115 results about "Uncertainty estimation" patented technology

The invention discloses a welded pipe conveying abnormity prediction method and system based on large model reasoning, and aims to solve the problems that multi-source data is difficult to align, cross-station false correlation is caused, prediction lacks executable positioning and time sequence, and linkage control reliability is insufficient. Event alignment is carried out by taking a controller edge signal and an encoder zero position as time anchor points, a production line topology semantic graph containing time delay, capacity and interlocking attributes is constructed, and topology reachability and physical time delay constraints are applied in a self-attention long sequence model to carry out multi-step rolling prediction. And outputting a risk probability, refining the risk probability to spatial positioning of a roller way section or a shaft and the minimum executable intervention time, and generating a risk interval in combination with uncertainty estimation and calibration so as to drive an upstream beat self-adaptive speed reduction, shunting or stopping strategy. The technical effects of improving accuracy and interpretability, reducing false alarm and missing alarm, ensuring that linkage can be executed in advance and meeting edge time delay budget are achieved.

The invention relates to the technical field of agent path planning, in particular to a multi-agent collision-free path planning method based on a fusion DQN algorithm. The multi-agent collision-free path planning method comprises the following steps: firstly, constructing a two-dimensional grid map as an environment, and carrying out feature extraction by utilizing a CNN (Convolutional Neural Network); then, behavior clone learning is carried out through the expert model to obtain a BC model; the core innovation lies in that a BC model and a CNNDQN model are fused, an adaptive strategy learning framework is constructed, and intelligent dynamic combination of expert experience and reinforcement learning exploration is realized by adopting uncertainty estimation, antagonistic knowledge distillation and performance perception sampling technologies; and finally, further processing an initial path output by the fusion model by a CBS algorithm, and completing multi-agent collision-free path planning. According to the method, the accuracy and efficiency of path planning are optimized through a mixed learning strategy.

The invention relates to an intelligent measurement and control optimization method and device for dynamic parameter adaptive calibration, equipment and a medium. The method comprises the following steps: executing invariance causal test based on an external environment context to obtain a cross-environment invariant explanatory variable subset, and performing anti-fact simulation on each parameter in the cross-environment invariant explanatory variable subset to generate a causal attribution report; based on an under-excitation parameter subset and a cross-environment invariant explanatory variable subset in the baseline recognizable atlas, performing safe active excitation planning on the parameters to obtain a safe micro-perturbation excitation plan; and based on the residual error, the uncertainty estimation, the delay cross-correlation feature, the hysteresis loop area feature, the causal contribution score and the excitation-response fragment, performing calibration estimation on the target parameter by using a hierarchical estimator, and generating a calibration packet according to a calibration estimation result. By adopting the method, self-adaptive calibration of intelligent measurement and control dynamic parameters can be realized through residual attribution and cross-environment invariance test in combination with safe perturbation excitation.

The invention provides an AI-driven equipment health state assessment method and system, and relates to the technical field of intelligent operation and maintenance. The method comprises the following steps: acquiring equipment operation data, and performing time and dimension unification and quality control to form a multi-source operation data set and an environment context; generating an initial state feature based on the mechanism feature library, and obtaining a general representation through self-supervised pre-training; executing calibration learning by using a preset health label, and establishing a fusion evaluation model containing time sequence consistency and physical boundary constraint; carrying out distribution alignment and uncertainty estimation on the basis of scene differences to obtain alignment characterization and credibility scores so as to optimize a model gating strategy; performing joint mapping on the new data, outputting health index, fault probability and residual life estimation, and generating a root cause clue; lightweight online updating is executed under drifting detection, health indexes and root cause clues are written back to a mechanism feature library, early warning levels and maintenance suggestions are generated, and therefore complete-cycle intelligent sensing and self-adaptive optimization of the equipment state are achieved.

In various examples, systems and methods for uncertainty estimation for object detection in autonomous and semi-autonomous systems and applications are provided. The systems and methods may use data from one or more sensors (e.g., camera(s) and / or LiDAR sensor(s) to generate a representation of features surrounding a machine. A model may be used to generate probabilities of objects being present in the representation of features and uncertainty estimates corresponding to the object presence probabilities. The uncertainty estimates may be used to identify scenes that are significantly different from the training data, detect errors in the bounding shapes for objects, and / or highlight areas where object detections may have been missed. The systems and methods may also be used to auto-label scenes associated with the representation of features, and the auto-labeled scenes may be used for training purposes.

