14results about How to "Reduce training complexity" patented technology

The embodiment of the invention provides a data processing method and a training method and device for an image text data quality enhancement model, and the method comprises the steps: obtaining an image and initial description information, inputting the image and initial description information into the image text data quality enhancement model, and obtaining an output result of the model; the image text data quality enhancement model is used for obtaining evaluation information according to the image and the initial description information, generating enhanced description information of the image based on the image when the evaluation information indicates that the matching degree of the initial description information and the image is low, and taking the enhanced description information and the evaluation information as an output result; and when the evaluation information indicates that the matching degree between the initial description information and the image is high, taking the evaluation information as an output result. If the output result is evaluation information, determining the image and the initial description information as training data of an MLLM model; and if the output result is the evaluation information and the enhanced description information of the image, determining the image and the enhanced description information as training data of the MLLM model. The method is used for enhancing data quality and improving data utilization rate.

The invention discloses an adjusting and optimizing method and system based on a large power dispatching model, and belongs to the technical field of power dispatching, and the method comprises the following steps: obtaining historical power dispatching knowledge, and dividing the historical power dispatching knowledge into regional knowledge and global knowledge based on regional characteristics; constructing a first teacher model based on the difference between the global knowledge and the regional knowledge, and constructing a student model based on the lowest fluctuation cost; obtaining regional scheduling correlation according to historical power scheduling knowledge, and constructing a second teacher model according to the regional scheduling correlation; associating the first teacher model, the second teacher model and the student model based on knowledge distillation, and constructing a large power dispatching model; and outputting a tuning strategy according to the large power dispatching model and the current power data. The beneficial effects of the invention are that the output of the power tuning strategy gives consideration to pertinence and efficiency.

The invention discloses a multi-modal target identification method based on anti-noise contrast loss, and the method comprises the following steps: S1, selecting different encoder architectures and decoder architectures for each modal according to the input multi-modal target information, and respectively taking the encoder architectures and the decoder architectures as a feature extraction network and a reconstruction network; s2, constructing a multi-modal target recognition network according to the number of input multi-modal target categories; s3, performing training according to the anti-noise comparison loss and the modal reconstruction loss, and updating the feature extraction network, the reconstruction network and the recognition network; s4, inputting modal data of a to-be-identified target into the feature extraction network, fusing the modal data, inputting the fused modal data into the identification network, finally obtaining dimension vectors with the same category number, and selecting the category with the maximum probability value as a prediction category of the to-be-identified target; according to the invention, the problem of low recognition accuracy in a traditional cross entropy-based multi-modal target recognition method in a noisy scene is solved, and an over-fitting phenomenon caused by tag noise is relieved.

The invention relates to the technical field of near infrared spectrum analysis, in particular to a near infrared spectrum analysis method and system based on spectrum image feature fusion. The method comprises the following steps: acquiring physicochemical properties corresponding to a batch of samples, measuring near infrared spectrums of the samples and forming a modeling sample set; randomly selecting two near infrared spectrums from the sample set to construct a two-dimensional grey-scale map; extracting near infrared spectrum characteristics from the two-dimensional grey-scale map, and preprocessing the near infrared spectrum characteristics; processing the near infrared spectrum features by using a principal component analysis technology to generate feature vectors; constructing a regression model based on the feature vector and the physicochemical property data of the sample to fit the relationship between the feature vector and the physicochemical property data of the sample; and inputting the feature vector of a to-be-analyzed sample to the regression model, and outputting a prediction result of the physicochemical properties of the to-be-analyzed sample. According to the method, the near infrared spectrum analysis precision is improved by adopting a directional rapid rotation brief feature image extraction technology.

This application provides a neural network operation method, apparatus, and storage medium, including: adding a random mask layer to a first neural network according to the type of variable dimensions to obtain a second neural network; the random mask layer is used to perform random masking processing on tensors input to the random mask layer; and training the second neural network using sample data. The neural network operation method, apparatus, and storage medium provided in this application, by adding a random mask layer to the neural network, only require training and deployment of one neural network to adapt to different input and output dimension requirements. This method has low training complexity, low storage overhead, and is easy to deploy and continuously evolve through online training.

