370 results about "Reference modeling" patented technology

The invention relates to the field of chip verification, and provides a chip verification method, device and equipment, a medium and a chip. The method comprises the following steps: establishing a dynamic simulation verification environment of a to-be-verified design, and generating a plurality of test scenes and corresponding test excitation according to a design specification definition; sending the test excitation to a driver through a sequencer, converting the test excitation into a signal conforming to a to-be-verified design interface protocol by the driver, and transmitting the signal to the to-be-verified design; the monitor collects an input signal and an output signal of a to-be-verified design in real time, converts the input signal and the output signal into transactions, and transmits the transactions to the scoreboard; comparing the reference model with an output signal of the to-be-verified design to generate a dynamic simulation verification result; according to the dynamic simulation result, the state space is reduced, state space traversal is carried out on a design part which is not covered by dynamic simulation in the design to be verified, and a formal verification analysis result is generated; and evaluating the to-be-verified design according to the formal verification analysis result.

The invention discloses an intelligent event studying and judging method based on machine learning, which comprises the following steps of: extracting features from multi-source event data, constructing an event fragment set with a uniform structure according to a preset time window, identifying fragments with stable features in the event fragment set as normal samples, constructing a reference model through an MCD algorithm, and analyzing the normal samples. The method comprises the following steps: extracting feature distribution in a normal state, comparing a to-be-analyzed event with a reference model, calculating feature offset, forming a behavior trajectory, determining an anomaly judgment range by adopting an elliptical envelope algorithm, and finally evaluating an event risk level and dynamically adjusting reference sample composition according to a continuous anomaly condition to realize intelligent identification and risk judgment of the event. According to the invention, dynamic perception and intelligent identification of complex events can be realized.

The invention relates to a motor energy-saving control method for adaptive parameter adjustment. The method comprises the following steps: constructing an operation parameter sequence and comparing the operation parameter sequence with a preset reference model by collecting the rotating speed, the torque and the current signal of a motor in real time to generate a parameter deviation vector; based on a deviation threshold comparison result, carrying out noise processing on the overrun deviation to obtain a filtered sequence, and calculating a time delay degree in combination with historical parameters; calculating a compensation gain according to the time delay, fusing the compensation gain with the filtered deviation to generate a control instruction sequence, correcting the compensation gain through fuzzy control to obtain a stabilization parameter, and further generating a correction control instruction sequence; and finally, acquiring load data, fusing the load data with the correction instruction to obtain a system operation characteristic value, and generating an optimization state identifier when a judgment standard is met. By adopting the method, adaptive adjustment of the control parameters can be realized, the dynamic working condition is accurately responded, and the energy-saving efficiency of the motor under the variable-load working condition is improved.

The invention discloses a 3D art model intelligent generation system and method based on multi-source data fusion, and relates to the technical field of computer aided design and graph generation, and the system comprises a multi-source data input module which is used for receiving a conceptual design drawing, text description and a reference 3D model; the feature extraction module comprises an image feature extraction unit used for extracting a style feature vector and a structure contour vector of the conceptual design drawing; and the text semantic analysis unit is used for outputting a key attribute tag vector described by the text. According to the 3D art model intelligent generation system and method provided by the invention, through multi-source data input including the concept design drawing, the text description and the reference 3D model, the design intention is comprehensively fused, the limitation of a single data source generation model is solved, and the dynamic weight distribution strategy and the closed-loop optimization mechanism of the 3D art model intelligent generation system and method are high in practicability. The geometric shape and style characteristics of the generated model are highly consistent with the design intention, and the workload of manual later correction is reduced.

The invention provides a preference alignment optimization method based on reward-driven selective punishment, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining intelligent question and answer training data, and constructing an intelligent question and answer training sample set which comprises a plurality of intelligent question and answer training samples; taking a to-be-optimized large language model as a strategy model and a reference model; and training the strategy model by using the intelligent question and answer training sample set, and measuring the offset amplitude of the strategy model before and after training by using the reference model to obtain an optimized large language model. According to the method, implicit reward signals in the model are introduced, preference data are divided into multiple categories according to implicit reward distribution generated by the model, a dynamic weight function is designed, differential weighted optimization is carried out on different categories of samples, reinforcement learning of high-quality samples and suppression of low-quality samples are achieved, and the method has the advantages of being high in robustness and high in robustness. Weight of low-quality or conflict samples is reduced while high-quality sample learning is enhanced, noise interference is suppressed, and overfitting is prevented.