A short-term scheduling rule extraction method for a water-optical complementary system comprises the following steps: correcting an existing business prediction result by adopting a probability prediction model based on historical observation data and business prediction data of runoff and optical power; generating multiple groups of future possible runoff and optical power scenes by adopting a data sampling method based on a commercial prediction result corrected by the probability prediction model in the step 1; constructing a water-light complementary peak regulation optimization scheduling model; 3, constructing a Bayesian gated loop network which can consider the commercial prediction result and the model parameter uncertainty, introducing a probability recalibration mechanism which can effectively relieve the uncertainty estimation deviation of the Bayesian gated loop network, taking the optimal scheduling solution set obtained in the step 3 as the basis, extracting a complementary system scheduling rule which considers the multiple uncertainties, and finally obtaining the optimal scheduling solution set. And the peak regulation scheduling of the water-light complementary system is guided by using a complementary system scheduling rule. The comprehensive benefits of the multi-energy complementary system can be improved.

The invention discloses a medical image analysis method and system based on a visual language model, and belongs to the technical field of medical image intelligent diagnosis, and the system comprises an image preprocessing unit which carries out the down-sampling of an original retina OCT image to 256 * 256 and carries out the normalization of the original retina OCT image; the feature encoding unit comprises an image encoder based on RET Found in combination with LoRA optimization and a text encoder based on BioClinicalBERT; the class balance comparison learning unit is used for adjusting loss through class balance coefficients so as to relieve the class imbalance problem; the uncertainty estimation unit is used for calculating confidence quality and uncertainty scores based on Dirichlet distribution, and determining a threshold value in combination with an improved Youden index; and the model training unit adopts a total loss function of class balance loss and uncertainty loss, outputs a diagnosis result and an uncertainty score through transfer learning, and further comprises an image input module, a result display module and a data storage module. Rare disease classification performance and reliability are improved, training efficiency is improved through LoRA optimization, and an accurate and reliable scheme is provided for detection of the rare retina diseases.

The invention provides a dynamic sparse observation-oriented deep neural process ocean data assimilation method, and relates to the field of ocean data processing, and the method specifically comprises the following steps: constructing a training data set; simulating actually observed non-uniform and uncertain characteristics through Gaussian nuclear diffusion; building an ocean assimilation network oriented to sparse dynamic observation, outputting an analysis field and estimating uncertainty; and performing end-to-end training on the ocean assimilation network model by taking the reanalysis true value field as a supervision signal, and optimizing network parameters by combining a minimum error term and a structure constraint term. And after training is completed, inputting the background field in the test stage and sparse observation into the ocean assimilation network model for reasoning to obtain an ocean state reconstruction field conforming to the actual physical quantity scale. According to the technical scheme, the problem that in the prior art, calculation feasibility, cross-scale correlation modeling and credible uncertainty output cannot be considered under the real conditions of sparse observation and dynamic change of spatial-temporal distribution is solved.

The invention belongs to the cross technical field of artificial intelligence and decision support systems, and discloses an intelligent decision support method based on a dynamic knowledge graph and multi-modal fusion, which comprises the steps of 1, acquiring and preprocessing multi-source heterogeneous original data, 2, constructing and updating the dynamic knowledge graph, carrying out multi-modal fusion, outputting multi-modal fusion features, and 3, carrying out multi-modal fusion on the multi-modal fusion features. The method comprises the following steps of: 1, obtaining a multi-modal fusion feature and a knowledge graph embedding matrix, 2, receiving the multi-modal fusion feature and the knowledge graph embedding matrix, and outputting a candidate decision path set, decision probability distribution and a state value, 4, outputting dual uncertainty of each path by an uncertainty estimation module, 5, calculating a comprehensive reward of each path, 6, carrying out meta reinforcement learning and strategy optimization, and 7, carrying out multi-modal fusion. And step 7, outputting a final decision result, an uncertainty evaluation report and an interpretable reasoning path, and completing the decision. According to the method, the defects in the prior art are effectively overcome, and more accurate, reliable and explainable decision support is provided for a complex scene.

