism is adopted, tasks with different complexities are subjected to corresponding technical layer processing, verification is carried out in combination with a multiple verification mechanism, and therefore an optimal foundation processing scheme is obtained; and training a large language model by adopting a large model training method based on a structured reasoning process, and dynamically updating standard terms and expert knowledge. According to the method, the accuracy of the rule engine, the professional property of the RAG and the flexibility of the large language model can be combined, safety is ensured, intelligence and efficiency are improved, and quality is guaranteed.

The invention belongs to the technical field of polarization codes, and discloses a large nuclear polarization code BP decoding algorithm based on a pruning permutation factor graph, and the algorithm comprises the steps: firstly, randomly generating a permutation factor graph; secondly, deleting nodes with low contribution degree and non-contribution degree to decoding through pruning operation; and when the stop condition is not met, replacing the factor graph to execute BP decoding of the pruned and permutated factor graph until the stop condition is met. For a large nuclear polarization code, a permutation factor graph can improve the decoding performance, and a pruning factor graph can reduce the decoding complexity. A simulation result shows that compared with a BP decoding algorithm, the algorithm provided by the invention can ensure that the complexity is not too high and the decoding performance is improved at the same time.

The invention discloses an underwater acoustic processing CPU-GPU dynamic load balancing method based on a task flow model, and belongs to the technical field of underwater acoustic processing. The method comprises the following steps: deconstructing an underwater acoustic processing application into a task flow model represented by a directed acyclic graph; establishing a feature portrait including calculation complexity, parallelism and data throughput for each task node, and constructing a cost prediction model; the CPU / GPU utilization rate and the data transmission performance are monitored in real time; a processor is distributed to each task node by using a dynamic programming algorithm in combination with a cost prediction model and a real-time system state with the goal of minimizing the total task flow completion time; and dynamically scheduling tasks through a central scheduler according to a decision result, and periodically updating a strategy. According to the method, the defects that static task division lacks adaptability and neglects task dependence and communication overhead are overcome, dynamic and efficient utilization of CPU and GPU resources is achieved, and the efficiency and real-time performance of underwater sound processing are remarkably improved.

The invention relates to the technical field of code generation, and provides a code generation method, a power distribution network fault monitoring method, equipment, a medium and a product, and the code generation method comprises the steps: responding to a received code generation instruction input by a user, and carrying out the semantic analysis of the code generation instruction; according to a semantic analysis result, decomposing the code generation task to obtain a plurality of sub-tasks; aiming at each subtask, determining a code generation mode for executing the subtask according to the task complexity of the subtask so as to generate a code of the subtask; and generating the code of the code generation task based on the code generated for each subtask. The problem that a traditional code generation scheme is difficult to deal with a complex code generation task can be solved, the complex and multi-dimensional code generation task can be dealt with, and the code generation efficiency is improved.

The invention discloses a large language model output quality guarantee method, device and system based on chain type deterministic reasoning verification, and belongs to the field of artificial intelligence safety, deterministic calculation and reasoning verification. The method comprises the following steps: classifying each reasoning step generated by LLM into a deterministic step (bT: deductive reasoning, verified facts and mathematical proof) or a non-deterministic step (bF: inductive reasoning, unverified references and creative guess), and accumulatively constructing a reasoning chain; chain certainty verification chain (chain) = foldr (AND, true, chain) and time complexity O (n) are carried out in each k steps; if the bF is returned, a chain pollution elimination theorem is applied to prove that once the chain contains non-deterministic steps, the whole chain loses deterministic guarantee; and a deterministic boundary report is generated, and a non-deterministic initial position is marked. The device comprises an inference classifier, a reference verifier, a mathematical checker and a chain verification engine, and all the modules can be independently realized (formalized verification / database query / rule engine / mixed mode).

The invention relates to a method for memory access of tensor data, a computing device, a computer readable storage medium and a computer program product. The method comprises the following steps: converting an input tensor into a one-dimensional array; determining non-tail elements and tail elements of the one-dimensional array based on a vectorization unit and the total element number of the one-dimensional array; processing non-tail elements of the one-dimensional array in a vectorized continuous memory access mode; and processing tail elements of the one-dimensional array in an element-by-element memory access mode. According to the method, the processing efficiency of the tensor data in the non-alignment scene can be remarkably improved, and the code complexity and the maintenance cost are reduced.