The invention provides a chip verification method and device, equipment, a medium and a program product, and the method comprises the steps: receiving a first operation instruction of a to-be-verified module and a second operation instruction of a reference model in parallel, the reference model being a function model used for simulating the expected behavior of the to-be-verified module; searching a first instruction record matched with the first operation instruction from a request queue of the reference model, and searching a second instruction record matched with the second operation instruction from a request queue of the to-be-verified module; and determining a verification result of the to-be-verified module according to a comparison result of the first operation instruction and the first instruction record and / or a comparison result of the second operation instruction and the second instruction record. According to the invention, a structured and reusable universal chip verification method is realized, so that a tedious customized scoreboard design is replaced, and the efficiency and accuracy of verification work are greatly improved.

The present disclosure provides a model training method and device, and a data processing method and device. The method of the present disclosure uses a reference model to screen an existing instruction dataset for an instruction whose model fitting difficulty meets a preset condition, and the method can screen the instruction dataset for a challenging instruction having high model fitting difficulty as a seed instruction. A plurality of similar instruction samples are generated by means of expansion on the basis of the seed instruction, thereby obtaining more challenging instruction samples, a training set comprising the instruction samples and reference responses for the instruction samples is constructed, and knowledge distillation can be achieved by using the reference model on the basis of the existing instruction dataset, thereby obtaining a training set containing higher-quality instruction data. Furthermore, using the training set to train a deep learning model enhances the capability of the deep learning model to handle more complex and challenging tasks, and improves the performance of a trained target model.

The invention discloses a power industry large model continuous pre-training method and system based on dynamic self-constraint, and belongs to the technical field of artificial intelligence, and the method comprises the steps: obtaining industry pre-training corpora and instruction training corpora, dynamically adjusting the mixing ratio of the industry pre-training corpora and the instruction training corpora through a curriculum-type strategy, and obtaining an industry pre-training corpora and an instruction training corpora; obtaining a dynamic mixed data set; configuring a reference model based on the dynamic mixed data set, and training a target power industry large model by adopting a differential loss function and the reference model for different types of data in the dynamic mixed data set; according to the differential loss function, self-adaptive KL divergence is calculated according to inter-partition optimization logic, and the self-adaptive KL divergence is adopted to construct a loss function; and obtaining the probability of the reference model through an online reasoning framework, and enabling the training to be continuously carried out based on the probability of the reference model. According to the method, the knowledge conflict problem in professional field training is effectively solved, and the generality of the model is kept while the professional property of the power field is improved.

The invention discloses a photovoltaic module construction quality identification method based on computer vision and three-dimensional modeling. The method comprises the following steps: S1, collecting an original image sequence of a photovoltaic construction site and carrying out image preprocessing; s2, outputting a semantic tag graph and a two-dimensional space boundary graph through an improved BiSeNet network; s3, using a mask constraint SIFT algorithm to extract cross-view key point features, generating a dense three-dimensional point cloud according to multi-view stereo reconstruction, and outputting a structure three-dimensional geometric model; s4, mapping the semantic tag graph and the two-dimensional space boundary to the structure three-dimensional geometric model to obtain a semantic three-dimensional model; s5, performing spatial registration on the semantic three-dimensional model and the reference model, and calculating spatial error parameters of the photovoltaic module; s6, performing compliance discrimination on the spatial error parameters, and outputting a quality discrimination result; and S7, generating a construction quality evaluation report. According to the invention, automatic identification and accurate evaluation of the photovoltaic construction quality are realized, and the detection efficiency and the discrimination accuracy are improved.