The invention belongs to the field of computer vision and artificial intelligence, and discloses a salient object detection method based on a diffusion model and semantic guidance. Comprising the following steps: automatically generating text annotations of a saliency object data set; constructing a cross-modal saliency pseudo mask; carrying out structure optimization on an initial pseudo mask; constructing a saliency object detection framework based on a diffusion model; and training and testing the salient object detection model based on the diffusion model and the semantic guidance by using the salient object data set. According to the method, the saliency segmentation result can be effectively recovered, verification is carried out on a plurality of public data sets, and the saliency detection precision and the boundary integrity can be remarkably improved.

The invention relates to an artificial intelligence medical auxiliary data processing method and system, and the method comprises the steps: constructing a medical data processing model which comprises a K + 1 channel feature fusion network model and a machine learning model; constructing a dynamic weight model; the input of the dynamic weight model is an L-element input vector, and the output of the dynamic weight model is a K-element output vector; the K-element output vector is input for constructing a (K + 1) th channel of the medical data processing model; acquiring to-be-processed medical data, and constructing K-mode medical data input based on the to-be-processed medical data; and inputting the auxiliary data into the 1-K channels of the medical data processing model, and obtaining corresponding auxiliary data. According to the method, through data dimension splitting, weight training is rapidly reduced, rapid training of the medical data model is realized under the condition of less medical data support and training complexity, and the medical data auxiliary efficiency is improved.

The invention provides a direct shear test soil parameter calibration method fusing numerical simulation and machine learning, and relates to the field of geotechnical engineering. According to the direct shear test soil parameter calibration method fusing numerical simulation and machine learning, high-fidelity finite element numerical simulation, Kernel PCA dimension reduction, a PINN agent model and Bayesian parameter inversion are subjected to systematic deep fusion for the first time. The progressive damage process from the edge to the interior and non-uniform stress-strain distribution of a soil body in a direct shear test are precisely restored through high-fidelity finite element numerical simulation, low-dimensional essential characteristics of a high-dimensional response curve are extracted through a dimension reduction method, and a constitutive model and a strength criterion are embedded into an agent model to guarantee physical rationality. And finally, parameter estimation is realized through a Bayesian parameter inversion framework. The combination solves the interference of non-uniform stress and strain of the soil body in the direct shear test on parameter calibration, so that the calibration result better fits the actual mechanical property of the soil body.

The application relates to the technical field of wireless communication, and provides a signal recovery method and device for an orthogonal frequency division multiplexing system and electronic equipment. The method comprises the following steps: collecting a received signal; the received signal is obtained by receiving a transmitted signal of the orthogonal frequency division multiplexing system; determining a signal recovery result of the received signal; inputting the received signal and a pilot sequence into a conditional generative adversarial network to obtain a signal recovery result output by the conditional generative adversarial network. In the application, the conditional generative adversarial network is used for signal detection and signal recovery of the received signal of the orthogonal frequency division multiplexing system; the multiple cascaded sub-layers of the generator in the conditional generative adversarial network are used for gradient updating based on a rank restriction reverse transmission method with a constant matrix rank during training; the multiple cascaded sub-layers are trained as a whole, and the training of a large dense layer is replaced by the training of a lightweight sub-layer, so that the network training complexity is reduced, the data processing speed is improved, and the calculation efficiency is improved.

The invention provides a ponding image day and night bidirectional conversion model construction method based on reflection map consistency and application, and the method comprises the following steps: obtaining a plurality of groups of real day and night ponding image pairs comprising real daytime ponding images and real night ponding images, and generating daytime scene cues for each real daytime ponding image, generating a night scene prompt word for each real night ponding image; constructing a waterlogging image day and night bidirectional conversion framework comprising a brightening conversion unit and a darkening conversion unit, wherein the waterlogging image day and night bidirectional conversion framework generates a simulated night waterlogging image and a simulated daytime waterlogging image; and performing iterative training on the ponding image day and night bidirectional conversion architecture by using a plurality of groups of real day and night ponding images to obtain a ponding image day and night bidirectional conversion model. According to the scheme, the reflection consistency decoder is constructed to extract the reflection maps of various ponding images and calculate the reflection consistency loss, the conversion process is constrained without changing the essential attribute of the ponding area, and semantic distortion is effectively prevented.

