121 results about "Backtracking" patented technology

The invention relates to the technical field of large language models, and discloses a zero-illusion large language model output method and system, and the method comprises the steps: sequentially carrying out the deterministic lexical analysis and standardized mapping of a received original query, and generating a standard term set; locking to-be-reasoned items through hierarchical matching to form a deterministic retrieval result set; executing five rounds of progressive verification processing on the result set to obtain an integrated result set with an LLM verification mark; and outputting a layered zero illusion result through multiple verification and auditing of structure compliance, semantic consistency and knowledge base double recheck. According to the method, probabilistic reasoning is replaced by full-link deterministic operation, so that each code can be traced back to a credible knowledge source, illusion is thoroughly eliminated, absolute reliability and complete traceability of an output result in a professional field are realized by a verifiable deterministic path, and a foundation is laid for zero-illusion application.

The invention discloses a large language model reasoning acceleration method and device based on two-stage speculative decoding and a storage medium, and the method comprises the steps: constructing and initializing a Trie tree, and inserting a historical corpus, and phrase sequences in a document library or a code library into the Trie tree one by one; in the reasoning process, longest prefix matching is carried out based on a Trie tree, and a candidate draft sequence is generated by adopting branch backtracking and recursive search; performing confidence evaluation on the candidate draft sequence, calculating a joint confidence score of the sequence through probability multiplication and a Top-K screening mechanism, and judging whether the joint confidence score reaches a confidence threshold; if the accumulated confidence of the candidate sequence reaches a threshold value, skipping a small model generation stage, and directly entering large model verification; otherwise, entering a small model draft completion stage; and the final large model takes the replaced and updated draft sequence as final output. According to the method, adaptive acceleration of the decoding process can be realized, and the long text reasoning delay of the large language model is remarkably reduced while the generation quality is ensured.

The invention discloses an interaction state backtracking and semantic reconstruction system based on a user operation log. The system comprises an operation log recording module which is used for capturing interactive operation of a user in an interface in real time and generating an event log with a timestamp; the state backtracking module is used for recovering the interactive semantic state of any time node according to the log; the semantic reconstruction module is used for allowing a user to derive a new interaction path on the basis of the backtracking state so as to realize branch evolution of a semantic structure; and the user control confirmation module is used for providing a confirmation mechanism in the state recovery and reconstruction process to ensure that final execution is triggered by a user. The technical problems that an existing system lacks historical operation tracking, and semantic rollback and reconstruction are difficult to carry out are solved.

The invention discloses a semantic unification and anomaly traceability analysis method and system for heterogeneous program logs, and the method comprises the steps: firstly collecting original logs in various formats, and packaging the original logs into standard objects in a unified manner; thirdly, identifying fields through a rule and a semantic model, uniformly mapping and labeling semantic tags, eliminating semantic differences, and constructing log data with uniform semantics; automatically constructing a cross-system call chain based on the request identifier and the like; secondly, extracting log features and comparing the log features with a normal behavior model to carry out anomaly detection, and generating an abnormal event; combining time, calling and dependency to construct an anomaly propagation path, and reversely backtracking and positioning an anomaly source; abnormal priority scores are calculated according to the propagation range, the influence degree and the like, and sorting and visual display are carried out; through unified semantic modeling and automatic association analysis of heterogeneous logs, accurate tracing and intelligent positioning of cross-service link anomalies are realized, the operation and maintenance efficiency is effectively improved, and the dependence on artificial experience is reduced.

The invention provides an intelligent course arrangement method based on multi-constraint hierarchical optimization and dynamic backtracking, and is applied to the technical field of data processing. According to the method, basic data of setups, classes, courses and the like are firstly obtained, and a multi-dimensional constraint model is established through preprocessing and is quantified into a time slot state value; based on a two-level layering strategy and dynamic teacher priority scheduling, independent classes are processed firstly, then sub-classes are processed, conflicts are cleared, and a stage course arrangement result is generated. In combination with the weekly class hours of the courses, time is distributed through an intelligent hall connection and degradation strategy, conflicts are detected in real time, and a backtracking mechanism is triggered. And verifying the conflict-free persistent data, and generating an initial course arrangement scheme. And finally, according to indexes such as course arrangement success rate and the like and application feedback, a constraint model, a scheduling algorithm and a class connection strategy are optimized, a multi-dimensional coupled intelligent course arrangement optimization scheme is formed, and the course arrangement problem under complex constraints is efficiently solved.