The invention discloses a deep neural network model freezing training optimization method based on tensor similarity. According to the method, a reference model generation module regularly generates a single-layer reference model to support real-time tensor similarity evaluation; the tensor similarity calculation module generates a normalized Gram matrix by using the activation output, and accurately evaluates the stability of the active layer; the freezing decision module can smooth instantaneous fluctuation in an evaluation process based on a moving average value of tensor similarity so as to make a steady freezing decision; the tensor I / O module caches the forward propagation result of the freezing layer, and repeated calculation is avoided. According to the method, the calculation amount of back propagation and forward propagation in the deep neural network training process can be effectively reduced, and the calculation resource utilization rate in the training process is improved. The method is remarkably superior to an existing freezing method in deep neural network model training tasks such as image classification, target detection and image segmentation, and the training stability and efficiency are improved while the final precision of the model is not sacrificed.

The invention relates to the field of function verification of digital integrated circuits, and discloses an atomic-scale computing chip verification method based on a UVM. The method comprises the following steps: firstly, building a UVM simulation verification platform comprising each core component; then, a software program is used for generating test excitation with physical significance, wherein the test excitation comprises information such as configuration, functional, pseudo potential and track files; then respectively inputting the test excitation into the established UVM verification platform and the software reference model; the UVM verification platform applies the test excitation to the to-be-tested design and outputs a calculation result; the reference model also correspondingly generates a reference result; and finally, comparing the output of the to-be-tested design with the output of the reference model, and verifying the correctness of the function of the atomic calculation chip through visual auxiliary analysis and quantitative error analysis. Through excitation generation based on physical constraints and an effective error evaluation method, the function verification problem of the atomic-scale computing chip can be solved.

The invention discloses a large model-based safety production hidden danger identification method, which comprises the following steps of: acquiring to-be-queried data, and identifying safety production hidden dangers by utilizing an identification model; wherein the recognition model carries out reinforcement learning through the following steps: carrying out multiple sampling on to-be-queried data by utilizing a reference model, and generating a response group; calculating an advantage value of each element in the response group; and optimizing the identification model based on each advantage value. According to the method, the accuracy, the compliance and the interpretability of object recognition and hidden danger analysis in a professional scene by the model can be effectively improved on the premise of not depending on large-scale labeled data.

The invention discloses a chip verification method and device based on simulation result comparison, equipment and a medium, and relates to the field of chip verification, and the method comprises the steps: obtaining a simulation result and an expected result generated by a to-be-tested chip and a reference model based on the same input excitation; the reference model is a pre-constructed model consistent with the function of the to-be-tested chip; partitioning the simulation result and the expected result, and performing Hash calculation based on each simulation data block and each expected data block obtained by partitioning to obtain Hash values corresponding to each simulation data block and each expected data block; and verifying the to-be-tested chip by comparing the consistency of the hash values of the simulation data blocks and the corresponding expected data blocks. According to the method, the partitioning technology and the Hash algorithm are combined, the Hash value of each simulation data block and the Hash value of the corresponding expected data block can be compared in parallel, the chip verification efficiency is improved, when the chip verification goes wrong, which data block goes wrong can be accurately positioned, and the chip verification accuracy is improved.

The invention discloses a direct preference optimization-based large model post-alignment training method and system. The method comprises the following steps of: (1) constructing a static reference model and a strategy model; (2) optimizing preference data by using an implicit reference model, and (3) adjusting a strategy model by optimizing an objective function so that the strategy model better fits the user preference. According to the method, the limitation of the existing DPO and SimPO methods is solved by introducing a new reference model expression, for example, the method depends on a suboptimal reference model or uses a fixed reward margin. By adjusting the balance between the strategy model and the reference model, personalized reference model setting for each pair of responses is realized, and meanwhile, a theoretical guarantee is provided and the KL divergence is effectively controlled. And the alignment performance and the winning rate of the model are remarkably improved, and the method is a robust LLM fine tuning method.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a generative machine learning model using multi-turn training examples that include sequences of example inputs and example outputs. In one aspect, a method comprises, at each of a sequence of training iterations: obtaining a plurality of example interactions, wherein each example interaction includes example model inputs and example model outputs for a plurality of time steps; obtaining one or more reference interactions for each example interaction, wherein each reference interaction includes reference model inputs and reference model outputs for a plurality of time steps; determining a preference measure for each example interaction based on a comparison between the example interaction and the reference interactions for the example interaction; and updating the target generative machine learning model to optimize an objective function that includes the preference measures for the plurality of example interactions.