The invention discloses a joint blind denoising method and system based on self-heuristic learning and Bayesian reasoning, belongs to the field of computational imaging, and solves the problems that in the prior art, the mixed noise modeling capability is insufficient, the performance is degraded under the condition of low signal-to-noise ratio, the combination of uncertainty quantization and regularization is lacked, and the generalization capability is limited due to data dependence. Comprising the following steps: collecting an original image and preprocessing; generating a noise data pair; an enhanced residual attention U-Net model is constructed; a noise estimation sub-network is adopted to extract noise features, the noise features are fused with original image features, and the model is trained; adopting the trained model to carry out multiple times of forward propagation on the same input image to obtain multiple groups of denoising results; calculating a mean value and a standard deviation to obtain a de-noising prediction and uncertainty heat map; and training the trained model again based on the uncertainty heat map, optimizing network parameters, and obtaining a final denoising prediction result and uncertainty estimation thereof. The method is suitable for complex noise distribution processing scenes.

The invention relates to a model based on multi-modal cross attention and uncertainty integral gradient and application, and relates to the technical field of bioinformatics. The model is obtained based on multi-modal cross attention and uncertainty integral gradient construction, the problems that in an existing drug sensitivity prediction method, multi-modal data fusion is difficult, the model generalization ability is weak, and interpretability is lacked can be solved, the model captures correlation between modals through a cross attention mechanism, and the prediction accuracy of the drug sensitivity is improved. According to the method, uncertainty estimation is realized by combining Monte Carlo Dropout (MC Dropout), feature importance is analyzed by using an integral gradient algorithm, and finally, unification of high-precision prediction and biological interpretability is realized.

The invention discloses a personalized exercise prescription generation method fusing deep learning, learnable similarity and multi-modal time sequence analysis. The method comprises the steps that firstly, static signs and health risk labels of a user and dynamic signals such as the heart rate, the HRV and the step number collected by wearable equipment in real time are integrated, and unified modeling is conducted through a multi-modal encoder to form individual state vectors; then outputting a preliminary motion scheme by a BERT-Transform text generation model, mapping a user to a hidden space taking a historical execution effect as supervision by using a comparative learning network, and accurately inferring an FITT core parameter through approximate nearest neighbor retrieval and effect weighting; performing consistency optimization on texts and parameters through a large language model, completing risk verification under a safety barrier constructed by a medical rule base, ensuring that output meets clinical taboo and individual tolerance, and finally continuously optimizing long-term health income under medical constraints in combination with an offline reinforcement learning strategy, and cooperating with meta-learning cold start and uncertainty estimation, so as to achieve the purpose of improving the safety of the health management system. Safe, accurate, self-adaptive and explainable personalized exercise prescription generation for new and old users is realized, and the method can be widely applied to intelligent fitness, chronic disease exercise intervention and digital health management scenes.

The invention relates to an air conditioner load prediction method based on a self-adaptive double-flow graph attention network, and belongs to the technical field of building energy conservation and intelligent control. The method comprises the following steps: collecting historical power and environmental data of an air conditioner, and after preprocessing and normalization, constructing an input sequence through a sliding window and dividing a data set according to time; a prediction model is constructed, and a causal graph learning module, a multi-scale graph structure learning module, a self-adaptive space-time attention module, an uncertainty quantization module and a self-adaptive sampling module are integrated; a training set and a joint loss function training model are adopted, and a load prediction result and uncertainty estimation are output through Monte Carlo Dropout during testing. According to the method, the dynamic causal relationship between variables and multi-scale space-time dependence can be adaptively learned, reliable uncertainty quantification is provided while the prediction precision is improved, and the method is suitable for intelligent regulation and control and energy efficiency optimization of the air conditioning system.

