25 results about "Model expansion" patented technology

The invention belongs to the field of system reliability design analysis, and particularly relates to a common cause failure parameter-based fault analysis method, which comprises the following steps: acquiring a fault tree model of an analysis object, and fusing the fault tree model with a common cause failure parameter model; constructing a fault tree model with a common cause failure parameter model; performing common cause failure analysis by using the common cause failure parameter model; expanding the constructed fault tree model with the common cause failure parameter model into an explicit expression common cause failure fault tree model by adopting a universal parameter model expansion method; solving the extended explicit expression common cause failure fault tree model to obtain a top event occurrence probability; according to the method, the implicit common cause failure parameter model is expanded into explicit expression, the automation of the reliability analysis process of the complex system considering the common cause failure influence is realized, the modeling difficulty is reduced, and the reliability analysis working efficiency is improved.

The invention discloses a category incremental target detection method and related equipment, and the method comprises the steps: firstly extracting the multi-scale features of an input image, coding a known category text label to generate an embedded feature, calculating an average feature, and then generating an unknown category pseudo text embedded feature; a cross-modal feature pyramid is formed through a cross-modal fusion network, so that a known target is predicted and an unknown target is positioned; self-supervised training is carried out on an unknown target to generate a pseudo label to finish category fine tuning; and finally, model expansion is realized through incremental updating of a low-rank matrix and a shared matrix. According to the scheme, through pseudo-text embedding features and cross-modal fusion, the model is endowed with the unknown target recognition capability, and the application limitation of the open world is broken; by means of self-supervised false label fine tuning and matrix increment updating, full-category retraining is avoided, computing resource consumption and training duration are reduced, interference on original category weights is prevented, and stability and reliability of a target detection system during category expansion are guaranteed.

The invention provides a large language model modal expansion method and device based on parameter fusion and decoupling, and relates to the technical field of large language models. The method comprises the following steps: performing fine adjustment on a pre-training language model to obtain a plurality of multi-modal large language models; task vector extraction is carried out on each multi-modal large language model; sparsification is carried out on the original task vector by adopting a sparsification strategy to obtain sparse vectors, and fusion is carried out on the sparse vectors to obtain a fusion task vector; constructing model parameters according to the fusion task vector; constructing a mode-exclusive binary mask for each multi-mode large language model according to the fusion task vector; and constructing a fusion model according to the model parameters and the binary mask. The invention provides a multi-modal language model extension method with non-training fusion, modal decoupling, performance retention and continuous extension capabilities, which is suitable for efficiently integrating a plurality of MLLMs, reconstructing an original model structure, coping with application scenes such as continuous integration of new tasks and the like.

Disclosed is a device and method for processing tasks through model-based offline learning. The device includes: a dataset input unit configured to receive an offline dataset for offline reinforcement learning; an initialization unit configured to initialize a world model and a model generation dataset for predicting a state transition and a reward without interacting with a real environment; a model rollout unit configured to expand the offline data of the offline dataset based on the world model to generate an imagined trajectory and generate imaginary data of the model generation dataset; and a learning update unit configured to perform critic update and actor update based on the offline data and the imaginary data.

The application discloses a self-evolution multi-agent event semantic understanding system and method based on a dynamic scalable model, which comprises a multi-agent cooperation framework modeling module, a demand understanding and task planning module, a capability verification module, a model dynamic expansion module, a task execution module and a completion module connected in sequence, wherein the multi-agent cooperation framework modeling module comprises a central control type multi-agent, a task execution type multi-agent and a model expansion type multi-agent. The method focuses on a news event semantic understanding scene, constructs a dynamic scalable system architecture through a multi-agent cooperation mechanism, introduces a capability verification link to optimize a task execution process, and realizes on-demand iteration of model capability by using an active learning framework, so that adaptive evolution of agent capability is realized, thereby solving the problems of model rigidity and poor adaptability of a traditional event semantic understanding system in a complex news event scene, and significantly improving the dynamic adaptability and accuracy of news event semantic understanding.

The invention provides a hierarchical text classification method and system based on a self-feedback neural network, and belongs to the field of hierarchical text classification. The method aims at solving the problem that when an existing hierarchical classification method processes multi-level labels, flattening deviation can be introduced in a prediction normal form for outputting all the labels at a time, and consequently the relation between upper and lower layers of labels is inconsistent. According to the method, hierarchical dependence modeling among labels is realized through a multi-stage encoder and a step-by-step neural network structure, and cyclic correction of prediction results is performed among different stages in combination with a multi-round self-feedback strategy, so that information of superior labels is effectively utilized when fine-grained labels are discriminated; the weighted multi-target loss ensures the robustness under the condition of label imbalance, and allows the contribution degrees of different rounds to be meticulously regulated and controlled; on the whole, the method can give consideration to the global consistency and the accuracy of deep-level fine granularity in a hierarchical text classification scene, and lays a good foundation for subsequent model expansion and performance improvement.