The invention relates to a mechanical arm self-adaption compliance control method and system based on model reference self-adaption and impedance parameter online identification, and belongs to the field of robots. The method comprises the steps that a mechanical arm task space dynamics model is established; designing a reference model; a recursive least square method is adopted, and an equivalent stiffness matrix and a damping matrix of the interaction environment are identified online; designing a composite adaptive law; the expected rigidity of the reference model is dynamically adjusted according to the environment rigidity obtained through identification; and according to sensor data collected in real time, the mechanical arm task space dynamics model, the robot parameter self-adaption law, the environment rigidity, the environment damping and the environment parameter identification robust item, control force is generated, and the mechanical arm is driven through the control force. According to the method and the system, the environment characteristics can be sensed online, and the compliant behaviors can be adaptively adjusted, so that safe, efficient and high-robustness interaction of the mechanical arm in a complex and unknown environment is realized.

The invention discloses a federal forgetting method and system based on projection gradient rise, and belongs to the technical field of machine learning, and the method comprises the steps: obtaining information of N clients and a central server in a federal, and carrying out the T-round training of the federal through a federal weighted aggregation learning algorithm; performing data distillation on each client to obtain a distillation data set; acquiring a federal global model and a local model of the forgotten client in T-1 rounds, and solving a reference model on the forgotten client; the client optimizes parameters in an l2 norm ball with the radius of delta around the reference model; sending the federal global model to all clients, and performing opposite training strategies on the forgotten clients and the remaining clients; re-aggregating the local models of the forgotten client and the remaining clients; and obtaining a final federal forgetting model until E rounds of training are completed. The problems that a non-distributed machine forgetting method caused by distributed knowledge penetration is difficult to be used for federal learning and disastrous forgetting is caused by gradient rise of federal forgetting are solved.

The invention provides a text-to-structured query statement method based on a large language model and reinforcement learning, and relates to the technical field of information, the method comprises the following steps: integrating a natural language problem and a database mode into a unified prompt template to obtain a candidate structured query, and generating a reference structured query by a reference model; constructing a multi-dimensional reward framework, and performing weighted aggregation on multi-dimensional rewards to obtain corresponding final reward scores; calculating a KL penalty value between the output of the strategy model and the output of the reference model based on the candidate structured query and the reference structured query to obtain a constraint of stable strategy update; updating the strategy model parameters through back propagation iteration to obtain an updated strategy model; and analyzing the prompt template by using the updated strategy model to obtain a text-to-structured query statement processing result, and completing the process from the text to the structured query statement. The problems that existing structured query generation is low in accuracy and semantic consistency is difficult to guarantee are solved.

The invention relates to the technical field of precision spring manufacturing and automatic control, in particular to a spring forming online control system and method. The system comprises a dynamic modeling module, a disturbance observation module, an impedance decoupling module and a trajectory correction module. The system calculates disturbance torque by constructing a mechanical transmission dynamics reference model and collecting servo parameters; the core of the method is to decouple a material deformation impedance component representing transient rheological characteristics of a wire rod from a disturbance torque, generate a position correction instruction based on an elastic-plastic compensation model, and dynamically adjust a forming track in a current period; according to the invention, the servo motor is multiplexed as a virtual probe, the hysteresis of visual detection is eliminated, the strength fluctuation of the wire rod can be effectively handled, and the product consistency is obviously improved.

The invention discloses a power grid impedance identification method based on model reference self-adaption, and belongs to the technical field of electrical engineering. The method comprises the steps of establishing a reference model, and adopting integral control to iterate the reference model. A three-phase grid-connected current estimation value is obtained by establishing a reference model, the three-phase grid-connected current estimation value tracks the sampled three-phase grid-connected current through integral control, and power grid impedance parameters in the reference model are iterated. And when the three-phase grid-connected current estimation value is consistent with the sampled three-phase grid-connected current, the iterative power grid impedance parameter in the reference model is the final power grid impedance identification value. According to the method, active injection harmonic disturbance is not needed, a complex signal processing method is not needed, grid-connected current control is not affected on the basis, and the power grid impedance identification result is accurate.