The invention discloses a high-speed aerodynamic derivative analysis modeling method and system based on symbolic regression, and belongs to the technical field of aircraft aerodynamic modeling and artificial intelligence cross, and the method comprises the steps: obtaining an aerodynamic data set of a high-speed aircraft, and carrying out the preprocessing; defining grammar for generating candidate symbol expressions and constructing an expression tree; modeling based on the training set by adopting a layered symbol regression framework to obtain an analytical model; the prediction precision of the analytical model is verified on the test set, and physical consistency analysis is carried out; carrying out uncertainty quantification on the analytical model to obtain predicted uncertainty estimation of the analytical model; and finally, outputting an analytic model in a mathematical expression form for describing the relationship between the aerodynamic derivative and the flight state variable and uncertainty estimation of the analytic model. The method overcomes the defects of a traditional method in precision, generalization and interpretability, realizes full-automatic efficient modeling from data to the analytic model, and improves the modeling efficiency. The method is suitable for control design, stability analysis and real-time simulation of the high-speed aircraft.

Certain aspects of the present disclosure provide techniques for uncertainty estimation, such as for deep learning (DL)-based object tracking systems. A method generally includes processing, with a deep learning network, an input state associated with an object to predict an output state for the object, wherein the output state comprises a plurality of state estimates for a first time period; and generating a state covariance based at least in part on estimated covariance between at least two state estimates of the plurality of state estimates, wherein the state covariance represents an estimated uncertainty associated with the output state.

The invention provides a space field domain magnetic field reconstruction method based on rapid conversion of a magnetic field, and relates to the technical field of electromagnetic field measurement and inversion calculation. According to the method, observation data are collected and calibrated under the unified time and space reference, a prior physical model is established in combination with geometry and materials, an energy conversion operator is constructed and set, data, energy and physical constraints are integrated in a unified objective function, and solving is achieved through a multi-layer grid and an alternating updating strategy. And correction is completed through residual error verification and a back-off mechanism, and finally spatial magnetic field distribution and uncertainty estimation are output. The method still has fast and steady reconstruction capability under sparse sampling and noise conditions.

A computer-implemented method utilizing a continuous discrete recurrent Kalman network, wherein the method includes receiving, at an encoder, an input from one or more sensors, wherein the input includes one or more time series data associating data at one or more points in time; outputting, to a Kalman filter, a latent observation and uncertainty estimate in response to the input at the encoder; determining a latent state prior and latent state posterior utilizing the Kalman filter; and outputting, via a decoder, a filtered observation utilizing at least the latent state posterior.

The invention discloses a multi-agent workflow automatic optimization generation system and method based on prediction driving, and relates to the technical field of artificial intelligence, in particular to an automatic workflow generation method based on a data processing flow and an algorithm model. Through a combination mechanism of lightweight quality prediction, uncertainty estimation and candidate selection based on an upper confidence boundary, efficient optimization search of the multi-agent workflow is realized. Different from an existing AFlow method which completely depends on real evaluation, the method introduces a recursive least square method quality prediction model and a residual exponential smoothing uncertainty estimation mechanism into automatic workflow generation for the first time, so that the evaluation cost is remarkably reduced, and the convergence speed is increased.

The application discloses a digital-analog fusion double-drive digital-real fusion evaluation method for aerospace equipment, and belongs to the technical field of intelligent manufacturing. The method comprises the following steps: generating enhanced product data covering extreme working conditions through a conditional generative adversarial network, and constructing an environmental parameter matrix to realize systematic representation of assembly working conditions; secondly, a dynamic data scheduler and a hierarchical reward mechanism are designed to realize intelligent fusion of real product data and enhanced product data in the reinforcement learning training process; finally, a bidirectional collaborative training framework is constructed, the feature extraction capability of the graph neural network is combined with the decision-making capability of the reinforcement learning, and the model generalization capability is improved through the meta-course learning of environment perception. In the online evaluation stage, a hybrid prediction mechanism based on uncertainty estimation is adopted to ensure the prediction reliability of the system under different working conditions. The application significantly improves the quality prediction accuracy under the condition of small sample, and is especially suitable for the fields such as aerospace and precision instruments which have extremely high requirements for assembly precision.

This invention relates to the fields of information security and artificial intelligence technology, specifically to a privacy-preserving open-set face recognition method based on LoRa and uncertainty estimation, comprising the following steps: generating a hash code set; pre-training a basic feature encoder to extract facial features; mapping facial features to binary hash codes in the hash code set, defining them as a target encryption key; and constructing a key space; selecting a query sample; fine-tuning the basic feature encoder multiple times to a personalized feature encoder; converging the facial features of the query sample under different conditions to the mapped target encryption key; during fine-tuning of the basic feature encoder, jointly learning evidence values ​​about the output with the total loss; in the authentication stage, calculating uncertainty; if the uncertainty value is greater than an uncertainty threshold, rejecting the access request; if the uncertainty value is not greater than a pre-uncertainty threshold, performing a bit-by-bit matching between the predicted key and the encryption key in the database.

