25results about How to "Reduce variance" patented technology

The invention provides a deep-learning artificial lens degree determination method and device for a equi syndrome patient and electronic equipment, and relates to the technical field of medical artificial intelligence, and the method comprises the steps: obtaining clinical data of the equi syndrome patient; preprocessing the clinical data of the equi syndrome patient to obtain an eight-dimensional feature data set; performing first-layer prediction on the eight-dimensional feature data set through a plurality of regression models to obtain a plurality of intraocular lens preliminary prediction degrees; and splicing the plurality of preliminary predicted degrees of the artificial lens with the eight-dimensional feature data set, and performing second-layer prediction on the spliced data to obtain a final predicted degree of the artificial lens. According to the method, the IOL degree of the CEL patient can be accurately and stably calculated, so that the technical bottlenecks of a traditional formula and an existing single AI model are overcome, and the success rate of an operation and the visual prognosis of the patient are assisted to be improved.

This invention discloses a method and system for labeling prompt words in a large-scale communication domain model based on knowledge graphs. Belonging to the fields of communication technology and artificial intelligence, it addresses the following technical problems: low efficiency in labeling specialized terms in communication domain prompt word annotation, low accuracy and efficiency of automatic recognition, limited prompt word dimensions, and lagging domain knowledge. The method includes: sequence labeling and entity extraction using a trained combined model to output entities; generating relation triples as candidate relations using a trained generator, and scoring and ranking the selected relation triples using a trained discriminator; semantic parsing of user input to extract key entities and intents; multi-hop reasoning from a dynamic knowledge graph to retrieve relevant entities and relations; eliminating illusory content using a trained GAN network; generating entity prompt words from three levels: basic, extended, and application dimensions; and introducing a triple quality verification mechanism.

The invention discloses an operation state detection method and system, and the method comprises the steps: monitoring natural gas transmission and distribution equipment, and collecting multiple groups of operation state data of the natural gas transmission and distribution equipment; performing preprocessing and data fusion on the multiple groups of operation state data to obtain a data set; and training the neural network model based on the attention mechanism by using the data set to obtain a fault category prediction model, and detecting the operation state of the natural gas transmission and distribution equipment by using the fault category prediction model. According to the invention, monitoring and fault detection of the operation state of the natural gas transmission and distribution equipment are realized.

This invention belongs to the field of optical component technology, and particularly relates to a light-emitting optical fiber, comprising a fiber core with an average refractive index of n0, and a microstructure cladding covering the fiber core. The microstructure cladding includes two phases, one of which is arranged in an array within the other phase. The two phases are a high-refractive-index phase and a low-refractive-index phase, respectively. The high-refractive-index phase has a refractive index of n1, and the low-refractive-index phase has a refractive index of n2. The refractive index n1 of the high-refractive-index phase is greater than the average refractive index n0 of the fiber core, and the refractive index n2 of the low-refractive-index phase is less than the average refractive index n0 of the fiber core. A first protective layer and a second protective layer are sequentially coated around the microstructure cladding. Compared to polymer fiber cores, the light-emitting optical fiber provided by this invention has better weather resistance, is less prone to aging, has higher mechanical strength, and a simpler fabrication process. It is more conducive to light emission within the optical fiber, has a smaller variance in luminous intensity, and achieves total internal reflection, thereby ensuring higher luminous brightness.

This application relates to a joint estimation method for wireless communication delay and frequency offset based on hill-climbing fine search. The method includes: setting a time offset search range, a frequency offset search range, a coarse search step size, and a fine search step size based on prior information; generating a two-dimensional grid point set within the search range according to the coarse search step size; calculating the cost value of each grid point based on a cost function; taking the time offset and frequency offset corresponding to the maximum value as the coarse estimate; using the coarse estimate as the initial point, performing fine search iterations using hill-climbing; calculating the cost value of multiple candidate points in the neighborhood in each iteration, comparing and determining the update direction, and updating the current time offset and / or frequency offset; repeating the iteration until a preset exit condition is met, and outputting the final estimate. This application effectively solves the parameter coupling problem of traditional step-by-step estimation in NTN scenarios, while balancing search range and computational complexity, facilitating engineering implementation, and being applicable to various NTN systems and terminal forms.