The present disclosure generally relates to wireless communication systems, and more particularly, to an apparatus and method for integrated inference using dual-sided machine learning in wireless communication systems. A method of operating a user equipment (UE) in a wireless communication system includes: transmitting capability information of the UE to a network; receiving at least one of a structure or parameters of a reference model, or receiving a learning data set from the network according to the capability information of the UE; configuring a machine learning (ML) model directly on the UE or through a UE-side learning server based 10 on the received information; and performing integrated inference based on dual-sided machine learning models with the network using the configured machine learning model.

The invention relates to the technical field of knowledge graph construction, and discloses a small model and large model combined campus security event knowledge graph construction method, which comprises the following steps: acquiring campus security event text data, and defining a campus security event type and a keyword dictionary; feature coding fusion is carried out by adopting an LEBERT model; identifying semantic features of the campus security event text through a BiLSTM model; performing optimal tag sequence decoding through a CRF model, and extracting common entity texts of the campus security events; extracting an initial campus security event triple by adopting the reference model; performing evaluation by adopting a reward model, and calculating a comparison reward by adopting an offline response mean value function to guide the strategy model to perform reinforcement learning optimization to obtain a campus security event triple after reinforcement learning fine adjustment; and constructing a campus security event knowledge graph. According to the method, the identification precision of the model on common entities of the campus security event is improved, and the identification accuracy of the model on the relationship between the entities is improved.

The invention relates to the technical field of dynamic system obstacle avoidance, in particular to a distributed multi-agent safety control method based on a non-smooth graph barrier function. Comprising the following steps: defining a node set, and constructing a communication graph topology and a neighbor interaction relationship according to the node set; boundary conditions required for constructing a non-smooth graph barrier function are given, and the non-smooth graph barrier function is constructed according to the boundary conditions; formalizing a multi-agent system kinetic equation with uncertain parameters according to a non-smooth graph barrier function; according to the multi-agent system kinetic equation, designing a reference model and an adaptive law, and constructing a stable reference trajectory; constructing and improving a filter according to the reference trajectory; and inputting the motion data of the unmanned aerial vehicle group to a filter to obtain a motion control result of the unmanned aerial vehicle. According to the method, an absolute continuity modeling tool is introduced, description of non-smooth safety areas such as polygonal obstacles is directly supported, and the dependence of a traditional method on a differentiable boundary is broken through.

The invention discloses a new energy automobile tire burst prediction method and system, a vehicle and a storage medium. The new energy automobile tire burst prediction method comprises the steps that the real-time state of a tire, vehicle operation and scene feature data are obtained and preprocessed; combining the weight, torque and heat source characteristics of the new energy automobile to construct a dynamic reference model to identify abnormal signals; the abnormal signals and the dynamic interaction parameters are fused through a space-time diagram neural network, the real-time risk level is evaluated, and the comprehensive tire burst probability is predicted; and generating a risk traceability report and a personalized maintenance scheme. According to the invention, through multi-modal sensing and dynamic reference modeling, the limitation of traditional single parameter monitoring is broken through, and accurate identification of new energy specific risks such as sidewall hidden damage and asymmetric wear is realized; in combination with the multi-scale feature fusion capability of the space-time diagram neural network, the comprehensiveness and timeliness of risk assessment are improved, the rate of missing report and false report is reduced, and a reliable guarantee is provided for the safety of new energy automobile tires.

A monitoring entity for an AI / ML model used in a wireless communication system is configured to receive model data associated with the model and to execute an evaluation of the model data with respect to a reference model to obtain an evaluation result. The monitoring entity has a plurality of analyser modules, wherein each analyser module is configured for evaluating an associated set of parameters derived from the model data with respect to an evaluated property to determine an associated monitoring metric associated with the analyser module, the associated monitoring metric indicating a performance of the model with respect to the associated set of parameters. The monitoring entity is adapted to request, from a different network entity, additional information relating to the model and to use the additional information relating to the model for obtaining the plurality of associated monitoring metrics.