The invention discloses a domain-adaptive span-level aspect emotion triple extraction model establishment method and application thereof, and belongs to the technical field of emotion analysis, and the method comprises the following steps: extracting semantic information by adopting a span-level feature extraction method combining syntactic and semantic information, and shielding the expression of a prediction enhancement feature on syntactic information through part-of-speech tagging to obtain a domain-adaptive span-level aspect emotion triple extraction model; syntactic information and semantic information are fully utilized by span features; the domain invariant features of the span are enhanced by taking the dependency relationship tree as the core, the feature distribution is optimized by utilizing a mutual information maximization loss function, the features aiming at the specific field are transferred to be field-independent, and the cross-field prediction is realized on the premise of not increasing the labeling cost; according to the method, simple span-level tags are designed and only comprise six tags, so that under the condition of ensuring accurate extraction of an emotion triple in the aspect of span level, the calculation burden in the training process is reduced, and a clearer learning target is provided. According to the method, on the premise that the labeling cost is not increased, cross-domain span-level aspect emotion triple extraction can be realized.

The application belongs to the field of information security, and provides a high-robustness image tampering detection method, which comprises the following steps: 1) obtaining multi-scale feature extraction by using a hole convolution, and using Concat to obtain a global feature representation, so that the loss of shallow layer features after multiple convolutions can be prevented, and a larger receptive field can be obtained, and the effect of small target tampering detection is improved; 2) extracting noise by using BayarConv, and the noise extractor is better than an SRM filter, and Bayarconv is developed to enhance the noise inconsistency between the manipulated area and the real area in a given image; 3) using a residual block, batch normalization and an activation function to capture spatial features, and the structure is a conventional encoder, and the main purpose is to cross-fuse with semantic features; 4) using a cross-attention mechanism to fuse spatial features and semantic features, and the fused features are input into an up-sampling layer and a convolution layer for detection, and a hybrid loss function is used to train the model. Experimental data on a large number of public data sets show that the cross-attention mechanism provided in the application can not only accurately predict the tampering position, but also more accurately identify the tampering type.

The invention discloses a data expansion and exoskeleton joint end-to-end torque estimation method based on a diffusion model, and the method comprises the steps: carrying out the normalization processing of time series data from a multi-source sensor, carrying out the fragmentation according to a fixed length, and constructing a training sample with a motion class label; then, training a classifier-free conditional diffusion model by adopting a sample, and simultaneously learning conditional and unconditional denoising mapping relationships in a manner of randomly inactivating category conditions; in the generation stage, based on a classifier-free condition guidance mechanism, multi-modal time series data with specified motion category features are gradually generated from random noise; and finally, fusing the generated synthetic data with real acquired data to train a joint torque end-to-end prediction network, thereby realizing joint torque estimation of input sensor time sequence data. The method improves the prediction precision, generalization ability and stability of the end-to-end torque estimation model in a multi-action and few-sample scene, and has a good engineering application value.

This invention relates to a text-driven immersive large-scene neural rendering and blending enhancement method, comprising: 1. Creating a dataset; 2. Reconstructing the large scene based on an improved progressive neural radiation field; 3. Predicting the foreground and background matte values ​​of the rendered image based on a convolutional neural network; 4. Generating the background from text using a stable diffusion model; 5. Calculating background motion through changes in camera parameters between adjacent frames; 6. Blending the rendered foreground and background and coordinating lighting. This invention achieves real-time text-driven, editable large-scene background enhancement, allowing rendering of immersive large scenes at any observation position at a city scale, generating observation images with effects consistent with the real scene, and enabling mixed reality on this basis. It can satisfy users' personalized customization of the scene and achieve visual effects at the level of film and television special effects. This technology can be applied to fields such as 3D visualization, digital mapping, and virtual reality games.