The invention relates to the technical field of data processing, and discloses a computing power server system layer optimization method and system oriented to a high-load scene. The method comprises the steps of collecting micro performance indexes such as the CPU instruction cycle number and the page table missing frequency, constructing a three-layer causal directed acyclic graph through Granger causal inspection, reducing a parameter search space based on bottleneck node reverse backtracking, generating an interpretable optimization decision with a causal path and contribution degree quantification, and carrying out optimization on the basis of the interpretable optimization decision. The problems that the performance bottleneck root cause cannot be accurately positioned and the optimization result lacks transparency in the prior art are solved. According to the method, bottleneck node reverse backtracking and parameter space pruning are performed based on the causal atlas, so that the problem of low optimization efficiency caused by incapability of accurately positioning a performance bottleneck root cause and blind exploration of a parameter space in the prior art is solved.

The invention relates to the technical field of software configuration management, and discloses a monorepo-based multi-version control and management method, which comprises the following steps: receiving a version switching instruction containing a target global version identifier and an environment identifier; loading the physically isolated environment version configuration file from a preset path according to the environment identifier; a target configuration node corresponding to the target global version identifier is positioned in the configuration file, recursive branch information merging operation is executed based on inheritance attributes and branch attributes in the node, and a final branch configuration set containing all sub-module target states is generated; and traversing the final branch configuration set, and calling a version control tool to switch the working area state of each sub-module to the corresponding target sub-module version identifier. Through the environment isolation and increment inheritance model, the problems of version definition redundancy and environment confusion in a multi-module warehouse are solved, and accurate backtracking and atomization switching of sub-module states are achieved.

The invention discloses a metadata version control and automatic structure comparison method and system based on incremental snapshot and intelligent decision, and the method comprises the following steps: managing a metadata version through an incremental snapshot model, and enabling the incremental snapshot model to support the complete backtracking of a historical structure through storing changed incremental information and associating with a version chain; carrying out difference analysis on the metadata structure by adopting a hierarchical progressive comparison algorithm; an intelligent DDL decision engine is adopted, and an executable DDL script is generated according to the difference analysis result; in the process of generating the DDL script, metadata consanguinity information is jointly checked so as to carry out change risk assessment. Through collaborative design of all the modules, double benefits of systematic breakthrough and conventional effect enhancement are brought, and the technical effect is directly associated with collaborative design.

The invention relates to the technical field of code security, in particular to a data stream compliance detection method and device based on code semantic rule-driven LLM, and the method comprises the steps: carrying out the normalization processing of a to-be-detected source code: removing annotation, null lines, unifying branch circulation into an if-else structure, and establishing lt; code lines and line numbers gt; performing mapping; constructing a code semantic rule knowledge base; generating a code intermediate expression by using LLM; generating an execution path for the intermediate expression, constructing a symbol table, and reasoning and positioning non-compliant variables and corresponding code lines according to a data flow rule; generating a detection report; the device comprises a code standardization processing function module, a knowledge base construction function module and other function modules for achieving the method, manual intervention is not needed, C / C + + code compliance can be automatically detected, the detection precision and efficiency are remarkably improved, and non-compliance positions can be accurately traced.

Methods, non-transitory machine-readable mediums, and system to provide historical positioning information are described. In an embodiment, a device detects one or more conditions in user contextual data that trigger collection of a set of historical positions for a lookback window, the device receives at least one historical position from the set of historical positions for the lookback window, the device classifies the at least one historical position based on one or more features as a candidate location for a backtrack route, and the device determines to provide the at least one historical position as part of the lookback window based on the classification.

