48results about How to "Improve experiment results" patented technology

The invention discloses a compression method of a neural language model based on a tensor decomposition technology. According to the method, starting from linear representation of an original attention function, it is proved that the attention function can be linearly represented by a set of standard orthogonal basis vectors, and then parameters are compressed by sharing the set of basis vectors under the condition that a multi-head mechanism is constructed; meanwhile, the neural network model can have stronger discrimination capability through modeling in a tensor slicing mode. A new idea isprovided for developing a neural network model with low parameters and high accuracy.

The invention discloses a high-low temperature magnetic field detection system and relates to a device for detecting magnetic element magnetic field distribution characteristics under high-low temperature environments. The high-low temperature magnetic field detection system comprises a box, a measured workpiece and a Hall probe. The box is made of a nonmagnetic heat insulation material, the Hall probe is installed a heat insulation carrier capable of freely extending in and being drawn out of the box, an infrared temperature detecting instrument for measuring the temperature of the surface of the measured workpiece is arranged outside the box, a distance measuring device for measuring the distance between the Hall probe and the measured workpiece is further arranged on the outer portion of the box, and the measured workpiece performs refrigeration or heating through a liquid nitrogen inlet and a hot nitrogen inlet which are arranged on the box and are symmetrically distributed. The high-low temperature magnetic field detection system is quick in heating or refrigeration efficiency, small in power consumption and suitable for batch production, high-low temperature switching is performed in the box, a temperature fatigue test can be performed, and the high-low temperature magnetic field detection system is simple and convenient and reduces cost; the Hall probe enters the box during the measurement and is drawn out at once after the measurement is finished, and the damage to the Hall probe caused by high temperature is avoided.

The invention discloses a consumption intention identification and prediction method based on a consumption cause map, and belongs to the technical field of consumption intention identification and prediction. According to the invention, the problem of low consumption intention identification and prediction accuracy in the prior art is solved. According to the main technical scheme, the method comprises the steps: step 1, performing event extraction based on a pre-training model; step 2, extracting a relationship between events based on the pre-training model; step 3, constructing a bipartitegraph as weak supervision information by adopting an unsupervised method based on the comment corpus; step 4, constructing a consumption reason graph based on the bipartite graph weak supervision information and the annotation data; and step 5, training a homogeneity relationship attention model by using the training data, and judging a corresponding relationship between the event and the consumption intention by using the homogeneity relationship attention model. The method can be applied to consumption intention identification and prediction.

The invention discloses a device for interactive reaction between heavy metals and particulate matter and belongs to the field of interactive reaction devices. The device comprises a flue gas mixing system, a temperature control hearth and a tail gas processing system, wherein a temperature control device is arranged in the temperature control hearth, an ejector, an atomizer and a sampling hole are respectively arranged outside of the temperature control hearth, the front end of the ejector is connected with one end of the flue gas mixing system and one end of the atomizer respectively, the other end of the atomizer is connected with a peristaltic pump, and the tail end of the temperature control hearth is connected with the tail gas processing system. The device can be used for researching a reaction mechanism of specific ingredients of particulate matter and heavy metals under different conditions in different concentrations of heavy metal atmospheres, and the peristaltic pump combines with an atomizer and double-by-path venturi ejector combined device, so that the particulate matter and the heavy metal atmospheres re evenly mixed, and control is convenient. In addition, feeding is performed from the cylindrical tangent plane of the hearth, so that mixing is violent, and the staying time of fine particulate matter and heavy metal flue gas in the hearth is also prolonged.

The invention discloses a rock experiment standard test piece drying and heat shrink tube baking device and an operation method thereof. The device is composed of a vacuum chamber, a drying chamber and a baking chamber. The drying chamber can be placed in the vacuum chamber and is used for drying a rock experiment standard test piece. The backing chamber can be placed in the vacuum chamber and is used for baking a heat shrink tube. Through the device, the experiment efficiency and experiment precision are greatly improved; the phenomenon of test piece and base dislocation caused by handholding and rotating an experimental device when a traditional heat gun is used for baking a rock experiment standard test piece and a heat shrink tube is effectively avoided; and the problem that the traditional heat gun scalds hands when used for baking the rock experiment standard test piece and the heat shrink tube is solved. Meanwhile, the drying chamber and the baking chamber can be rotated so as to achieve the even heating effect. Accordingly, the success rate of experiments is increased, and the accuracy of experiment results and data is improved.