Embodiments of the present application provide a pre-training model expansion method and device, electronic equipment, a storage medium and a computer program product, wherein the pre-training model expansion method comprises: sequentially subjecting a first model to an expansion operation corresponding to each of a plurality of set dimensions to obtain a second model; wherein the first model represents a pre-training model after the Nth expansion, the second model represents a pre-training model after the N+1th expansion, and N is an integer greater than or equal to 0.

The invention discloses a self-evolution multi-agent event semantic understanding system and method based on a dynamic extensible model, and the system comprises a multi-agent cooperation framework construction module, a demand understanding and task planning module, a capability verification module, a model dynamic extension module, a task execution module, and a completion module which are connected in sequence. The multi-agent collaboration framework construction module comprises a central control type multi-agent, a task execution type multi-agent and a model expansion type multi-agent, the method focuses on a news event semantic understanding scene, a dynamic extensible system architecture is constructed through a multi-agent collaboration mechanism, a capability verification link is introduced to optimize a task execution process, and the task execution efficiency is improved. And on-demand iteration of the model capability is realized by using an active learning framework, and adaptive evolution of the agent capability is realized, so that the problems of model stiffness, poor adaptability and the like of a traditional event semantic understanding system in a complex news event scene are solved, and the dynamic adaptability and accuracy of news event semantic understanding are remarkably improved.

The present invention provides a method and device for expanding the modality of a large language model based on parameter fusion and decoupling, which relates to the technical field of large language models. The method includes: obtaining multiple multimodal large language models by fine-tuning a pre-trained language model; extracting task vectors for each multimodal large language model; sparsifying the original task vectors using a sparsification strategy to obtain sparse vectors, and fusing the sparse vectors to obtain a fused task vector; constructing model parameters according to the fused task vector; constructing a modality-specific binary mask for each multimodal large language model according to the fused task vector; and constructing a fused model according to the model parameters and the binary mask. The present invention proposes a method for expanding a multimodal language model with capabilities of fusion without training, modality decoupling, performance retention, and continuous expansion, which is applicable to application scenarios such as efficiently integrating multiple MLLMs, reconstructing the original model structure, and continuously integrating new tasks.

This disclosure relates to a model expansion method, apparatus, device, storage medium, and computer program product. The model expansion method includes: during model training using first training data, monitoring the data compression capacity of each of k feedforward modules; in response to the detection of a first feedforward module among the k feedforward modules, adding a second feedforward module to the decoding module containing the first feedforward module; based on the expanded feedforward module, continuing to train the model using second training data, and monitoring the data compression capacity of each of the k feedforward modules; repeating the above process until the number of feedforward modules in the decoding module is expanded to a target number. This disclosure improves the utilization rate of training data by dynamically setting the model parameters of each added feedforward module, saves training resources, improves the continuity of the model's pre-training process, and thus improves the stability of the model.

The invention provides a novel prompt engineering method based on twinborn prompt, and relates to the technical field of prompt engineering. According to the method, a high-quality data set is constructed through real query and collection, multi-round screening and professional labeling, and then model expansion and lyric filtering, so that the data accuracy is improved; then, LLMs are adopted to construct a hierarchical reasoning model, and hierarchical reasoning comprises the steps that in the first stage, the heavy query is guided through a professional view angle, and LLMs are helped to focus on core problems in the dual-carbon field; in the second stage, specific example optimization response is supplemented, accuracy and understandability are both considered, domain knowledge is deeply activated, and response quality is remarkably improved; finally, the final response is evaluated through multiple indexes, the incredibility degree of the LLMs response is quantified through word amazing degree distribution, the illusion risk recognition accuracy is improved, and the information reliability is improved.

This invention provides a multi-teacher joint distillation and plug-in model expansion method, belonging to the field of data processing technology. Specifically, it includes: constructing a decoupled student model; performing joint distillation training on the decoupled student model using multiple teacher models, applying feature decoupling constraints to the common features output by the shared backbone network and the task-specific features output by the task-specific adapter during training; in the inference phase, dynamically calling the corresponding task-specific adapter based on the task identifier using the trained decoupled student model, jointly processing the input data with the shared backbone network to obtain the inference result corresponding to the task identifier; and performing plug-in incremental expansion, whereby, in response to the access of a new teacher model to introduce a new task, the shared backbone network remains unchanged, and a new task-specific adapter corresponding to the new teacher model is added and trained. This invention improves flexibility and adaptability.