The invention belongs to the technical field of image processing, and particularly relates to an insulator size deviation analysis method and system based on point cloud data, and the method comprises the steps: obtaining an insulator point cloud, carrying out the slicing processing, and calculating the form distortion index of each slice; based on a gravity torque distribution model, the distortion index is normalized into a gravity impedance coefficient, and the gravity compliance confidence coefficient is determined according to the difference value of the impedance coefficients of the adjacent slices; and driving a standard CAD model skeleton to perform weighted deformation by using the confidence coefficient, generating a deformation reference model, and registering the deformation reference model with the point cloud to obtain the dimensional deviation. According to the method, natural falling and structural defects of the insulator caused by gravity can be distinguished, gravity bending interference is eliminated by constructing the self-adaptive reference model, and high-precision detection of abnormal conditions such as dirt accumulation, icing or damage of the insulator is achieved.

The invention discloses a chip verification method, device and equipment and a storage medium, and relates to the field of chip verification, and the method comprises the steps: constructing corresponding sub-reference models based on the unassigned input signals and output signals of each composition module in a to-be-tested chip, so as to obtain a target reference model corresponding to the to-be-tested chip; in a chip verification process, assigning the input excitation signal value to an input signal of a corresponding composition module in the chip to be tested so as to carry out signal processing, and obtaining a simulation signal value of each composition module; assigning the input excitation signal value to an input signal of a corresponding sub-reference model in the target reference model so as to carry out signal processing, and obtaining an expected signal value of each sub-reference model; each of the simulation signal value and the expected signal value comprises an input signal value and an output signal value; and comparing the simulation signal value of each composition module with the expected signal value of each sub-reference model to verify the to-be-tested chip, and accurately positioning a specific error module when the chip verification has an error.

The invention relates to a non-signalized intersection traffic decision-making method based on a large model enabling reinforcement learning decision-making framework, and belongs to the technical field of automatic driving. The method comprises the following steps: firstly, acquiring an environment map and a vehicle state, providing coarse-grained reference actions based on a navigation reference model of classical path planning and trajectory tracking, and then generating a reward function through a large model by adopting a chain reasoning structure based on designer-suggestor-optimizer; based on the reference action and the reward function, the vehicle decision-making agent learns a driving strategy based on a reinforcement learning framework; and finally, establishing a rule-based interaction field, and finely adjusting the driving strategy output by the reinforcement learning framework to obtain a final control strategy. The reward design of the invention gets rid of excessive dependence on expert knowledge, and the design of the reference action can significantly reduce the exploration dimension and accelerate strategy learning.

The invention relates to the technical field of digital advertising, and discloses a digital advertising method and system based on a virtual community, and the method comprises the steps: displaying the digital modeling of a target commodity through a commercial tenant virtual unit, enabling a customer virtual unit to observe the commodity in multiple dimensions, obtaining offline observation information from the customer virtual unit according to the transaction record of the target commodity, and enabling the customer virtual unit to obtain the offline observation information; and on the basis of the reconstructed commodity digital modeling, generating an actual measurement reference modeling set, carrying out parallel matching on the commodity digital modeling of different commercial tenants in the virtual community and the actual measurement reference modeling set, generating a personalized temporary community scene for a customer virtual unit, and according to registration information of customers, carrying out real-time registration on the virtual unit. According to the method, logistics transportation prediction is carried out on commodity information in a temporary community, personalized commodity prediction modeling is generated, customers are helped to make shopping decisions more accurately, the method enhances the personalization and accuracy of advertisements, the shopping experience and advertising efficiency of the customers are improved, and the problem that in the prior art, the effectiveness of the advertisements is low is solved.

The invention provides a large model fine tuning method and device, electronic equipment and a computer storage medium, and the method comprises the steps: generating a candidate answer sequence according to question information through employing a strategy model after the question information is received; afterwards, generating an evaluation value sequence by using the reference model, and generating a reward value sequence under different reward functions by using a reward model comprising the multi-dimensional reward functions; performing group advantage evaluation according to the reward value sequence under all reward functions to obtain an advantage value sequence; performing difference analysis on the dominant value sequence by using the evaluation value sequence to obtain a difference analysis result; and finally, carrying out fine adjustment on the strategy model based on a difference analysis result to obtain an optimized strategy model. By fusing a multi-dimensional reward mechanism, the accuracy, logicality and robustness of a large model in complex logic and multi-scene tasks are effectively improved.