The method discloses a method of cell counting, which comprises the steps: the adherent culture of the cells of an experimental group to be counted and the cells of a criterion group with a known number is carried out on a plane surface; an original culture medium is soaked up, then the cells are washed, after that, glutaraldehyde solution is added for soaking, and the cells are dried in the air after washing by phosphate buffer solution; the fluorescence intensity of samples is detected by iraser; a standard curve is drawn by combing the known number of the cells and the fluorescence intensity so as to obtain a relation formula between the fluorescence intensity I standard and the number of the cells N standard: I standard equals to a multiplied by N standard and then plus b; and according to the standard curve and the fluorescence intensity of the cell samples in the experimental group, the number of the cells in the experimental group is calculated, namely, N experiment equals to (I experiment minus b) / a. The invention reflects the beneficial effects of simplifying experimental procedures, reducing cost, improving experimental precision and result parallelism.

The invention discloses an image generation method based on a Fourier transform attention mechanism, and belongs to the field of computer vision. The method comprises the following steps: firstly, selecting a generative adversarial network as a basic framework, normalizing training pictures, and sampling normal distribution to obtain noise; improving the original attention mechanism, using the characteristic that discrete Fourier transform can be combined with all position information to carry out calculation for replacing an original calculation pixel point correlation feature map part with high complexity; and obtaining a feature map combining all position information. During training, noise and picture categories are input into the network at the same time, and a generative adversarial network algorithm is utilized to train the model. After the network is trained, the image generation task can be completed by inputting noise and picture categories in the generative adversarial. According to the invention, the calculation complexity and the time complexity of a self-attention mechanism can be remarkably reduced, and the quality of images generated by an existing method and the diversity of the images are improved.

The invention relates to an apparatus and method for precisely detecting Bauschinger effect of repeated bending of a metal plate, the apparatus comprises a mechanics loading module and a plate clamping adjustable module fixed at the bottom of the mechanics loading module; the mechanics loading module is fixed and suspended through a testing machine and is vertically arranged, X-direction regulating blocks are symmetrically arranged at left side and right side along X-direction of the mechanics loading module, the front end and rear end of each X-direction regulating block are respectively fixed on the Y-direction regulating blocks, and is capable of respectively being driven by the Y-direction regulating blocks to horizontally slide on a pedestal along the Y-direction back and forth; the operation is easy and has automation function, force and displacement during a loading process can be measured, the coordinate figure, curvature radius and strain value at random point on both surface of the plate during the loading process can be measured, the error of test data is less, and the flexibility requirement of the Bauschinger effect for the plate can be satisfied, so that accurate calibration basis for determining material parameter of Bauschinger effect for the plate can be provided.

The human immunity function reconstructed mouse endometriosiss model and its constructing method features that human early pregnancy peripheral blood mononuclear cell is fed into non-obese diabetic serious combined immune deficiency mouse body to reconstruct immunity function and the decidual tissue of the same human is planted in the mouse body to establish human immunity function reconstructed mouse endometriosiss model. The model establishing method features that the peripheral blood and the decidual tissue are taken from the identical woman in artificial abortion operation. The peripheral blood mononuclear cell is fed into mouse body, the decidual tissue is planted under skin of mouse abdomen, and the cut is sawn with bacteria-free suture. The NOD-SCID mouse has human immune, transplanted tissue maintaining the tissue state of endometrium, obvious and stable characteristic, short observation period and experiment result with great clinical reference significance.