The application discloses a kind of anti-degradation multi-sensor fusion SLAM method and system, belong to robot synchronous positioning and mapping technical field.The method includes obtaining original data from camera, laser radar, inertial measurement unit and global navigation satellite system, based on VIG tight coupling optimization unit, original data is preprocessed and initial estimate in front end, obtain global pose, based on LIG tight coupling optimization unit, motion distortion correction and feature extraction processing are carried out by original data and global pose, based on two-stage factor graph optimizer, laser feature and GNSS observation information are fused in two stages, global optimization is executed, and high-precision, globally consistent pose and map are output.The application overcomes the limitation of single sensor, suppresses cumulative error, realizes high-precision, high robustness and global consistency positioning and mapping in large-scale complex scene outdoors.

The present application belongs to the technical field of circular lifting belt, and particularly relates to a production method of circular lifting belt, comprising the following steps: using a first filament to bundle a plurality of parallel arranged filaments into a filament bundle body; winding the filament bundle body into a ring shape after passing through a leather, and the front end of the filament bundle body passes through the leather multiple times; winding a second filament from the front end of the filament bundle body before the front end of the filament bundle body passes into the leather, and the second filament is continuously wound outside the multiple turns of the filament bundle body during the multiple times of the filament bundle body passing into the leather; after the filament bundle body is wound for a specified number of turns, the tail of the filament bundle body is cut off to splice the head and tail of the filament bundle body; and the leather is stretched and sewn to obtain the circular lifting belt. The present application realizes the control of the neatness of the filaments by bundling the plurality of filaments and bundling the multiple turns of the filament bundle body, can make the lifting belt achieve higher filament utilization rate in the stretching use, and improve the breaking load.

The invention relates to the technical field of garment sewing, in particular to a cloth and sewing line colorimetric method, which comprises the following steps: acquiring a cloth image, extracting cloth colors of the cloth image based on a plurality of color extraction algorithms, and obtaining a cloth color value corresponding to each color extraction algorithm; obtaining suture data in a suture database, wherein the suture data comprises suture color values of different colors and corresponding suture color numbers; performing color difference calculation on the cloth color value corresponding to each color extraction algorithm and the stitch color value of the stitch data to obtain a similarity value; obtaining weight information of each preset color extraction algorithm, and calculating and obtaining a weighted similarity value based on the weight information and the plurality of groups of similarity values; and obtaining the sewing line color number with the highest similarity based on the weighted similarity value. The intelligent colorimetric operation is realized, and the time consumption and the difference of manually finding the nearest color sewing line are reduced.

The utility model provides a kind of tubular object manufacturing device, the device of the utility model includes: the first slitting assembly of original strip material is divided into at least one wide strip and at least two narrow strips;The glue assembly of one strip surface of the wide strip or narrow strip is coated with glue;The glue coating surface of each wide strip is bonded with at least two adjacent narrow strips with the way that narrow strip is in the same plane through glue coating surface to be double-layer bonded strip bonding assembly;The second slitting assembly of double-layer bonded strip is divided into double-layer strip along the gap between adjacent narrow strips;Forming assembly is curled into continuous double-layer tubular object with the axis parallel to the conveying direction of double-layer strip.This utility model's device effectively improves tubular object roundness, hardness, reduces raw material consumption, and the two layers of double-layer tubular object are completely consistent, can reduce product difference rate, improve product quality, while multi-channel parallel forming can multiply production efficiency.

The application discloses a method and system for realizing sleep staging by using Bayesian uncertainty, comprising: obtaining electroencephalogram signal data and preprocessing to obtain an unlabeled data set; using the data set to preliminarily train a Bayesian sleep network comprising two Bayesian sleep sub-networks with the same structure, outputting Bayesian uncertainty quantification results according to Bayesian entropy; selecting part of important samples according to the quantification results to label sleep staging labels, which are used for fine-tuning the network; passing the remaining samples which are not labeled through the two fine-tuned Bayesian sleep sub-networks respectively, outputting predicted sleep staging results, calculating Bayesian uncertainty quantification results, assigning weights to the predicted sleep staging results according to the quantification results and fusing the results to obtain overall predicted sleep staging results, selecting part of samples to correct the results and outputting final sleep staging results. The application can effectively reduce the dependence on sleep staging label labeling and improve sleep staging accuracy.

