20 results about "Generalization error" patented technology

The invention provides a cross-domain equipment fault diagnosis method and system based on large and small model cooperation, and relates to the technical field of equipment fault diagnosis. According to the method, the causal field generalization structure is introduced into the small model, explicit decomposition is carried out on the stable causal law and the field specific difference, and meanwhile, the causal field generalization structure is corrected by using the large model, so that the small model can automatically identify and retain the causal relationship which is universally applicable to each device and each field; therefore, the influence of inter-domain distribution difference is effectively eliminated. Theoretical analysis shows that the generalization error of the model mainly depends on the accuracy of the stable causal item, and the structure can minimize error drift caused by distribution drift. Therefore, the robustness of health state evaluation and fault prediction can be remarkably improved in a cross-domain scene, and the fault diagnosis model can still keep the prediction capability close to the training domain level under the condition of no target domain annotation data.

The application relates to the field of news recommendation technology and natural language query, and discloses a multi-domain and multi-behavior adaptive news recommendation system and method. Through the double-layer mechanism of the BIPN network, invalid behavior noise such as accurate filtering of false clicks and quick removal can be filtered, the effective browsing prediction accuracy can be improved, the recommendation deviation caused by noise can be avoided, the behavior noise suppression effect is remarkable, through the multi-domain mixed expert mechanism, the preference of a certain domain can be seamlessly transferred to other domains, the bad experience caused by the fragmentation of multi-end recommendation can be avoided, the cross-domain preference transfer capability is strong, through the GCN enhancement layer, high-order neighbor correlation is realized, preference information is supplemented for new users and small theme news, the situation that a hot news monopolizes a recommendation list is broken, the individualization coverage rate is improved, the AutoML double-layer optimization does not need manual parameter adjustment, can automatically adapt to the domain distribution and behavior distribution of different news platforms, compared with manual parameter adjustment, the model generalization error is reduced, and the training efficiency is improved.

The application discloses a machine learning-based retrospective clinical data management method. The method comprises the following steps: obtaining an electronic medical record log file and parsing unstructured text to generate a word vector; calculating the Hamming distance between the word vector and the knowledge graph standard entity vector to extract standardized test indicators; establishing a two-dimensional physiological correlation mask matrix according to the pathology dependent edge; truncating the time series data into a sequence feature matrix and performing linear transformation to generate a query matrix, a key matrix and a value matrix; calculating the dot product value of the query matrix and the transpose of the key matrix, adding the scaling factor and the mask matrix after exponentiation, and then multiplying the value matrix to obtain the context feature representation; and generating a dense feature tensor data through a feedforward neural network to predict the interpolation value. The application can effectively curb the overfitting of the time series model under sparse data, improve the physiological authenticity of the feature array, and reduce the generalization error of the downstream prediction model.

This invention discloses a generalization adjustment method for federated learning. First, the generalization variance is calculated on the local terminal. Then, training is performed on the local terminal for the r-th round to obtain the local model for the r-th round. The local model for the r-th round and the generalization variance are uploaded to the server. Finally, the global model parameters are aggregated on the server to obtain the global model for the (r+1)-th round. This method can minimize the variance of the generalization variance among the local terminals during federated learning training, thereby effectively reducing the generalization error of the global model across domains and improving the generalization problem in federated domains.

The application provides a pre-distortion method, system, device and storage medium. The method is applied to a pre-distortion system, the pre-distortion system comprises a pre-distortion multiplier, a complex neural network and a radio frequency power amplifier output feedback loop, and the method comprises the following steps: inputting a training complex signal into the pre-distortion system and outputting a corresponding complex scalar; training the pre-distortion system based on the training complex signal and the complex scalar until the generalization error vector amplitude and the generalization adjacent channel leakage ratio of the pre-distortion system reach a set requirement; and inputting a service complex signal into the trained pre-distortion system to obtain a pre-distortion corrected complex scalar. The above technical scheme effectively solves the problem of insufficient nonlinear representation capability of traditional pre-distortion models and power amplifier models, and solves the problem of poor pre-distortion correction effect caused by the lag of the dynamic change reaction of the power amplifier system due to the separate processing when the neural network is applied.