The invention discloses a method for converting a natural language into a structured query language by considering accuracy and reasoning cost. The method comprises the following steps: firstly, performing task difficulty judgment on an input natural language query statement through a multi-dimensional scoring method; then diversified example data are generated through a data enhancement method, and an enhanced example library is formed; a semantic vector retrieval method is adopted, and a sample most relevant to input is selected; different cue words are generated according to different complexity levels of the query task, and large language models of different parameter scales are called for SQL generation; and finally, carrying out structure and semantic consistency verification on the generated SQL statement, if inconsistency exists, automatically constructing a correction prompt word, guiding the model to carry out correction, and finally outputting the SQL statement with a correct structure and accurate semantics. According to the method, the reasoning cost is remarkably reduced while the generation accuracy is improved, and the method has good practicability and expansibility.

The invention is applicable to the technical field of computers, and provides a big language model reasoning acceleration method based on bidirectional block floating points, which comprises the following steps: encoding an input text into a Token sequence, representing hidden representation and logits in a bidirectional block floating point number format in a Transform reasoning process, and combining a Softmax normalization method of a table look-up method based on bidirectional block floating points to obtain a big language model reasoning acceleration model; and low-bit efficient reasoning is realized, and a reasoning result is generated. According to the method, the calculation complexity and the storage overhead of the reasoning process are remarkably reduced while the generation precision is kept, and the reasoning speed and the energy efficiency ratio of the large language model are improved.

The invention provides a symbolic execution tree and LLM vector space heterogeneous fusion program defect detection method, relates to the technical field of program static analysis, and solves the problems of path explosion and overhigh constraint solution complexity in C / C + + program defect detection of traditional symbolic execution. The method comprises the steps that firstly, a symbolic execution tree of a target program to be detected is constructed, key feature information is extracted from the symbolic execution tree and converted into multi-dimensional feature representation, and corresponding symbolic execution feature vectors are formed; then constructing a mapping model, realizing a mapping process from the symbolic execution feature vector to an LLM vector space, and obtaining an LLM mapping result; symbolic execution analysis is achieved based on the symbolic execution tree, LLM analysis is achieved based on the LLM mapping result, and finally two kinds of analysis results are fused to obtain a program defect detection result. For optimization training of the mapping model, a comparative learning strategy is also adopted. According to the invention, accurate detection of complex program defects can be effectively realized, and the detection efficiency is improved.

The invention discloses a dynamic sparse cross-modal fusion data feature extraction method and system, and the method comprises the steps: multi-modal feature coding and alignment, and dynamic sparse cross-modal fusion, which are cooperatively completed by a near-end operator sparse controller and a Top-k sparse cross-modal attention module. After dynamic sparse fusion, features containing highly concentrated cross-modal information are obtained, and through a hierarchical data encoder module, the features of the cross-modal information are converted into data representation with a compact structure and rich semantics; a hybrid expert architecture is adopted; the system comprises a modal feature encoder, a dynamic sparse fusion layer, a data encoder and a task adapter. According to the invention, the calculation complexity of traditional cross-modal fusion is greatly reduced; the data has global consistency while keeping fine-grained details; and meanwhile, the richness of cross-modal semantics is kept, and the method is suitable for feature extraction tasks of multi-modal data such as images, audios and the like.

The invention discloses a two-stage algorithm selection and hyper-parameter joint optimization method, which comprises the following steps of: in the first stage, processing a training set and a test set through row sampling operation and column dimension reduction operation to form a reduced data set; randomly sampling a certain number of configurations in the hyper-parameter space of each candidate algorithm, evaluating the performance of each candidate algorithm by using the reduced data set, and extracting an optimal performance score; in the second stage, a previous algorithm is screened according to the optimal performance score to form a candidate set, and a pruned hyper-parameter search space is formed so as to reduce the calculation complexity of processor hyper-parameter search; and performing hyper-parameter optimization on the pruned hyper-parameter search space by using the original data set, and outputting an optimal algorithm adaptive to the target technical task and hyper-parameter configuration thereof. Algorithm screening and hyper-parameter tuning adaptive to a specific scene are realized through a two-stage optimization strategy, and meanwhile, the method is suitable for a traditional table type dichotomy task and aims at improving the deployment efficiency and performance of a machine learning model.

