Abstract: Disclosed is a method for meta-knowledge fine-tuning and platform based on domain-invariant features. According to the method, highly transferable common knowledge, i.e., domain-invariant features, in different data sets of the same kind of tasks is learnt, the common domain features in different domains corresponding to different data sets of the same kind of tasks learnt in the network set are fine-tuned to be quickly adapted to any different domains. According to the present application, the parameter initialization ability and generalization ability of the universal language model of the same kind of tasks are improved, and finally a common compression framework of the universal language model of the same kind of downstream tasks is obtained through fine tuning. In the meta-knowledge fine-tuning network, a loss function of the domain-invariant features is designed in the present application, and domain-independent universal knowledge is learn.

Abstract: A co-frequency co-time full duplex (CCFD) signal receiving method includes: taking the sent baseband signal as the self-interference reference signal, reconstructing self-interference, and then performing primary self-interference cancellation on the received signal; processing, by using a timing synchronization loop, the signal after the primary self-interference cancellation, realizing timing recovery at the optimal sampling point of the useful signal through resampling a, and controlling resampling b1 and resampling b2 after performing low-pass filtering on the timing error signal in the timing synchronization loop, to recover the optimal sampling points of the self-interference reference signal and the received signal respectively; and performing joint self-interference cancellation and equalization on the resampled self-interference reference signal and the resampled received signal, and receiving the useful signal through signal demodulation.

Abstract: The present invention discloses an edge calculation-oriented reparametric neural network architecture search method, including the following steps: S1: designing linear operators and multi-branch block structures; S2: constructing a hypernetwork by stacking the multi-branch block structures; S3: training the hypernetwork through a gradient-based first-stage search algorithm; S4: deleting redundant branches in the hypernetwork to construct an optimal subnetwork; S5: converting the multi-branch optimal subnetwork into a single-branch network; and S6: completing task reasoning by using the single-branch network. The method is used to search the neural network structure capable of performing reparameterization, and ensures the reasoning real-time performance and the high efficiency of model operation while ensuring the reasoning precision.

Abstract: A phase noise suppression method for a multiple-input multiple-output (MIMO) system with a plurality of co-reference channels includes: dividing the phase noise of each channel in the MIMO system into common phase noise and independent phase noise, and constructing a certain number of joint phase states for the independent phase noise; inserting a pilot sequence into the sent signal based on a preset cycle, obtaining the common phase noise based on the pilot at receiver, and performing compensation; and performing signal demodulation on each joint state of the independent phase noise, and comparing the posterior log likelihood values to select the optimal result to output. The above method can significantly improve the phase noise suppression performance of the MIMO system with a plurality of co-reference channels, thereby providing support for improving the system capacity by using MIMO technology.

Abstract: A co-frequency co-time full duplex (CCFD) signal receiving method includes: taking the sent baseband signal as the self-interference reference signal, reconstructing self-interference, and then performing primary self-interference cancellation on the received signal; processing, by using a timing synchronization loop, the signal after the primary self-interference cancellation, realizing timing recovery at the optimal sampling point of the useful signal through resampling a, and controlling resampling b1 and resampling b2 after performing low-pass filtering on the timing error signal in the timing synchronization loop, to recover the optimal sampling points of the self-interference reference signal and the received signal respectively; and performing joint self-interference cancellation and equalization on the resampled self-interference reference signal and the resampled received signal, and receiving the useful signal through signal demodulation.

Abstract: Disclosed are a labeling method and apparatus for named entity recognition of a legal instrument. The method includes steps: step S1: acquiring a legal text, and transforming the legal text into an index table; step S2: outputting a sentence feature encoding result; step S3: performing training and prediction; step S4: obtaining a set; step S5: obtaining a multi-head score transfer matrix; step S6: obtaining a score transfer matrix corresponding to the legal text; step S7: determining a recognized nested entity; and S8: constructing an entity labeling template by using the recognized nested entity. According to the present disclosure, a user tries to complete recognition of nested entity labeling by changing an input of the BERT model, and a multi-head selection matrix labeling thought of the present disclosure is used to relieve the difficulty in recognizing a long text and a nested entity in an NER task to a larger extent.

Abstract: A method for enhancing vacuum tolerance of optical levitation particles includes steps of: (1) turning on a trapping laser to form an optical trap, loading the particles to an effective capture region of the optical trap, and collecting scattered light signals; (2) turning on the preheating laser, and directing a preheating laser beam to the captured particles; (3) adjusting a power of the preheating laser until a particle heating rate is larger than a heat dissipation rate; (4) turning on the vacuum pump, and stopping evacuating when a vacuum degree is greater than a vacuum inflection point of a first reduction of the effective capture region of the optical trap; and (5) turning off the preheating laser when the scattered light signals collected by the photodetector no longer changes. The present invention improves a stable capture probability of the particles in high vacuum environment.