The invention discloses a multi-agent collaborative power station index anomaly root cause step-by-step tracing method, and relates to the technical field of power system operation monitoring and intelligent fault diagnosis, and the method comprises the steps: collecting the operation index data of power station multi-level equipment, and constructing an index association topological structure based on the causal relationship between indexes; receiving a traceability request packet identified by the abnormal index, initializing an analysis agent of a corresponding level, and triggering collaborative analysis by taking the abnormal index as a traceability starting node; the analysis agent performs step-by-step traversal analysis on the upstream indexes with causal association along the index association topological structure to obtain an abnormal influence evaluation result; and aggregating the analysis results of each hierarchy, forming a causal propagation path from the abnormal indexes to the candidate root cause indexes, and outputting corresponding index identifiers, influence evaluation results and hierarchy information. According to the method, the power station abnormal index root causes are efficiently explained and traced.

According to the graph retrieval enhanced post recommendation method and system based on reinforcement learning, a high-quality post portrait knowledge graph can be automatically constructed based on a desensitization post recruitment data set, efficient community summary indexes and metadata backtracking are established, downstream retrieval and reasoning support are facilitated, and on this basis, the post recommendation efficiency is improved. The ranking model of recalled posts is trained by a reinforcement learning low-rank fine tuning framework based on the GRPO algorithm, accurate post pushing is realized, and the method has great advantages in interpretability and recommendation precision.

The application provides a multi-agent cooperative task state management and abnormal backtracking method and system, relates to the technical field of artificial intelligence, and comprises the following steps: obtaining interaction data, state data and execution logs, constructing an initial task knowledge graph based on an ontology model, establishing a bidirectional causal chain by using a Bayesian network to form a dynamic knowledge graph, analyzing state migration rules when an abnormality is detected, identifying a propagation node, constructing an abnormal propagation subgraph, generating an influence chain, determining a recovery priority and triggering state rollback. The method can quickly trace the abnormal source, effectively control abnormal diffusion, and improve system robustness and task recovery efficiency.

The invention provides a logic analyzer trigger point accurate backtracking control method based on hotspot scoring and boundary learning, and the method comprises the steps: carrying out the popularity scoring according to the ratio of the number of channel compression units in each message segment to the length of an occupied byte, and screening out a hotspot candidate set; analyzing the structure of the hot area message segment and decoding, and realizing time alignment of a multi-channel sampling sequence in combination with the position offset adjusting quantity to form a synchronous sampling matrix; calculating local change intensity based on the matrix, introducing an inter-channel jump consistency constraint to construct a jump probability score, and accurately positioning a trigger point time index and a message segment where the trigger point time index is located; and finally, mapping a logic time position into a physical storage address in combination with a compression structure non-balance compensation parameter, so as to realize efficient reverse reading. According to the method, a mark or an additional address table does not need to be inserted, and high-precision backtracking control of the trigger point can be realized under the conditions of high compression ratio and multiple channels.

The invention provides a 3D printing task scheduling method based on an optimization backtracking algorithm. The 3D printing task scheduling method comprises the following steps that all to-be-printed task information and equipment resource information are collected and modeled; calculating the priority weight of each task based on the task attributes in the agricultural scene; task constraints adaptive to agricultural application requirements are constructed; an optimization backtracking algorithm is adopted, and an initial scheduling scheme is generated by taking minimization of the total completion time and the switching cost as targets; carrying out conflict detection and correction on the scheduling scheme; outputting an optimal task scheduling result adaptive to the agricultural production rhythm; the invention aims to solve the problems of insufficient agricultural time adaptation, relatively high loss, single constraint, poor flexibility and the like of the existing scheduling technology aiming at the requirements of agricultural time sensitivity, material particularity and task diversity of the agricultural scene 3D printing task, realizes efficient, accurate and flexible scheduling of the agricultural 3D printing task by optimizing the backtracking algorithm and customizing the model design, and improves the scheduling efficiency of the agricultural scene 3D printing task. The farming progress is guaranteed and the production loss is reduced.