The present application relates to the field of image data processing technology and provides a method and apparatus for detecting the print quality of 3D-printed concrete. In this scheme, the material properties of freshly mixed concrete are used as simulation input parameters, computational fluid dynamics software is used to simulate the printing process of a 3D concrete model, and the theoretical three-dimensional characteristics of the concrete 3D model are calculated based on the simulation results of the 3D printing process. Actual printing is carried out to generate a 3D printed concrete part, and an in-situ three-dimensional scan is performed on the 3D printed part. The macro-deviation, fine-deviation, surface defect rate, theoretical undulation, average expansion rate, and model expansion rate of the 3D-printed concrete are calculated using the theoretical three-dimensional characteristics and the actual three-dimensional characteristics of the 3D printed concrete part. These indicators can comprehensively and quantitatively detect and evaluate the print quality of 3D-printed concrete from various aspects, ensuring that the evaluation results are comprehensive and accurate, and improving the accuracy of 3D-printed concrete print quality detection.

The application discloses an information extraction algorithm based on large model extended label enhancement. The application comprises the following steps: taking relation extraction as an example, first, using the text in the training data set and the corresponding head-tail entity pair to expand the predefined relation type; second, using the expanded relation to extract the head-tail entity pair from the text in the verification data set, if the extracted head-tail entity pair is consistent with the real entity pair, the corresponding expanded relation is retained; finally, integrating the retained expanded relation into the predefined relation set, using the large model to perform the relation extraction task on the test set, thereby enhancing the performance of the model on the relation extraction task. The application is suitable for information extraction tasks in the field of natural language processing using large models, by designing an information extraction algorithm based on large model extended label enhancement, it is proved that the large model has the phenomenon of "false association" when performing information extraction tasks, and further using the extended type label to improve the performance of the large model.

The invention provides an intelligent system fusing structure perception fuzzy division and modular neural regression modeling, belongs to the field of learning model construction, and constructs an intelligent learning system which is high in precision, explainable and efficient in calculation by fusing structure perception fuzzy division and hierarchical neural regression modeling. Wherein the structure sensing mechanism dynamically adapts to data local density distribution, and the naturalness and boundary adaptability of rule division are remarkably improved; the sparse superficial neural sub-network is used as a consequent model, and the interpretability is maintained through parameter constraint while the nonlinear mapping capability is reserved; the unified fusion framework supports heterogeneous model extension and dynamic weight optimization, so that the system still keeps low reasoning overhead in a complex scene. Finally, collaborative optimization of interpretability, precision and calculation efficiency is realized, and a deployable solution is provided for industrial control, real-time decision and other high-dimensional nonlinear tasks.

The application discloses a learning index model facing update distribution, comprising the following steps: (1) using an overhead model to construct a key into an RMI structure; (2) uniformly dividing the range of data by using the cumulative distribution function of data at each internal node; (3) constructing a monotonous gapped unary linear regression model at each leaf node; (4) learning the update distribution of data during the construction of the index; (5) point query and insertion; and (6) model expansion and splitting. The application uses a machine learning model to replace a traditional B-tree-based database index structure, simultaneously learns the update distribution of data, utilizes the characteristics of the update distribution of data and the advantages of the machine learning model, and greatly reduces the storage overhead of the index structure and the overhead during data insertion.

The invention discloses a multi-modal large model expansion method fusing a multi-head cross attention mechanism and a Q-former, and belongs to the field of work intelligence and machine learning. The method comprises the following steps: encoding text data and corresponding multi-modal data, and aligning an encoded multi-modal feature vector with a text feature vector to form a text-multi-modal data pair; performing multi-head cross attention calculation on the aligned text-multi-modal data pair; performing Q-former query mechanism calculation on each type of multi-modal data to obtain a result which has no strong correlation with the text data but retains a large amount of information of the multi-modal data; the result of the cross attention calculation and the result of the Q-former calculation are fused; and according to different inputs of the selected large text model, setting an input transformation layer to carry out transformation and then inputting the large text model. According to the multi-modal information processing method, the self-adaptive capability of the model is enhanced, so that the behavior of the connector can be dynamically adjusted according to different tasks and inputs, and more efficient multi-modal information processing is realized.