The application provides a process parameter screening method and device for an anti-penetration high-entropy alloy, and relates to the field of armored materials.The method comprises the following steps: obtaining performance indexes based on theoretical and target shooting experiment analysis of an alloy material target plate under normal penetration, and taking heat treatment process parameters of solid solution and aging as characteristic factors to design an orthogonal experiment to obtain a data set; performing standardization processing on characteristic values in the data set, and performing transformation on the characteristics; performing redundancy judgment and selection on time characteristics and temperature characteristics before and after the transformation; establishing a machine learning model, and performing generalization performance evaluation on the model through a root mean square error and a determination coefficient; and establishing a multi-objective optimization method based on a Pareto front and a multi-dimensional joint probability distribution function to realize multi-objective fusion and screening of process parameters.The data set in the application is completely obtained from experiments, which reduces generalization errors generated when a machine learning model is built for a small data set, and can accurately screen required process parameters in a short time.

Disclosed are a predistortion method and system, a device, and a non-transitory computer-readable storage medium. The predistortion method is applicable to a predistortion system which may include a predistortion multiplier, a complex neural network, and a radio frequency power amplifier output feedback circuit. The method may include: inputting a training complex vector to the predistortion system to obtain a complex scalar corresponding to the training complex vector, which is output by the predistortion system; training the predistortion system based on the training complex vector and the complex scalar until a generalization error vector magnitude and a generalization adjacent channel leakage ratio corresponding to the predistortion system meet set requirements; and inputting a service complex vector to the trained predistortion system to obtain a predistortion corrected complex scalar.

The invention discloses a federated learning-based stealth structure multi-target distributed optimization method, which comprises the following steps of: constructing a star federated learning architecture which comprises a central server and N distributed nodes; each node carries out model training based on local data and uploads a gradient to the central server; the central server aligns electromagnetic domain and mechanical domain feature distribution of each node through domain adversarial training based on the gradient of each node to realize cross-domain parameter optimization; performing weighted aggregation on each domain result to obtain a global gradient; global model parameters of the central server are updated by using the global gradient and then issued to each node to complete a round of iteration; and the central server periodically acquires the generalization error of the local model of each node, and determines the communication priority of each node according to the generalization error. According to the invention, the data volume is compressed to 10MB / time through gradient transmission, and the communication efficiency is greatly improved.

The invention belongs to the field of new energy power generation prediction and edge calculation application, and discloses a distributed photovoltaic field station short-term power generation prediction method based on federated learning, which comprises the steps of collecting and normalizing multi-source heterogeneous photovoltaic data, processing the normalized multi-source heterogeneous photovoltaic data through variational mode decomposition and double branches of an LSTM-GRU hybrid model, and predicting the short-term power generation of a distributed photovoltaic field station. And outputting a future power point prediction value and a double-confidence interval boundary value to reduce a prediction error. The LSTM-GRU hybrid model is trained and updated through combination of federated learning and a differential privacy technology, original data are not leaked, the system availability is high, and meanwhile, cross-station cooperative training is performed through federated learning to solve data isomerism and reduce generalization errors.

The invention discloses a deep belief network C concentration prediction method based on quantifiable generalization errors, and relates to the field of artificial intelligence driven environment monitoring, and the method comprises the steps: collecting C concentration and historical data of associated variables, carrying out the preprocessing, and constructing a training data set through the preprocessed C concentration and historical data of the associated variables; calculating the causal contribution strength of each correlation variable relative to the C concentration by using the training data set and a causal discovery algorithm, and generating a causal weight vector; constructing a deep belief network model, and performing supervised training on the deep belief network model by using the training data set to determine a reference deep belief network model; and monitoring the online monitoring data stream in real time, updating the generalization error upper bound by calculating the data distribution difference degree of the online monitoring data stream and the training data set, and generating a dynamic generalization error upper bound value. According to the method, the minimization of the upper bound value of the dynamic generalization error is taken as a target, and constraint optimization is carried out on the model parameters according to the causal weight vector.