The invention discloses a heterogeneous fuzzy and game-based double-layer adaptive evolution CAD modeling command generation method, and belongs to the technical field of artificial intelligence and computer-aided design intelligent modeling. The method solves the technical problems that in an existing natural language driven CAD command generation technology, fuzzy semantics are difficult to quantify and map, heterogeneous commands are poor in adaptability, the executable performance of generated commands is low, and the reasoning process is black-box-shaped. According to the method, a semantic analysis-game generation-evolution refinement-verification backtracking-knowledge precipitation full-process closed-loop architecture is constructed, the method can be applied to the fields of industrial software, three-dimensional modeling, intelligent design systems and the like, the engineering performability, generalization ability and reasoning interpretability of command generation are improved, the generation diversity and feasibility are balanced, and the method is suitable for popularization and application. And the model iteration cost is reduced, and technical support is provided for industrial landing of the text-driven CAD modeling technology.

The invention discloses a coherent radial velocity spectrum target detection method based on a backtracking filter. The method comprises the following steps: constructing a sea surface echo model and a multi-frame joint detection problem; the method comprises the following steps: based on a sea surface echo model and a multi-frame joint detection problem, performing coherent accumulation on a radar receiving pulse sequence, and calculating test statistics of intra-frame coherent accumulation under each frame, each distance unit and each radial speed; based on the inconsistency of the sea clutter structural texture between the slow time Doppler velocity and the frame time radial velocity and the velocity consistency of the target on two time scales, inter-frame accumulation is carried out on the test statistics of each frame, each distance unit and each radial velocity in a distance-frame time plane along the radial velocity direction; constructing a coherent radial velocity spectrum; and in an inter-frame accumulation process, replacing a linear search accumulation mode of the coherent radial velocity spectrum between scanning by using the backtracking filter bank to obtain a coherent radial velocity spectrum based on the backtracking filter bank, and carrying out target detection by using the coherent radial velocity spectrum.

The application relates to the technical field of natural language processing, and particularly discloses a group risk cognitive consensus construction method based on risk metaphor, multi-source interactive texts are collected, and after extraction and filtration, an initial metaphor cognitive set is formed; a dynamic relationship network is constructed based on the set, responsibility is quantified as an attribute quantity transmitted between nodes, and a responsibility distribution graph is generated through iterative calculation; the transmission path is calculated, the dispersion and backtracking features thereof are quantified, and a quantitative analysis report for identifying a high-risk responsibility dilution path is generated; a responsibility anchoring coefficient is calculated according to the report, semantic reconstruction and iterative optimization are performed on the initial metaphor, and multiple candidate consensus expressions are generated; the robustness of the responsibility distribution under each candidate expression is tested by simulating an extreme cognitive bias and a long-time-span stress situation, and an expression that is most stable and satisfies responsibility convergence and traceability conditions is selected as an output of a final consensus framework; the application outputs a consensus framework with clear responsibility anchor points and anti-interference capability.

This invention provides a method, apparatus, electronic device, and storage medium for complex event evolution deduction. It generates a macro-planning tree based on scene description information, containing temporal stages and macro-objectives. Under the constraint of the macro-objective at the current time step, it drives multiple agents to generate micro-level actions, thereby determining candidate states for the next moment. It extracts micro-evolutionary trends and calculates a consistency metric between these trends and the macro-objective. When the consistency metric is low, it triggers dynamic adjustments to actions or plans to obtain the next state of the target. Finally, it updates the deduction results of the complex event, constructing a complete closed-loop deduction framework that unifies macro-planning, micro-behavior generation, state transition synthesis, and value backtracking arbitration. This significantly reduces deviations caused by the emergence of multi-agent behaviors in long-chain deductions and improves the credibility, stability, and interpretability of deduction results in open environments.