This invention relates to a method and system for enhancing low-light videos based on multi-scale spatiotemporal feature fusion, belonging to the field of computer vision technology. During the training phase, the low-light video is preprocessed according to its different data formats, separating different pixels into different channels using color filters. A low-light video enhancement algorithm generates an enhanced video from the input low-light video and compares it with a real normal-light video. The loss function is calculated, and the parameters in the algorithm model are updated through backpropagation. When a preset termination condition is reached, the model structure and the updated model parameters are saved. During the usage phase, a low-light video enhancement model is constructed based on the model structure and parameters saved during the training phase. The low-light video to be processed is then fed into the model for low-light enhancement. This invention can restore low-light videos to high-quality normal-light videos with excellent evaluation metrics and good subjective visual experience.

The embodiment of the utility model provides a wiredrawing aluminum plate, and relates to the technical field of aluminum plate strips. The wiredrawing aluminum plate provided by the utility model comprises a substrate, a wiredrawing aluminum layer and a passivation layer, the first surface of the substrate is used for being attached to a to-be-mounted surface, the wiredrawing aluminum layer is arranged on the second surface of the substrate, and the surface, away from the substrate, of the wiredrawing aluminum layer is provided with decorative threads; the passivation layer at least covers the surface, deviating from the substrate, of the wiredrawing aluminum layer; and the decorative threads can be exposed through the passivation layer. The substrate is arranged as a structure supporting layer, the surface flatness and the material structure stability of the substrate provide a uniform adhesion foundation for the wiredrawing aluminum layer, and the consistency of the overall surface quality is guaranteed. The wiredrawing aluminum layer and the base plate are arranged in a split mode, decorative threads do not need to be manufactured on the base plate, the decorative threads of the wiredrawing aluminum layer can penetrate through the passivation layer to be exposed, the wiredrawing aluminum layer is directly arranged on the base plate, the consistency of the visual effect presented by the wiredrawing aluminum plate is guaranteed, and the difference of wiredrawing aluminum plates of different batches can be reduced.

The present invention discloses an electroplating device for automotive parts, comprising: a suspension rod, the upper end of which can be connected to an external bracket so that the lower end thereof can be suspended in an electroplating bath; a non-contact hanger, connected to the lower end of the suspension rod and having one or more upward working surfaces, the working surfaces being arranged to match the outer surfaces at corresponding positions on the product, and the working surfaces having jet holes for forming a dynamic flow layer capable of carrying the product between the working surfaces and the outer surfaces of the product; a water pump, arranged outside the electroplating bath and communicating with the jet holes for providing the plating solution required to form the dynamic flow layer. The present invention provides an electroplating device for automotive parts, which can ensure that there is no direct contact between the hanger and the outer surface of the product during the electroplating process, so that the coating on the outer surface of the product is continuous, complete and free of indentations.

This invention discloses a diluent for drug fluorescence detection, its preparation method, and its application. The diluent contains a buffer, a stabilizer, an inorganic salt, and a dispersant. The buffer includes any one or a combination of at least two of tris(hydroxymethyl)aminomethane, phosphates, or carbonates. The stabilizer includes any one or a combination of at least two of bovine serum albumin, casein, or PEG. The dispersant includes any one or a combination of at least two of polyvinyl alcohol, polyethylene glycol, or lauryl alcohol. This invention creatively discovers that a diluent composed of these four components—buffer, stabilizer, inorganic salt, and dispersant—can significantly improve the specificity and uniformity of drug detection reagents, increase product stability, reduce inter-product variability, and improve detection accuracy. It is suitable for rapid screening in public places. Furthermore, the components are relatively safe and will not cause environmental pollution.