The invention relates to the technical field of computer programming languages, and particularly discloses a computer language programming method based on business logic. Aiming at the defects of deep coupling of business logic and technology implementation, high cross-domain adaptation cost, hardware operation code redundancy, high asynchronous programming complexity and the like in the prior art, the invention provides the following core schemes: a modularized development framework, a cross-platform adaptive mechanism, an event hierarchical model, a hardware intention analysis engine and a code, namely a document system. According to the method, the development complexity of a multi-field system is remarkably reduced, the business logic expression efficiency and maintainability are improved, the method is suitable for cloud micro-services, embedded equipment and mixed language scenes, and a technical basis is provided for intelligent programming.

The invention relates to a power network structure weakness detection method based on multi-diagonal-block matrix decomposition, and belongs to the technical field of super-large-scale integrated circuits. According to the method, an original large-scale sparse matrix is converted into a band edge diagonal block structure and divided into a plurality of sub-matrixes capable of being solved independently by constructing a layering and blocking strategy for eliminating tree drive, and redundancy of full-matrix calculation is avoided. Meanwhile, a local approximate inverse algorithm of column norm truncation is designed, and target elements are calculated on the premise that a preset error threshold value is met. According to the method, the parallel computing architecture of the multi-core processor is fully utilized, the computing complexity is reduced by 2-3 orders of magnitude, the computing efficiency is effectively improved, the memory occupation is remarkably reduced, and a high-precision and high-efficiency solution is provided for detecting the weakness of the power network structure of the super-large-scale integrated circuit.

The invention discloses a method and device for determining the number of test cases and electronic equipment. The method comprises the following steps: generating an initial orthogonal table of a target software system; determining a target service operated by the target software system, obtaining a complexity index variable, and determining a service influence index variable of the target service; determining a complexity factor value according to the complexity index variable and the service influence index variable; acquiring historical operation information of the target software system, acquiring an error rate index variable from the historical operation information and the software code, and determining an error rate factor value of the target software system according to the error rate index variable; and adjusting the preset number according to the complexity factor value and the error rate factor value to obtain a target number of the test cases of the target software system. By means of the method and device, the problem that in the related technology, due to the fact that the accuracy of determining the number of supplementary cases of the orthogonal table is low, the accuracy of software testing according to the orthogonal table is low is solved.

The invention discloses a layered LDPC (Low Density Parity Check Code) decoding method based on syndrome feedback self-adaption, and aims to solve the problem of poor decoding performance caused by fixed correction factors and poor adaptability to different check matrixes and channel conditions in the existing LDPC decoding algorithm. The method comprises the following steps of: firstly, dividing rows of a check matrix into a plurality of independent layers by adopting a hierarchical scheduling architecture so as to accelerate propagation and convergence of decoding information; secondly, in each iteration process of hierarchical decoding, a self-adaptive mechanism based on syndrome feedback is used, and a core offset factor beta in an offset minimum sum algorithm is adjusted in real time; the mechanism intelligently increases or decreases the offset factor by diagnosing the current decoding state (i.e., the number of unsatisfied check equations) to help the decoding process jump out of local optimum and stably converge. A simulation result shows that compared with a traditional offset minimum sum algorithm, the decoding performance is effectively improved and the bit error rate is reduced on the premise that the complexity is not obviously increased, and particularly, the performance gain is obvious in a medium-high signal-to-noise ratio region, and the robustness is higher.

The embodiment of the invention provides a model training method, an information recommendation method, equipment, a storage medium and a program product. According to the embodiment of the invention, a multi-encoder-multi-sub-decoder-total decoder hybrid model architecture is provided, sample data of different information modes correspond to different encoders-sub-decoders, and a mode of processing all training sample data by a single model is converted into a divide-and-conquer mode. The internal complexity of each encoder-sub-decoder is relatively low, the complexity of model training can be reduced, resource consumption can be saved, and different encoder-sub-decoders can be trained in parallel, so that the model training time can be shortened; and furthermore, by utilizing a dual decoding mechanism of the sub-decoder and the global decoder, parameters of the encoder can be continuously adjusted through local optimization and global optimization, the performance of the model is optimized, the accuracy of a reasoning result is improved, the convergence speed of the model is accelerated, the model training efficiency is further improved, and the model training time is saved.