One embodiment of the present invention provides a marker pose uncertainty estimation method. The method comprises the steps of: receiving a single image that includes a reference marker, and deriving a corner position for the reference marker on the basis of a pixel intensity variation of the reference marker; deriving corner position uncertainty for the corner position on the basis of the pixel intensity variation, and deriving the position and the pose of the reference marker on the basis of the corner position; and deriving the position and the pose uncertainty of the reference marker on the basis of the corner position and the corner position uncertainty.

The invention relates to the technical field of electrical equipment defect detection, and discloses an electrical equipment infrared defect image expansion method, which comprises the following steps: carrying out compressed sensing on an infrared image of target equipment based on a compressed sensing network model to obtain compressed low-dimensional data of the infrared image; establishing a generative model based on a variational auto-encoder VAE and a generative adversarial network GAN, and introducing a loss function of uncertainty estimation UGAN to the generative model; performing loop iteration alternate training on the uncertainty estimation UGAN and the variational auto-encoder VAE until new feature image data generated based on the compressed low-dimensional data through the generation model reaches a target value; and performing diverse new defect sample reconstruction based on the new feature image data through the trained generation model.

The application discloses a kind of based on mixed domain enhancement's thyroid nodule semi-supervised segmentation method and device, it is related to image segmentation technical field.The method includes: constructing semi-supervised segmentation model, including unified bidirectional copy and paste enhancement module, cross-teaching model, patch enhancement module based on information entropy and frequency domain enhancement module based on Fourier transform;Using labeled data and unlabeled data to train the model constructed, by unified bidirectional copy and paste, generate intermediate sample between labeled data and unlabeled data in space domain;Using the uncertainty estimation based on information entropy, exchange low-entropy patch in labeled data and high-entropy patch in unlabeled data;The low-frequency component of labeled data and unlabeled data is exchanged by fast Fourier transform, and the low-frequency prediction is supervised in consistency based on the pseudo-label generated based on full frequency;The thyroid nodule ultrasound image to be segmented is input into the trained segmentation model, and the nodule segmentation result is obtained.The present application can improve segmentation accuracy.

The invention discloses a photosynthesis leaf chamber carbon dioxide steady state concentration prediction method and system, and the method comprises the steps: collecting the dynamic time sequence data of the leaf chamber carbon dioxide concentration, the dynamic time sequence data comprises the sequence of the descending segment and the ascending segment of the carbon dioxide concentration and the environmental data, and carrying out the preprocessing; processing the data by using a pre-trained double-branch time sequence coding network model to obtain a steady-state concentration predicted value and prediction uncertainty, and comparing the steady-state concentration predicted value and the prediction uncertainty with a preset uncertainty threshold; and judging prediction reliability according to a comparison result, outputting a predicted value if the prediction reliability is reliable, and continuing to collect data and re-predict if the prediction reliability is unreliable. The system comprises a data acquisition and preprocessing module, a prediction and comparison module and a judgment and control module. Through the double-branch time sequence coding network and the uncertainty estimation mechanism, the steady-state value can be predicted in advance in the dynamic change process of the carbon dioxide concentration, the measurement waiting time is remarkably shortened, the measurement efficiency is improved, and meanwhile the prediction precision and reliability are guaranteed.

The present application relates to a multi-modal data inconsistency detection and cleaning method and system based on adaptive energy guidance, belonging to the field of artificial intelligence and data mining technology. The present application proposes a multi-modal data inconsistency detection method based on adaptive energy guidance. First, a multi-expert uncertainty estimation framework is constructed to effectively mine multi-dimensional modal semantic correlation information such as prediction uncertainty, cross-modal consistency and teacher-student difference. Second, according to the multi-expert evaluation signal, an adaptive energy scoring mechanism is designed to effectively aggregate and quantify the sample modal consistency by constructing a comprehensive energy function and a weighted function based on Boltzmann distribution. Finally, a dynamic data cleaning and screening strategy based on energy score is constructed to improve the quality of the data set and the robustness of the downstream model, and better achieve the purpose of data service.