The application discloses a model instance adjustment method and an electronic device, relates to the technical field of artificial intelligence, and comprises the following steps: performing aggregate calculation on original performance index data by means of instantaneous value index items and incremental value index item classification calculation to obtain service performance index data of a model; setting a model expansion and contraction limit threshold value through an open-source expansion and contraction component; in response to the service performance index data reaching the model expansion and contraction limit threshold value, adjusting the service instance of the model; in response to a new valid request of the model service and the service performance index data value being greater than a first threshold value, expanding the model container service instance copy; and in response to the service continuously having no valid request within a preset time or the service performance index data value being less than a second threshold value, reducing the number of model container service instance copies to a preset minimum value to release display memory and computing resources. The application can optimize the performance and resource use of a large model inference service and make it easier for users to analyze and maintain the service.

This disclosure provide a method and a related apparatus. The method includes: determining, from a learnware library, a first reuse model matching a target data set, where the learnware library includes a plurality of learnwares, each of the learnwares includes one reuse model and one model specification, the first reuse model is a reuse model included in one learnware in the learnware library, the target data set includes a plurality of pieces of target data, and the plurality of pieces of target data are tabular data; performing shrinkage adjustment on the first reuse model based on the target data set, to obtain a second reuse model; performing model expansion based on the target data set and the second reuse model, to obtain a third reuse model; and integrating the second reuse model and the third reuse model, to obtain a target reuse model corresponding to the target data set.

This application discloses a model access method, apparatus, storage medium, and computer device, relating to the computer field. The method includes: receiving an access request through a RESTful interface; wherein the access request carries a user identifier, model name, business capability name, and business parameters; determining a target model in a model set based on the model name; wherein multiple models in the model set share the RESTful interface; using reflection to invoke a class object indicated by the business capability name in the target model; and returning the invocation result obtained by the class object in response to the business parameters to the user indicated by the user identifier. This application can save time in model development and expansion, avoid adding a large number of interfaces during model expansion, improve development efficiency, and ensure code reuse.

The embodiment of the application provides a model expansion method of a face access control device, a computing device and a storage medium. In the embodiment of the application, a file of a to-be-downloaded model is generated by a server device, and a description feature for describing input data of the to-be-downloaded model is stored. The to-be-downloaded model refers to a model for updating the face access control device. The file and the description feature are acquired by the face access control device from the server device. According to the file, the corresponding model is loaded on the face access control device to update the model. Therefore, the face access control device can dynamically update the model through the file of the model of the server device, so that different application scenarios can be supported, and the capability of the face access control device, such as accuracy, is improved, thereby improving the experience of the user.

This application discloses a method for extending a model, electronic equipment, and a storage medium, which belong to the art of computers. The method includes the steps of: setting an index field in a TR181 extended model file for target traffic based on the type of target traffic; setting at least one description field in the TR181 extended model file for target traffic based on the description parameters of the target traffic, wherein one of the description fields is for expressing the characteristics of the description parameters; and setting the length of each of the description fields based on the globalization adaptability and global standards of the target traffic, thereby realizing the extension of the TR181 extended model file.

The invention discloses a flexible robot motion control fault diagnosis method and system, and the method comprises the following steps: carrying out the single-section kinetic analysis of a flexible robot according to main observation data and a model simplification direction, and building a single-section kinetic model of the flexible robot; carrying out model expansion on the basis of the single-section kinetic model to obtain an overall kinematics model of the flexible robot, and converting the overall kinematics model into a standard form; determining a robot fault diagnosis strategy based on test statistic comparison based on the analysis strategy; judging whether the flexible robot has a fault or not based on the overall kinematics model and a robot fault diagnosis strategy; according to the method, a hypothesis testing idea can be adopted, and a fault detection logic with statistical significance is constructed by comparing statistical characteristics of real-time measurement data with a reference value in a fault-free state; the detection precision and the false alarm probability are balanced by introducing a threshold value, and the health state of the robot can be accurately judged.

The application discloses a kind of electromagnetic log auxiliary ship integrated navigation method, first according to the influence of ocean current and measurement noise time-varying to electromagnetic log speed measurement, establishes a fixed model set, includes two sub-models, and is adaptively generated a model set by the method of expected model expansion;Secondly, the coupling term of state vector and measurement noise covariance matrix is decoupled using variational bayes based on mean field theory, and a new adaptive filtering formula is derived;Again the estimation fusion of the interactive multiple model (IMM) adaptive filtering result of fixed model set and expected model set is carried out respectively;Finally, the fused estimation result is used to correct the attitude, velocity and position calculated by the strapdown inertial navigation system.The application can eliminate the influence of ocean current and measurement noise time-varying, effectively improve the positioning accuracy of ship integrated navigation.