37 results about "Correctness" patented technology

The application discloses a kind of compiler automatic debugging method and system for VLIW and SIMD architecture, the method of the application includes the semantic correctness verification for the program to be checked to judge whether the program to be checked exists semantic error relative to source program, if semantic correctness verification finds that there is semantic error, then determine that debugging is not passed, otherwise the physical register verification for the program to be checked to judge whether the program to be checked exists physical register allocation error, if there is physical register allocation error, then determine that debugging is not passed, otherwise determine that debugging is passed;When determining that debugging is not passed, then generate feedback verification report.The application can automatically check the semantic correctness of program in the process of compiling, and provide accurate verification report (including error occurrence position and type etc.) for developer, improve the efficiency of compiler development, reduce the burden of developer.

An adaptive redundancy-scheduling system for column-oriented neural-network compute arrays is disclosed. The system periodically verifies correctness of compute columns through concurrent redundant-column execution and adjusts test intervals and substitution thresholds in response to observed error statistics. Redundant columns are activated only when measured fault probability exceeds a programmable confidence level, maintaining yield and reliability with minimal performance impact. Verification tasks occur during non-critical compute intervals, and error history is retained for long-term reliability prediction.

This application relates to the fields of artificial intelligence and data processing technology, specifically providing a data management method, system, device, and medium based on a dual-track verification and self-healing mechanism, aiming to solve the problems of compatibility and logical errors between multiple databases. To this end, the data management method of this application includes: generating first and second target structured query statements based on a user-provided natural language question; verifying the completeness and consistency of the first target structured query statement to obtain a first verification result; verifying the correctness of the execution logic of the second target structured query statement to obtain a second verification result; if the first verification result and / or the second verification result are verification failures, obtaining verification error information, regenerating new first and second target structured query statements based on the verification error information and the natural language question, and continuing verification; otherwise, retrieving target data from the target database based on the second target structured query statement to improve query accuracy.

This invention presents a confidential image similarity determination method, belonging to the field of collaborative computation technology for multi-party privacy data. To address the low execution efficiency of existing secure multi-party computation methods for image similarity determination, and the significant impact on efficiency with increasing data volume, this invention combines the Paillier homomorphic encryption algorithm with Gödel coding technology to design a confidential image similarity determination protocol under a semi-honest model. The security and correctness of the protocol are analyzed. Furthermore, this invention designs a confidential image similarity determination protocol under a malicious model. This protocol uses zero-knowledge proofs and a segmentation-selection method to detect malicious behavior, and its security is proven using an ideal-practical paradigm. Verification shows that this invention exhibits very good execution efficiency.

Embodiments of the present disclosure provide a scheme for code generation. A method includes receiving a query indicative of a code generation problem; generating, using a target machine learning model, a first code segment based on the query and a prefix code, wherein the target machine learning model is constructed based on a language model and configured to iteratively generate code segments to form computer code for the code generation problem; determining, using a trained reward model, a first reward score corresponding to the first code segment based on an updated prefix code, the first reward score indicative of correctness of the first code segment with respect to the code generation problem, wherein the first code segment is appended to the prefix code to obtain the updated prefix code; and fine-tuning the target machine learning model based on at least the first reward score through reinforcement learning.

This application provides a mathematical problem-solving method, an electronic device, and a storage medium. On one hand, the electronic device or server can verify the correctness and completeness of the solution steps corresponding to a first mathematical problem input by the user in natural language, making the logic of the solution steps more rigorous and improving the accuracy of the solution. On the other hand, the electronic device can output solution steps in both natural language and formal language, or output verification results in both natural language and formal language, based on the first mathematical problem input by the user in natural language, demonstrating the solution process or verification process to the user, thus improving the authority and interpretability of the mathematical solution.

This invention provides an intelligent RTL code generation method and system based on MCP and Agent collaboration. The method involves: acquiring user-input circuit design requirements; initializing the Agent and large language model environment, configuring the Agent's role definition and system prompts; performing compilation and syntax checking; obtaining execution feedback results from EDA simulation tools to determine if compilation passed; generating a structured error correction report; the Agent receiving the structured error correction report and generating corrected Verilog RTL code; resubmitting the code until it passes compilation checks or reaches a preset iteration threshold. This invention significantly improves the compilability and correctness of generated code and reduces the threshold and cost of manual design by constructing an automated closed-loop process of "code generation – syntax checking – error parsing – iterative repair."