This invention belongs to the field of CAD manufacturing information technology and relates to an efficient interference inspection method for complex structural CAD models. It employs a collision detection optimization algorithm using a layered bounding box assembly tree and synchronous recursive descent, comprising the following steps: Step 1: Establishing quantitative evaluation indicators; Step 2: Optimizing the layered bounding box assembly tree algorithm; Step 3: Recursively applying the results of the optimized layered bounding box assembly tree algorithm using a synchronous recursive descent algorithm. This invention uses a collision detection optimization algorithm combining a layered bounding box assembly tree and synchronous recursive descent, and uses synchronous recursive descent to calculate the nodes to be compared, thus solving the problems of high time complexity and inability to effectively balance efficiency and accuracy in existing methods.

The invention provides a Python algorithm development debugging system and method based on a simulation platform, the system comprises a TonderPy simulation engine service, a platform interface dynamic library file TonderPy.pyd, a Python interpreter and a Python algorithm model, the TonderPy simulation engine service is used for establishing communication connection with the simulation platform, driving the Python algorithm model, and dynamically optimizing an interface calling time sequence based on a collaborative scheduling algorithm; the platform interface dynamic library file TonderPy.pyd packages an interactive interface of the simulation platform and is used for obtaining situation information of the simulation platform, transmitting the situation information to the Python algorithm model and sending a task instruction to the simulation platform; the Python interpreter is arranged in a TonderPy simulation engine service and is used for loading and executing a Python algorithm source file; the Python algorithm model generates a task instruction according to the situation information; according to the system, standardized calling of a simulation platform interface is realized, and the algorithm development complexity is reduced; the data interaction efficiency between the algorithm and the platform is improved; a real-time closed-loop verification mechanism is constructed, the debugging period is shortened, and the utilization rate of system resources is improved.

The embodiment of the invention provides a model parameter compression method and device for a large language model, equipment and a storage medium. The method comprises the steps of obtaining verification data and inputting the verification data into a large language model to obtain an input activation tensor received by each network layer; obtaining the reasoning confusion degree of the large language model for reasoning the verification data, and taking the minimization of the reasoning confusion degree as an optimization target to carry out iterative cutting decision to obtain a cutting decision vector; performing outlier clipping processing on the input activation tensor of the network layer based on the clipping decision vector to obtain a target activation tensor; for each network layer, calculating an importance score of a model parameter based on the target activation tensor, and pruning the initial model parameter tensor in combination with the importance score to obtain an intermediate model parameter tensor; and performing quantization processing on the intermediate model parameter tensor of the network layer to obtain a target large language model. Therefore, the storage and calculation complexity of the large language model can be reduced while the performance of the large language model is maintained.

A method for adaptive loop filtering is provided that includes determining a coefficient value for each coefficient position of an adaptive loop filter, applying the adaptive loop filter to at least a portion of a reconstructed picture using the coefficient values, and entropy encoding coefficient values into a compressed bit stream using predetermined short binary codes, wherein the short binary code used depends on the coefficient position of the coefficient value.

The invention belongs to the technical field of power grid dispatching, and particularly discloses a dual-port equivalent cluster aggregation method. According to the method, an independent aggregation operation interval is established for two ports by solving a specific robust optimization problem, an initial aggressive boundary is formed, and an initial high-dimensional polyhedral model is established. The key improvement is that the condition extreme value of the power of the other port is solved under the constraint of the operation interval of one port, so that the physical coupling relation is converted into a series of accurate boundary points, and meanwhile, the calculation complexity is also reduced. And finally, cutting the initial and aggressive high-dimensional model by using the boundaries, eliminating an infeasible space generated by neglecting coupling, and quickly generating a feasible dual-port aggregation interval. The problems of large model error and low aggregation efficiency caused by dual-port interconnection are systematically solved.