The invention discloses a federal learning parameter aggregation method and system based on an optimal probability flow. The method is used for federated learning of a central server and a plurality of clients, the central server is provided with a flow matching model and is used for learning empirical distribution of parameters uploaded by all the clients, and in each round of communication, the following steps are executed: the central server distributes parameters of a current global machine learning model to the selected clients; each selected client performs multiple rounds of training on the local model on the local data set to obtain updated local model parameters, and uploads the updated local model parameters to the central server; and the central server collects parameters from all participating clients, and executes a stream aggregation operation by using the stream matching model to generate a new round of parameters of the machine learning model until a set convergence condition is met. According to the method, the generalization performance and robustness of the federated model in a heterogeneous scene are improved.

This invention relates to the fields of clinical laboratory medicine and artificial intelligence, and provides a risk prediction method and device based on machine learning for clinical mass spectrometry. The risk prediction method includes: defining N+M dimensions of features based on a liquid chromatography-tandem mass spectrometry system to obtain an N+M dimension feature vector structure; obtaining a virtual training dataset based on the N+M dimension feature vector structure, acquiring parameter values ​​of each dimension of the current batch through a data acquisition interface, and assembling them into N+M dimension feature vector data; performing format verification and invalid value filtering on the N+M dimension feature vector data using the N+M dimension feature vector structure to obtain a feature matrix; and obtaining a standardized real-time risk score based on the feature matrix and a trained fusion-integrated risk prediction model. This invention utilizes a machine learning model to uncover the complex nonlinear relationship between configuration parameters and dynamic parameters, thereby achieving more accurate and forward-looking risk warnings than traditional single-threshold methods.

This application discloses a diffractive waveguide and an augmented reality display device. The diffractive waveguide includes a waveguide substrate and a first grating and a second grating disposed on the waveguide substrate. The first grating is configured to couple input light carrying image information into the waveguide substrate and propagate it through total internal reflection. The second grating includes a diffraction relief structure configured to at least dilate the pupil of light propagating therein. The diffractive waveguide also includes an optical film layer that at least covers the second grating and includes a filling structure. The filling structure fills the gaps in the diffraction relief structure of the second grating and forms an overflow layer that extends beyond the diffraction relief structure. The refractive index of the filling structure is greater than the refractive index of the ambient medium and less than the refractive index of the second grating. According to embodiments of the present invention, the transmittance of ambient light in the grating region can be effectively improved, and the difference between the grating region and the non-grating region can be reduced, thereby improving the overall visual experience.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training diffusion neural networks using differentiatable reward functions.

The invention provides a reinforcement learning-based large model intention understanding fine tuning method and device, and relates to the technical field of artificial intelligence, and the method comprises the steps: processing a current input based on a large model, and generating a local sample group containing a plurality of candidate responses; determining a reward value of each candidate response based on the intention matching reward value and the slot position matching reward value of each candidate response; based on the difference between the reward value of each candidate response and the reward mean value, determining the relative advantage value of each candidate response; and updating the parameters of the large model based on the relative advantage value of each candidate response. According to the method and the device provided by the invention, strategy gradient updating is carried out by adopting the relative advantage value, so that the variance of the reward estimation value is effectively reduced, and the instability problem possibly caused by independent establishment of a value network is avoided; and in combination with a refined reward function which is specially designed for the intention understanding task and comprises an importance level and a multi-dimensional matching strategy, the accuracy of executing the intention understanding task by the large model is improved.

The invention discloses a target recognition method and device based on a scene, and aims at improving the target recognition efficiency in an image for the scene. The scheme provided by the invention comprises the steps of obtaining a to-be-recognized feature vector of a to-be-recognized image; consistency comparison is carried out on the to-be-recognized feature vector based on a preset feature library through a preset comparison rule, so that whether a preset target feature vector consistent with the to-be-recognized feature vector exists or not is determined, and the preset feature library comprises a preset target feature vector corresponding to a preset target; if not, identifying a target scene category to which the to-be-identified image belongs based on the to-be-identified feature vector; and performing consistency comparison on the to-be-identified feature vector based on a preset feature library through a comparison rule matched with the target scene category, and determining a preset target corresponding to a preset target feature vector which is consistent through comparison as a target corresponding to the to-be-identified image.