The invention discloses a program control instruction generation method based on large model federation prompt fine tuning, which realizes the collaborative utilization of multi-party data on the premise of protecting the data privacy of each party by constructing a multi-level conditional diffusion model federation learning architecture. The problem of rapid adaptation of new equipment is solved through a heterogeneous document-oriented hierarchical retrieval enhancement generation method, and the problems of information fragmentation and text discontinuity in the traditional RAG technology are solved by constructing an operation process knowledge graph and a multi-granularity semantic representation learning framework and adopting a continuous perception retrieval mechanism. And finally, solving a fundamental contradiction between large model probability generation and test instruction preciseness through a constraint satisfaction deterministic generation mechanism, and converting a probability generation process into a strict logical reasoning process by adopting a beam search algorithm guided by a complete deterministic constraint. By constructing a three-dimensional constraint system of grammar constraint, semantic constraint and security constraint and a multi-stage verification mechanism, each generated instruction is ensured to have uniqueness, correctness and security.

The present application relates to a cross-language code intelligent auditing and automatic repair system. The system forms a quality and safety governance closed loop through code acquisition and preprocessing, multi-language unified intermediate representation (fusion AST / CFG / DFG / call graph / symbol table), general rule library detection, AI semantic enhancement, repair patch generation and multi-level verification, and feedback learning. The core is to constrain the repair generation of large language models with an interpretable evidence chain, and through automatic verification and failure rollback mechanism containing syntax, test, performance and security, the correctness and minimum change of the patch are ensured. The present application also relates to a corresponding method, device, processor and computer readable storage medium thereof. The cross-language code intelligent auditing and automatic repair system, method, device, processor and storage medium thereof adopt the present application, realize consistent risk detection across languages, significantly reduce the false positive rate and the risk of introducing new defects by repair, and maintain long-term effectiveness through a continuous learning mechanism.

This application provides a method, electronic device, and storage medium for testing the correctness of vectorization processing. The method includes: automatically enumerating multiple parameter combinations of test condition items (such as data type, broadcast rules, transpose state, etc.) related to the type of the operator to be tested; automatically generating a corresponding input tensor for each parameter combination; and performing vectorization processing and sequential loop processing respectively. By comparing the differences in the outputs of the two processing methods, the correctness of vectorization processing under all parameter combinations can be systematically verified. This overcomes the limitation of manually writing test cases that cannot fully cover multi-dimensional test conditions, and can efficiently and automatically discover boundary errors and numerical instability problems caused by different parameter combinations, thereby more accurately and reliably determining the correctness of vectorization processing.

The application discloses an electronic voting method and system based on homomorphic secret sharing. The method is as follows: reserving bit width for each candidate according to the set maximum number of votes, determining the total bit width according to the number of candidates, and setting the single vote code as a binary string with a length; a group of value points are disclosed, each value point corresponding to a collection center; each voting terminal constructs a random polynomial with a constant term and a degree, and obtains shares by evaluating the random polynomial at different value points and sends the shares to the corresponding collection center; the integer corresponding to the single vote code; each collection center adds the shares received to obtain the corresponding share sum; the value points and share sums of any collection center are used to generate a total code, the total code is sliced and decoded according to the reserved bit width, and the total number of votes for each candidate is obtained. The application significantly simplifies implementation and operation and maintenance while ensuring privacy and correctness.

This invention relates to the field of data security and privacy computing, specifically to a secret-sharing-based method for solving the k-th minimum value in a distributed dataset D, efficiently solving for the k-th minimum value under a semi-honest security model. The core of this method lies in adaptively managing the data state of the participants and designing and utilizing a lightweight secure comparison algorithm to determine each bit value of the k-th minimum value sequentially from the most significant bit to the least significant bit. While ensuring the correctness of the result, it reduces the number of interaction rounds and message volume, avoiding a linear increase in communication volume with the size of the participants and the data bit width, thus reducing the heavy reliance on multiple secure comparisons and bit-by-bit determinations. Simultaneously, it tightens the information leakage surface, allowing only the minimum necessary information that can be derived from the result to be output, avoiding the leakage of additional information.

The embodiments of the present specification provide a data verification method and device. According to the method, a computing device first acquires function information of a target function group of a target arithmetic circuit, wherein the target arithmetic circuit is used to perform a target computing task instructed by a verification device, and the target function group is used to describe a circuit structure of the target arithmetic circuit. The computing device determines a first function according to a current computing process of the target computing task, wherein the first function is used to describe input and output values of each gate in the target arithmetic circuit in the current computing process. Then, the computing device proves that the first function satisfies a plurality of preset constraint relationships by performing an interactive argument protocol with the verification device, thereby proving the correctness of the current computing process, wherein the plurality of constraint relationships include a constraint relationship between the first function and the function information of the target function group.