This invention discloses an insurance product recommendation method and apparatus based on a heterogeneous ensemble algorithm. The method acquires historical behavioral data of insurance customers and segments customer groups, constructing classification labels based on the types of insurance customers have historically purchased. It employs a multi-scale time sliding window mechanism to extract dynamic temporal features of customer behavior, setting different weights based on the distance of the sliding window from the prediction window to generate dynamic temporal features. A two-layer Stacking model framework containing heterogeneous base learners is constructed. The first-layer model takes the original training data and dynamic temporal features as input, while the second-layer model fuses and trains the prediction results from multiple base learners in the first layer, outputting a prediction result of the customer's purchase intention. Target insurance products are sorted, generating a personalized recommendation list and pushing them according to the sorting priority. This invention can capture the dynamic temporal features of customer behavior, and combined with the heterogeneous ensemble Stacking framework, improves the accuracy and personalization level of insurance product recommendations while reducing model generalization error.

The invention discloses a user privacy regression modeling method fusing multi-platform data, and relates to the technical field of privacy modeling, and the method comprises the steps: employing a hierarchical attention mechanism to carry out the preprocessing of user behavior data, outputting a feature vector, and according to the attention weight generated by the hierarchical attention mechanism, obtaining a feature vector; and dynamically distributing privacy budget by adopting an exponential decay function, carrying out cross-platform balance through a global privacy budget pool to obtain a personalized privacy budget distribution scheme, and locally carrying out distributed regression training on each data source based on the feature vector and the personalized privacy budget distribution scheme to obtain a personalized privacy budget distribution scheme. After noise adaptive to the privacy budget is added to the gradient, a global regression model is formed through secure aggregation; according to the method, local generalization errors of all platforms are summarized into a global error signal, and the global error signal is reversely propagated to an attention layer and closed-loop optimization of an exponential attenuation function, so that real-time cooperative adjustment of the attention weight and the privacy noise intensity is realized.

This invention belongs to the field of shale gas reservoir development technology and proposes a method for predicting shale gas well production, including the following steps: Step 1, collecting geological data, engineering data, and production data of shale gas wells to generate basic data; Step 2, preprocessing the basic data to identify the main controlling factors affecting shale gas well production; Step 3, constructing a sample set; Step 4, establishing a shale gas production prediction model using an improved B-L-A algorithm; Step 5, using the shale gas production prediction model established in Step 4 to predict the production of the fracturing section of gas wells in the target block. This invention establishes a shale gas production prediction model that deeply integrates the mathematical advantages of BP neural networks and LSTM neural networks, and uses an Arps model as an engineering experience model for constraint, effectively reducing generalization error, avoiding overfitting, and improving model prediction accuracy, thus providing reliable production prediction for the large-scale development of shale gas.

Human knowledge may be injected in an explainable AI system in order to improve the model's generalization error, model accuracy, interpretability of the model, avoid or eliminate bias, while providing a path towards the integration of connectionist systems with symbolic and causal logic in a combined AI system. Human knowledge injection may be implemented by harnessing the white-box nature of explainable / interpretable models. In one exemplary embodiment, a user applies intuition to model-specific cases or exceptions. In another embodiment, an explainable model may be embedded in workflow systems which enable users to apply pre-hoc and post-hoc operations. A third exemplary embodiment implements human-assisted focusing. An exemplary embodiment also presents a method to train and refine explainable or interpretable models without losing the injected knowledge defined by humans when applying gradient descent techniques. The white-box nature of explainable models allows for precise source attribution and traceability of knowledge incorporated into the model.

The application relates to the field of artificial intelligence and big data processing technology, and discloses a user portrait construction method based on AI big data, which comprises the following steps: performing differential privacy injection preprocessing on original user behavior logs; constructing a dynamic generalization unit satisfying k anonymity and having minimum generalization error; mapping the unit to a low-dimensional embedding space to generate an anonymous sequence; performing high-fidelity sequence reconstruction by using a conditional generative adversarial network; separating long-term interest and short-term intention by multi-granularity time sequence decomposition, respectively aggregating features by using a double-channel attention mechanism, and finally fusing to generate a high-fidelity user portrait vector. The system comprises corresponding function modules. The application significantly improves the behavior fidelity of the user portrait and the recommendation recall rate under the premise of meeting the differential privacy and k anonymity legal standards.