A method and system for uncertainty quantification approach for federated learning that enables the distinction between aleatoric and epistemic uncertainties, as well as between local and global in-and out-of-distribution data. The method and system offer permit selecting the appropriate model to predict on a given input based on these uncertainty estimations. This comprehensive framework contributes to enhancing the robustness and reliability of federated learning models in real-world applications, effectively addressing the challenges that arise due to the heterogeneity and diverse nature of data distributions.

The invention discloses a dual-group neutron diffusion equation solving method based on hybrid experts and EMA-NLL. The method comprises the following steps: determining a to-be-solved two-group neutron diffusion equation set; constructing a data set; constructing an equation solving neural network, wherein the equation solving neural network comprises a hybrid expert network, a fast group prediction network, a hot group prediction network and an uncertainty estimation network; constructing a total loss function Ltotal; and training and solving the neural network by using the data set and minimizing Ltotal to obtain an equation solving model and using the equation solving model for prediction. According to the method, multiple key improvements are introduced on a neural network structure and a physical modeling strategy, so that the expression ability and the solving precision of the model in a multi-region, non-uniform physical property and remarkable coupling effect scene are improved; and fine modeling of the problems of strong regional difference and severe flux change can be realized, the adaptability to error distribution of different regions can be enhanced, and the sensitivity and training efficiency of the model to a high-gradient region can be improved.

The invention discloses an environmental water quality monitoring and predicting method based on deep learning, and relates to the technical field of environmental monitoring. The method comprises the steps of collecting multi-site time series data and constructing a space-time input matrix and a space relation matrix; a time sequence feature is extracted by using 1D convolution and an improved Mama network; spatial features are extracted through multi-head local attention with distance punishment; carrying out multi-index multi-step prediction by adopting a multi-task learning framework after fusion; a multi-objective loss function is constructed, and a GRPO optimization strategy is adopted for training; during reasoning, uncertainty estimation is carried out through Monte Carlo Dropout, and a prediction mean value, a confidence interval and a risk level are output. According to the method, deep space-time joint modeling and multi-index collaborative prediction can be realized, the prediction precision and the training stability are improved, credible early warning information can be provided, and the method can be widely applied to water quality prediction of various environmental water bodies such as rivers, lakes and reservoirs.

The invention discloses a category rebalance overflow state identification method based on underground imbalance monitoring data, and relates to the technical field of petroleum drilling engineering and intelligent well control. The method comprises the following steps: acquiring multi-source underground monitoring data generated in a drilling process, and constructing an initial labeled sample set and an unlabeled sample pool; a joint self-supervision feature learning model is constructed, and depth characterization of drilling features is realized by introducing a feature extraction network based on an attention mechanism and potential feature space modeling; under an active learning framework, performing category grouping on unlabeled samples according to a model prediction result, and adaptively allocating a sample query number through a category rebalance quota optimization strategy; and further introducing an uncertainty estimation and mutual information measurement method based on random reasoning, preferentially selecting samples with relatively high information amount for manual annotation, and carrying out iterative updating on the model. According to the method, the problems of extremely unbalanced overflow sample distribution and high manual labeling cost are effectively relieved, the recognition precision and the model generalization ability of minority overflow events are remarkably improved under the condition of limited labeling budget, and the method has good engineering application value.

This invention relates to the field of computer vision and discloses a facial expression recognition method based on enhanced action units-guided causal inference. The method includes extracting high-dimensional semantic feature maps from the original image; performing reparameterized sampling to output a sequence of local visual feature blocks; calculating the geometric relationship between the feature sequence and facial key points to generate a position-enhanced feature sequence with superimposed embeddings; predicting the activation intensity and uncertainty of action units, and combining a static prior association matrix to perform causal intervention to generate a corrected feature sequence and a counterfactual feature sequence; and aggregating the corrected sequences to output the expression classification result. By dynamically adjusting the sampling distribution, the method focuses on high-discriminative regions and reduces noise; combines spatial structure and uncertainty estimation to suppress low-confidence features; and utilizes prior knowledge to perform causal inference to remove spurious correlations, verify causal sufficiency, and improve the model's generalization performance in complex scenarios.