The application belongs to the field of natural language processing, and specifically discloses a code generation hallucination mitigation method based on multi-source uncertainty modeling, which comprises the following steps: obtaining token probability distribution of the final output layer of a main model, an intermediate hidden layer and an auxiliary light model in parallel; constructing a multi-source trusted candidate set; calculating direction-aware asymmetric comparison weight, constructing a direction-aware factor according to the probability ratio of the intermediate layer and the final layer; estimating the slope based on the information entropy sequence of the historical decoding steps, and adjusting the timing trend gain; calculating the multi-source comparison decoding score and selecting the token with the highest score. The application also discloses a corresponding system. Through mechanisms such as direction-aware asymmetric comparison, timing trend adjustment and confidence gating, the application can effectively suppress hallucination in code generation without relying on additional training, and improve the grammatical correctness, logical consistency and executability of the generated code.

The application discloses an automatic optimization method and system for a template calculation core carrying communication semantics, wherein interval of a Stencil calculation core and communication associated information are explicitly marked in source code through compilation indication; semantic information necessary for optimization in the calculation core is extracted, and IR is used for abstraction; for the required IR, control flow stripping and conditional lifting are performed, boundary strips and internal domain calculation are explicitly separated, and internal domain calculation is abstracted; based on environment variables related to hardware characteristics, a discrete optimization problem with resource constraints is solved to generate deterministic scheduling information, a Halide generator source file and a glue code are generated, and the original program is safely replaced. When any link does not meet the condition, only the original source code execution path is retained, and the backtracking reason is recorded; the remaining calculation cores meeting the condition are still processed according to the flow, so that the Stencil kernel optimization and code generation flow with the characteristics of correctness priority, backtracking and reproducibility are realized.

This invention discloses a prefix caching method and apparatus suitable for linear attention model architectures, introducing a linear attention state snapshot mechanism based on hash indexes. During the capture phase, the system extracts the convolution and loop states of linear attention at block boundaries as snapshots and stores them in a mapping table associated with the block prefix hash value. During the recovery phase, runtime slots are allocated for new requests that hit the prefix, and the snapshots are written in parallel to the global state tensor by index using a state scattering kernel engine, thereby skipping the linear attention computation for already hit prefixes. The snapshot pool uses a least recently used strategy for capacity management and automatic eviction, and performs binding validity checks before pre-filling computation. If a snapshot is asynchronously evicted, reference stripping and state clearing are used to fall back to the full pre-filling mode to ensure correctness. This method effectively avoids redundant computation, reduces first-word latency, and improves inference throughput, while balancing system stability and resource utilization efficiency.

This application discloses a testing method and related apparatus for a multi-level storage lookup architecture. The method includes: generating a random sequence based on random stimuli in the first simulation; simulating the multi-level storage lookup architecture based on the random sequence; forcibly writing configuration data associated with the random sequence into each level of memory in the simulation to obtain a configuration file; and performing functional verification of the multi-level storage lookup architecture based on the random sequence and the configuration file in the second simulation. In the first simulation, the application forcibly writes associated configuration data into each level of memory to obtain the configuration file; in the second simulation, it accurately reproduces the key fields and timing of the test messages generated in the first simulation; and performs functional verification of the architecture based on the configuration file. The configuration file can be dynamically and flexibly configured to ensure comprehensive coverage of various test scenarios, verify functional correctness, and flexibly adjust test scenarios, effectively solving the problems existing in traditional testing methods.

This application provides a method, apparatus, device, and readable storage medium for optimizing UI lag in a Swing system. The method includes: an Event Deployment Thread (EDT) detecting whether each code block in the target user code is a time-consuming code block, wherein the target user code is bound to the UI event currently being responded to by the EDT; the EDT encapsulating the time-consuming code block into a time-consuming task and submitting it to an asynchronous task executor, which allocates a background thread independent of the EDT for execution; after submitting the time-consuming task, the EDT calls the `enter` method of the `SecondaryLoop` interface to start a secondary EDT event loop, blocking the execution of subsequent code blocks in the target user code while responding normally to new UI events; after receiving feedback that the time-consuming task has completed execution, the EDT calls the `exit` method of the `SecondaryLoop` interface to exit the secondary EDT event loop, resuming the execution of subsequent code blocks in the target user code. This application can solve the lag problem while ensuring the correctness of the original logic, without requiring large-scale modification and refactoring of the user code.