The application discloses a user privacy regression modeling method fusing multi-platform data, relates to the technical field of privacy modeling, and comprises the following steps: pre-processing user behavior data by adopting a hierarchical attention mechanism, outputting a feature vector, dynamically allocating a privacy budget by adopting an exponential decay function according to attention weights generated by the hierarchical attention mechanism, balancing cross-platforms by a global privacy budget pool to obtain a personalized privacy budget allocation scheme, performing distributed regression training locally at each data source based on the feature vector and the personalized privacy budget allocation scheme, adding noise adaptive to the privacy budget to a gradient, and forming a global regression model by safe aggregation; and the application realizes real-time collaborative adjustment of attention weights and privacy noise intensity by collecting local generalization errors of each platform into a global error signal and reversely propagating the global error signal to a closed loop of an attention layer and an exponential decay function.

The application discloses a domain self-adaption target detection method based on target domain generalization estimation, relates to the field of computer vision, and particularly relates to a target detection method for unsupervised domain self-adaption in deep learning.The application uses an auxiliary module to perform inconsistency prediction, thereby avoiding huge computing cost caused by multiple training of the whole detection model.The generalization error on the target domain is estimated according to the inconsistency predicted by the auxiliary module, and then the target detection performance on the target domain is directly optimized, so that the method is more effective than an indirect optimization method through adversarial training;the classification and regression subtasks of the target detection are respectively optimized, so that the classification accuracy is better and the frame positioning is more accurate.The high-confidence main classifier prediction is screened as a pseudo label through a confidence module, and source domain knowledge is fully explored to help improve the cross-domain detection performance.

The invention provides a mesothelioma prediction method based on CGAN-SVDD-DBN-ELM-BP, and the method comprises the steps: 1, obtaining the detection data of a mesothelioma patient, and carrying out the data preprocessing, and the data preprocessing comprises the filling of a missing value, the deletion of an abnormal value, and the normalization processing; 2, generating more patient samples based on mesothelioma definite diagnosis patient data and mesothelioma related characteristics by using a conditional generative adversarial network, and solving the problem of imbalance between data; 3, performing anomaly detection on the mesothelioma data after data enhancement through a support vector data description method, establishing a minimum hyper-sphere including as many training samples as possible, and removing abnormal data samples outside the hyper-sphere; and finally, using mesothelioma data after abnormal data elimination to train a DBN-ELM-BP deep learning algorithm, the algorithm combining DBN unsupervised feature extraction, ELM fast learning speed and generalization ability, solving the problems of slow convergence speed and falling into local minimum caused by parameter random initialization, effectively reducing training errors and generalization errors, and improving the robustness of the algorithm. And the prediction performance is improved. And classifying the samples by using the trained classifier and outputting a mesothelioma prediction result.

The invention relates to an aerodynamic characteristic six-component modeling method based on a functional learning method, which belongs to the technical field of aircraft aerodynamic characteristic modeling, and comprises the following steps: firstly, establishing an initial data set, performing data identification and screening, and removing unreasonable data; performing mathematical transformation, and dividing the data into a learning set and a verification set; constructing an initial model according to the data in the learning set, and performing verification through a verification set; then evaluating the model; and finally, carrying out lightweight processing on the verified model, and finally outputting an aerodynamic coefficient display function model. Cross-wind-tunnel and cross-scale data training is fused, the model can break through limitation of a single wind tunnel flow parameter coverage range, six-component prediction under the conditions of a wide-domain Mach number and an attack angle is achieved, and generalization errors are reduced compared with a traditional empirical model. According to the method, the dependence on mass data is remarkably reduced through feature parameter dimensionality reduction and function space constraint on the basis of a physically guided symbol regression framework, and the efficiency is greatly improved by relying on analytic model characteristics.

The invention provides a material constitutive equation construction method based on a large language model, which can be applied to the technical field of artificial intelligence. The method comprises the following steps: acquiring a stress-strain experimental data set of a material; performing data analysis on the experimental data set and the current equation set based on a first large language model to generate a natural language strategy; inputting the natural language strategy into a second large language model to output a candidate equation set; performing multi-target Pareto optimization evaluation of fitting error, generalization error and equation complexity on the candidate equation set, and updating a Pareto optimal equation set through non-dominated sorting; feeding back the updated Pareto optimal equation set to the first large language model, and repeatedly executing strategy generation and candidate equation generation operations until a preset convergence condition is met; and outputting the Pareto optimal equation set to form the Pareto leading edge of the material constitutive equation. The invention further provides a material constitutive equation construction device based on the large language model.