This invention belongs to the field of integrated circuit design technology, specifically relating to a method for optimizing integrated circuit design data. This method first constructs a "tool-layer-semantic" data acquisition architecture, integrating multi-channel parallel acquisition units to collect heterogeneous data from multiple sources. Compliance pre-verification is performed through hardware logic circuits and a verification rule base, extracting key data identifiers. A three-level incremental compression strategy based on key data priority, combined with an intelligent index integrating multi-dimensional indexes, is used for compression and storage. The stored data undergoes multi-dimensional optimization, including semantic deduplication, priority optimization, format conversion, and source quality backtracking, forming a unified format optimized semantic. Through an interactive interface, it is adapted and interacted with electronic design automation tools. After consistency verification and performance testing feedback loops, the final design data and full-process traceability information are obtained and stored. This invention achieves unified management, efficient compression, intelligent optimization, and full-process traceability of design data.

The invention discloses a loose coupling unified task arrangement method supporting multi-dimensional expansion, which comprises the following steps of: firstly, defining a transmission agreement of task arrangement: constructing a unified intermediate representation structure for describing an arrangement process, the intermediate representation structure being used for recording a process topological structure, a node type and parameter configuration; secondly, on the basis of the intermediate representation structure, an executable directed acyclic graph is automatically constructed by adopting a child node matching construction algorithm based on queues and backtracking; and thirdly, executing an arrangement process based on the constructed directed acyclic graph: starting from all write nodes, recursively calling a data processing method of a father node until reaching a read node, and executing a complete data processing link in a bottom-up manner. Through a unified intermediate representation structure and a queue backtracking matching algorithm, automatic topology construction and dependency management of a complex process are realized, flexible expansion of node types is supported, development complexity is reduced, and adaptability, efficiency and maintainability of a data management system are improved.

The invention relates to a backtracking video generation method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: in response to a trigger operation of a target object for a subprogram, generating and reporting a first record identifier matched with a native page of the subprogram under the condition that the subprogram starts a backtracking function; in response to a first interaction event triggered on the native page, generating and reporting native page record data of the native page; responding to a second interaction event triggered on the webpage linked with the subprogram, and generating and reporting webpage record data of the webpage; and under the condition that a backtracking request carrying the first record identifier is received, generating a target backtracking video for the target object based on the native page record data and the web page record data which are stored in association with the first record identifier. By adopting the method, the accuracy of backtracking video generation can be improved.

The invention relates to the technical field of computers, and discloses a large model reasoning service dynamic connection number limiting method based on service attributes, which comprises the following steps of: 1, carrying out initialization setting on service attributes, resource parameters and a weight model; 2, collecting multi-dimensional data, wherein the collected multi-dimensional data comprises load data, service data and resource data; step 3, carrying out data preprocessing and data fusion on the collected multi-dimensional data to form a three-dimensional fusion data set; step 4, calculating a three-dimensional dynamic weight based on the three-dimensional fusion data set; 5, generating a connection number threshold value according to the three-dimensional dynamic weight, and carrying out connection request distribution and over-limit processing; and step 6, forming closed-loop feedback through real-time monitoring, historical backtracking and algorithm tuning. By utilizing the scheme of the invention, the refined management and control of the connection number in the large model reasoning service are realized, and the resource cost is optimized while the high-priority service quality is ensured.

This invention provides a path planning method based on the DYPIBT algorithm, which can effectively solve the problem of path collisions between agents in narrow and busy terrain, reducing ineffective agent movements and chain congestion. It dynamically calculates priorities based on four dimensions: task urgency, remaining distance, local congestion coefficient, and unique ID, ensuring that each agent has a different priority at the same time step. For each location on the map, it performs predictive backtracking judgment, anticipating the next congestion risk before the agent moves and triggering backtracking in advance to reduce ineffective agent movements. To address the path conflict problem, it designs a high-priority priority rule and a priority inheritance rule. After a high-priority agent occupies the target location, a low-priority agent inherits the priority of the high-priority agent, giving agents that have already yielded the right-of-way an opportunity to gain priority in the next path selection, thus reducing ineffective movements for individual agents.