This application discloses a method, apparatus, device, and medium for assertion code verification based on a pre-trained model, relating to the field of code generation technology. The method includes: extracting timing relationships from hardware design code; obtaining corresponding reference assertion code based on the timing relationships, and inputting the timing relationships into a pre-trained model after describing them in natural language to generate assertion code to be tested; integrating both into the same verification environment and applying an excitation signal, recording each successful trigger moment during simulation; comparing the successful trigger moments of the assertion code to be tested with those of the reference assertion code, and determining the functional correctness index of the assertion code to be tested based on the number of time matchings and the total number of triggers of the reference assertion code; and evaluating the functional correctness of the assertion code generated by the model based on the functional correctness index. Through the technical solution of this application, assertion logic errors can be avoided, establishing a benchmark and feedback loop for the reliable application of pre-trained models in the verification field.

The application discloses a self-adaptive fine-tuning intelligent question reasoning and feedback method and system, and relates to the technical field of intelligent text analysis. The method comprises the following steps: system initialization; receiving user input, extracting question text and mathematical formula, converting the question text and the mathematical formula into LaTeX and natural language description, and integrating the question text and the mathematical formula into structured data; generating a problem solving thought chain containing question understanding, step decomposition, step-by-step reasoning and result verification; positioning an error step and determining the type by means of semantic similarity calculation, sequence alignment and step matching degree evaluation, comprehensively obtaining an answer correctness conclusion and missing step information, generating improvement suggestions and outputting in a multi-modal form; and simultaneously, performing self-adaptive fine-tuning on a basic model by using LoRA according to a trigger condition, so as to adapt to the question style and difficulty of a user. The application improves the accuracy and individualization of correction, and simultaneously considers the reasoning process explainability, deployment cost controllability and multi-modal input processing capability.

The application discloses an error correction code standardization verification component and a verification method thereof, wherein the error correction code standardization verification component comprises a test object, an error injection module and a result expectation module; the error injection module monitors the behavior of the test object, and when it is monitored that the test object is performing a write operation of RAM, injects an error correction code error into the RAM; the result expectation module, when it is monitored that the test object is performing a read RAM operation, judges whether the read RAM operation request will occur an error correction code error and specific error information according to the error injection information received from the error injection module, obtains an expected value, and when the test object reports error information, compares the reported error information with the expected value to judge the correctness of the error information reporting behavior of the test object; wherein the error injection module and the result expectation module are both designed by using a parameterized class, and are packaged with the test object into an independent error correction code standardization verification component. The error correction code standardization verification component has high reusability and efficient error detection.

The application discloses a kind of operator automatic tensor promotion and mathematical semantic verification method and device, belong to program analysis and formal verification field.This method includes: first, the bottom layer implementation of target operator is carried out symbol execution, extracts the symbol constraint between its input and output as behavior specification;Subsequently, based on the program synthesis strategy of combination of divide and conquer search and enumeration, candidate mathematical formula satisfying specification is automatically synthesized in high-level tensor operator language;Then, the correctness of synthesis formula on any scale input is formally verified in combination with Hall logic and property transformation technology;Finally, using algebraic rewriting technique based on equivalent saturation to simplify complex formula automatically, restore its natural, intuitive mathematical expression form.The present application can not only significantly improve the understandability and reliability of operator implementation, but also assist developers in auditing, debugging and reusing complex operators, improve the reliability and security of deep learning system.

Provided in the present application are a mathematical problem processing method, an electronic device and a storage medium. On the one hand, an electronic device or a server can verify, on the basis of a first mathematical problem in natural language input by a user, correctness and completeness of problem solving steps corresponding to the first mathematical problem, allowing the logic of the problem solving steps corresponding to the first mathematical problem to be more rigorous, thereby improving the accuracy of solutions to mathematical problems; on the other hand, the electronic device can, on the basis of the first mathematical problem in natural language input by the user, output problem solving steps in natural language and problem solving steps in formal language, or output a verification result in natural language and a verification result in formal language, so as to display a problem solving process or a verification process to the user, thereby improving the authoritativeness and interpretability of mathematical problem solving.

The invention relates to the technical field of natural language processing, and discloses an SQL generation method and system based on multi-agent interaction clarification, a terminal and a medium, and the method comprises the steps: obtaining original information of a user, obtaining an intention result of the user according to the original information, if the intention result is intention fuzziness, receiving a first clarification reply of the user, updating the original information according to the first clear reply until the intention result is normal; performing validity verification on the original information to obtain a verification result of the user, if the verification result is data fuzzy, receiving a second clear reply of the user, updating the original information according to the second clear reply until the verification result is valid, and obtaining historical information; generating matching data according to the historical information; and generating an SQL statement according to the historical information and the matching data. According to the method, the table retrieval accuracy and the SQL execution accuracy under the fuzzy query scene are improved.