The application discloses a digital person video updating method and device, equipment, medium and product, and relates to the technical field of video generation. The method comprises the following steps: acquiring current input content comprising current script text; determining the relevance of the original script text and the current script text and a video adjustment scheme based on the text correspondence relationship between the original script text and the current script text; determining the lower body movement animation sequence of the current script text based on the video adjustment scheme; and generating a target digital person full-body movement video based on the lower body movement animation sequence and the current input content. The application determines the video adjustment scheme according to the modification of the script text, then determines the lower body movement animation sequence, and finally determines the target digital person full-body movement video by using the lower body movement animation sequence and the current input content. The qualified lower body movement animation sequence in the initial digital person full-body movement video is reused as much as possible, the video updating efficiency is improved, and the resource consumption of the video updating work is reduced.

The invention discloses a quantitative seismic sedimentology method, which relates to the technical field of oil-gas exploration, and comprises the following steps of: establishing a nonlinear learning model of logging curves (Vp, Vs and RHOB) and a gamma curve (GR) by adopting a random forest learning method, and then applying the model to a three-dimensional elastic parameter body (Vp, Vs and RHOB) to obtain a three-dimensional gamma (GR) body; a three-dimensional gamma (GR) body with clear geological significance is obtained through a data driving mode, and the problem of obtaining a three-dimensional geological parameter body (GR) through three-dimensional seismic data is solved; secondly, replacing a three-dimensional seismic data volume adopted by the existing method with a three-dimensional gamma (GR) data volume in the process of seismic sedimentology analysis and stratigraphic slice analysis, and realizing quantitative seismic sedimentology interpretation under the constraint of a lithology quantitative interpretation standard; and finally, since the random forest algorithm carries out sampling every time to train the model, the generalization ability is very strong, and the method plays a role in reducing the variance of the model.

This invention relates to the field of energy management technology. It discloses a method and system for determining environmental-economic coordination degree based on a coupled model. The method includes addressing the potential for boundary diffusion and sample mismatch caused by global smoothing and single-scale alignment; constructing a de-jittering filtering link around the operating condition set; generating a window table using energy estimation vectors and contrast vectors; performing edge-preserving filtering on the encoded amplitude sequence to obtain a smooth set consistent with the operating condition set; converging neighborhood oscillations during start-stop and load switching; preserving boundary transitions; reducing segment variance; obtaining low-noise input for the short window table and sample matrix; reducing coordination vector fluctuations; generating an alignment table through joint modeling of alignment scale and hierarchical index; performing minimum cost execution under hierarchical records and combining anomaly index removal to obtain a delay vector; and publishing it window by window according to a stable Boolean sequence, outputting only the delayed and stably coupled coordination vector, thereby suppressing three-channel sampling delay and achieving stable short-window output of coordination degree.

The application belongs to the technical field of natural language processing, and discloses a federated learning optimization method and system for entity extraction, a device and a terminal. In the shared data meta information transmission stage, the server sends a shared data request to each selected client participating in federated learning, each client calculates shared data set meta information, and uploads the meta information to the server. In the approximate IID entity annotation data construction stage, the server constructs an approximate IID entity annotation data index set according to the meta information uploaded by the client, and requests the corresponding client for data corresponding to the index set. In the model training stage, the server is regarded as a client participating in federated learning training, and the approximate IID entity annotation data set is used for model training to obtain an approximate centralized model W IID The application reduces the difference between the model parameter spaces of the clients, improves the accuracy and convergence speed of the global model, and reduces the training communication cost.

The embodiment of the application provides a continuous sign language recognition method and device based on sign language word prior guidance, relates to the technical field of continuous sign language recognition, and specifically comprises the following steps: acquiring a continuous sign language recognition data set, fine-tuning the continuous sign language recognition data set by using a pre-trained BERT model to extract sign language word features with action invariance, and constructing a visual feature extraction model to extract sign language visual features; constructing a cross-modal feature alignment target function and a CTC target function to construct a target function, and training the visual feature extraction model to form a cross-modal feature alignment model. Through the method and device provided by the application, the technical problem of weak generalization ability of visual features in continuous sign language recognition in the prior art can be alleviated, the sign language word features with action invariance are used as priori, the multiple video frame visual features corresponding to the sign language word features are constrained, and finally the difference between the visual features is reduced.