The invention provides a method for searching a longest circuit link based on a layer connection file, and the method comprises the steps: constructing a bidirectional connected graph model based on all layers and connection relationships in the layer connection file; traversing the graph model by adopting an improved depth-first search algorithm to search for a longest path; the improvement comprises the following steps: managing via hole layer nodes and physical layer nodes as access states together, and executing complete backtracking containing the states; and after all traversal is completed, outputting the finally obtained longest path. According to the method, the complex topology can be robustly processed and the longest circuit link can be accurately extracted by constructing the bidirectional connected graph model and bringing the via hole layer into a complete backtracking mechanism of state management.

The application discloses an intelligent terminal semantic understanding optimization method based on deep learning, and relates to the technical field of intelligent terminal semantic understanding. When a current short instruction is received, the method reads multi-source interface events in a preset time window and constructs a short-time interface state transition graph. Based on the short-time interface state transition graph, a local causal chain is extracted and an interface attribution clue sequence is formed. Double-granularity semantic coding and slot constraint decoding are performed on the current short instruction text, and cross alignment is performed with the interface attribution clue sequence to generate a task attribution candidate result. Path backtracking and consistency checking are then performed to determine a target attribution result. Finally, a call entry is determined according to the target attribution result and written into a short-range semantic memory cache. The application is suitable for multi-application switching, page jumping and short instruction continuous input scenes.

According to the system bottleneck root cause positioning method and system provided by the invention, through calculation of the criticality index, a key thread which causes abnormal performance of the system bottleneck is quickly identified. And constructing a dynamic waiting graph containing a waiting relationship between the threads, and finding out a sleep thread causing blockage in combination with the key thread and the waiting graph. And performing deep backtracking along a dependency chain of the sleep thread, determining a root cause thread which has the greatest influence on the overall performance based on the calculated root cause score, performing accurate CPU hotspot portraying on behaviors of the root cause thread in a resource holding period, and directly positioning a performance bottleneck to a specific source code. According to the scheme, hardware does not need to be customized, the application range is wide, a clear causal chain can be obtained, the threshold of a developer for understanding and repairing complexity performance problems is lowered, source codes can be directly positioned, and a complete attribution solution is achieved.

The invention discloses a fuzzy rule-based weight coefficient design method, which comprises the following steps of: after a system is initialized, acquiring a plurality of network parameters and a plurality of auxiliary parameters in a measurement window, and standardizing the plurality of acquired auxiliary parameters; performing adaptive reward aggregation on the plurality of standardized auxiliary parameters to obtain a final reward; an objective function is established according to the final reward, and candidate weight coefficients are generated; performing KL constraint on the candidate weight coefficient to obtain KL divergence, and when the KL divergence is smaller than or equal to a KL threshold value, performing the next step; when the KL divergence is greater than a KL threshold value, carrying out step reduction to recalculate a candidate weight coefficient or carrying out rollback to calculate a safety weight coefficient; and smoothing the candidate weight coefficient or the safety weight coefficient to obtain a final weight coefficient. According to the design method provided by the invention, a better weight coefficient can be automatically approached in different service scenes, a sawtooth effect and periodic degradation caused by multi-stream synchronization are reduced, and frequent manual parameter adjustment is not needed.

The invention discloses a reverse debugging increment recording method based on dynamic layering and predictive filtering, which comprises the following steps of: analyzing a binary file division region of a target program, establishing a global region mapping table among an address range, a region type and debugging attributes, executing the binary file by a reverse debugger, constructing a first mapping table for modifying a record circular queue, according to the global region mapping table, performing differentiation recording on the region to form a first record, a stack frame snapshot, a second record and a register block snapshot, and performing repeated coverage filtering to reduce the recorded data volume to form an execution state track; and performing intelligent compression on stack frame snapshots, heap memories, global variable records, register snapshots and variable values in the execution state trajectory, and finally forming a low-overhead and accurate execution state trajectory to support reverse debugging backtracking.