The application discloses a deep learning kernel fusion method and device based on portrait driving analysis, and belongs to the technical field of automatic code generation. The method comprises the following steps: obtaining a computation graph representation of a to-be-optimized deep learning model, performing conflict and dependency analysis, and obtaining a candidate fusion subgraph; performing benefit-driven cost evaluation on the candidate fusion subgraph to obtain a high-value candidate subgraph set; constructing a structured code generation context for the high-value candidate subgraph, calling a large language model to generate candidate fusion kernel code; performing correctness verification, portrait feedback and iterative repair on the candidate fusion kernel code, and outputting a target fusion kernel that passes the verification. The application can improve the fusion success rate under the premise of guaranteeing semantic correctness, reduce the performance loss caused by intermediate tensor materialization, register overflow and memory access mismatch, and is suitable for deep learning compilation and inference optimization scenes such as PyTorch and Triton.

An information processing system includes a processor configured to: accept, as a piece of teacher data, a piece of first table data and a piece of second table data between which results of matching indicate correctness; combine each of pieces of first column data in the piece of first table data and each of pieces of second column data in the piece of second table data, the each of the pieces of first column data and the each of the pieces of second column data being identical to each other in data type, to generate a piece of second teacher data; and use a model having undergone learning, the model being generated from the piece of second teacher data, to present candidates for associating each of the pieces of first column data and each of the pieces of second column data with each other.

The application relates to the technical field of natural language processing, and discloses a guided question and answer system and method based on implicit premise triple extraction, which comprises the following steps: receiving user question text and performing implicit premise triple extraction to construct a premise triple set; performing multi-layer conflict detection on the set, including internal mutual consistency verification and external domain knowledge collision detection, labeling internal mutual consistency states and external collision states for each premise triple, and generating a premise verification report by summarizing; generating a counter-question sentence according to the states by using a corresponding guided counter-question generation strategy, verifying a conflict premise correction state according to a user feedback reply text, executing a closed-loop correction according to the state, and outputting a final answer; and the application eliminates logical misdirection caused by a large language model blindly obeying input conditions through structured verification and closed-loop correction of a question premise, and ensures the correctness of a question and answer conclusion in a complex professional scenario.

This invention relates to the field of automated code generation technology, providing a code generation method and system based on data augmentation and hybrid similarity negative sampling. The method includes: generating multiple sets of candidate codes and test cases based on a variant question set using multiple different large language models; iteratively calculating the credibility score of the candidate codes and the quality score of the test cases; selecting the candidate code with the highest credibility score as the standard answer code; and determining the test cases that pass the standard answer code as valid test cases; generating multiple extended codes and verifying their correctness using valid test cases; selecting the code with the shortest execution time as a positive sample from the correct codes; and selecting erroneous codes whose hybrid similarity meets a preset threshold as negative samples; and fine-tuning and training the target code generation model before using the target code generation model to generate code. This method combines high accuracy with high execution efficiency.

The application provides a tool calling capability learning training method and device and a storage medium. The method comprises the following steps: constructing initial context information to be trained; inputting the initial context information into a language model agent to generate a response result; performing multi-dimensional quantitative evaluation and processing on a tool calling instruction to generate fine-grained reward information; training the initial context information based on the response result, an execution result and correctness reward information to obtain target context information; and iteratively training the language model agent according to trajectory data generated by historical interaction information, the fine-grained reward information and the target context information. The application constructs initial context information containing a user question, a tool set and historical interaction information, generates a response result containing a tool calling instruction and performs multi-dimensional quantitative evaluation, and iteratively trains the language model agent in combination with a fine-grained reward mechanism, thereby improving the tool calling reliability and adaptability of the language model agent.

The application discloses a model-oriented generation and variation method for a deep learning compiler, and comprises the following steps: step 1, constructing an ONNX operator rule library and performing initialization configuration; step 2, constructing a computation graph DAG based on the operator rule library; step 3, constructing a constraint model based on the generated DAG; and step 4, performing oriented generation and model variation based on the constraint model constructed in step 3. The application can improve the code coverage rate and defect discovery efficiency of the compiler under the premise of guaranteeing the semantic correctness and structural legality of the model.