The invention relates to an instrument for sectioning and dyeing a biological sample on an electronic microscope, which belongs to the field of biological sample preparation. The instrument is composed of a locating handle, a water outlet, a groove flap board, a wattle, a built-in coverless box, a top cover, a stable gasket and a groove frame, wherein the groove flap board is arranged in a top groove of the groove frame; the locating handle is arranged on the outer side of the groove frame; the wattle is arranged above the groove flap board; the built-in coverless box is located in the top groove of the groove frame and arranged on one side of the groove flap board; the top cover is covered on the top groove of the groove frame; and the bottom part of the groove frame is provided with the stable gasket. According to the invention, the instrument can perform dyeing on biological sample parity for many times without repeated manual operations; thus, the time difference can be reduced; the generation of pollutant is eliminated; a high-quality experiment result is ensured to obtain; and the instrument can be used for immunohistochemical incubation of a histochemical microscope and preparation of macromolecule sample drop tablets and the like.

The invention discloses a pedestrian attribute recognition system and method based on multi-layer feature learning. The system includes a bottom-up feature extraction module, a top-down feature fusion module, a feature prediction module, a multi-layer prediction fusion module and a test module. , the specific steps of this method are: process the image layer by layer from bottom to top to obtain multi-layer features; fuse the features of adjacent layers from top to bottom, compress the channel of the feature map obtained by the higher layer, and combine it with the upper layer The sampled feature map is subjected to feature fusion and channel dimensionality reduction, and the current layer features are output; the fused features and the extracted uppermost layer features are passed through the maximum pooling layer and the fully connected layer to obtain preliminary prediction results at different levels; The preliminary prediction results are superimposed, and each attribute of each layer of prediction is given a corresponding weight value to obtain the final prediction result; the prediction result corresponding to the picture is extracted, and the results of each index are calculated. The present invention learns a set of specific weights for each attribute based on the predicted value obtained by the fused features, so that each attribute can better use multi-layer features to obtain better recognition effect.

The invention discloses a high-temperature tensile test clamp suitable for a steel wire with a special-shaped section, relates to the field of clamp design, and aims to solve the problems that an existing clamp cannot clamp and fix the steel wire with the special-shaped section, greatly damages a detected material during high-temperature tensile test, and further causes inaccurate test results anderrors. A chuck of the clamp is cylindrical, a boss is arranged in the center of the lower surface of the chuck, a square through hole is arranged in the center of the upper surface of the chuck, twopin holes are uniformly and symmetrically arranged along the circumferential outer wall of the chuck, a lantern ring is in a cuboid shape with a cavity, n square bosses are arranged in the top of thelantern ring, n represents a positive integer, groups of through holes are symmetrically arranged in the two side surfaces of the lantern ring, and the chuck penetrates one end of the lantern ring and is fixedly connected via the pins. The clamp is suitable for clamping the steel wire with the special-shaped section, and can be used for high-temperature tensile test.

A method for extracting news events of Internet politics and diplomacy disclosed by the present invention aims to provide an extracting method capable of improving the accuracy of event identification. The word set defines event categories, constructs trigger vocabulary and event category templates containing trigger words and event argument roles for each type of event; combines text-dependent syntax to analyze, identify and extract event elements in foreign and political fields. The single-text preprocessing operation is completed, and the category event trigger words are calculated and expanded based on the sememe similarity; the sentence that meets the threshold of similarity is taken as the candidate event sentence. Filter the event elements that satisfy the event category template, and extract the entity elements in the event sentence; then fill the event elements into the corresponding argument roles according to the event template; filter the candidate event elements that meet the category template; generate the event structure according to the event template Description file to build an event library in the field of foreign affairs.

The disclosure discloses an active prediction method and system for abnormal calls based on ranking learning and integrated learning, including: collecting phone samples, selecting features of the phone samples based on analysis and combination; dividing the collected samples into a training set and a test set ;For the training set samples, sorting learning is used to process the data, and the obtained results are used as a new test set, and then n sets of new training sets are formed to continue through the learning model to obtain n sets of results, and then these n sets of results are passed through ensemble learning , output the final test result. The disclosure has beneficial effects: the accuracy rate of predicting abnormal calls using ranking learning and integrated learning is higher than that of single using ranking learning, and the conventional method is more active in predicting abnormal calls than our method, and can solve large-scale data problems.