Abstract: The present invention discloses an edge calculation-oriented reparametric neural network architecture search method, including the following steps: S1: designing linear operators and multi-branch block structures; S2: constructing a hypernetwork by stacking the multi-branch block structures; S3: training the hypernetwork through a gradient-based first-stage search algorithm; S4: deleting redundant branches in the hypernetwork to construct an optimal subnetwork; S5: converting the multi-branch optimal subnetwork into a single-branch network; and S6: completing task reasoning by using the single-branch network. The method is used to search the neural network structure capable of performing reparameterization, and ensures the reasoning real-time performance and the high efficiency of model operation while ensuring the reasoning precision.

Abstract: The present invention discloses a cloud-side collaborative multi-mode private data circulation method based on a smart contract, including: S1, a system is initialized; S2, the original data are encrypted into private data, an encryption certificate z? for storage is generated, and z? includes metadata and a data certificate key?; S3, the DO calls a smart contract program to realize uplink of the encryption certificate z? and releases z? to a block chain through a smart contract, wherein the smart contract is open to all user accounts; S4, rapid data circulation is realized: when DO releases the data certificate, DU has been identified, a DU's account IDDU is set through an access policy, the DU obtains an encryption key for data access by executing a smart contract and a key algorithm, private data are obtained through metadata and decrypted to obtain a plaintext; and S5, the data circulation is confirmed.

Abstract: Provided is a cross-departmental decision support system for early diagnosis of a chronic kidney disease based on knowledge graph, which comprises a patient information model building module, a patient information model library storage module, a knowledge graph association module, a knowledge graph inference module and a decision support feedback module. According to the present application, by constructing a patient information model and utilizing an OMOP CDM standard terminology system, patient electronic medical record data is constructed into a patient information model with unified concept coding and unified semantic structure; making full use the advantages of semantic technology in data interactivity and scalability, so that the system has better adaptability and scalability to heterogeneous data in different hospitals.

Abstract: The present invention discloses an automatic pancreas CT segmentation method based on a saliency-aware densely connected dilated convolutional neural network. Under a coarse-to-fine two-step segmentation framework, the method uses a densely connected dilated convolutional neural network as a basis network architecture to obtain multi-scale image feature expression of the target. An initial segmentation probability map of the pancreas is predicted in the coarse segmentation stage. A saliency map is then calculated through saliency transformation based on a geodesic distance transformation. A saliency-aware module is introduced into the feature extraction layer of the densely connected dilated convolutional neural network, and the saliency-aware densely connected dilated convolutional neural network is constructed as the fine segmentation network model. A coarse segmentation model and the fine segmentation model are trained using a training set, respectively.

Abstract: The present disclosure belongs to an identity authentication technology in network security field, and relates to a lightweight identity authentication method. The method utilizes lightweight operations of the physical unclonable function, Hash operation, XOR operation, etc.

Abstract: Disclosed are a parallel method and device for convolution computation and data loading of a neural network accelerator. The method needs two input feature maps and two convolution kernel cache blocks, and sequentially stores the input feature maps and 64 convolution kernels into cache sub-blocks according to a loading length, so as to execute convolution computation and simultaneously load data of a next group of 64 convolution kernels.

Abstract: The present invention relates to the technical field of network communication, in particular to a method and system for controlling ID identifier network mobility based on a programmable switch. The system includes mobile terminal nodes, mobile access points, programmable switching nodes and control nodes, wherein the control nodes include local control nodes and a global control node, the mobile terminal nodes are connected and communicated with the mobile access points through wireless data links, the mobile access points are connected and communicated with the programmable switching nodes through wired data links, and the programmable switching nodes, the local control nodes and the global control node are connected and communicated in order through control links.

Abstract: Disclosed is a method for automatically compressing multi-task oriented pre-trained language model and a platform thereof. According to the method, a meta-network of a structure generator is designed, a knowledge distillation coding vector is constructed based on a knowledge distillation method of Transformer layer sampling, and a distillation structure model corresponding to a currently input coding vector is generated by using the structure generator; at the same time, a Bernoulli distribution sampling method is provided for training the structure generator; in each iteration, each encoder unit is transferred by Bernoulli distribution sampling to form a corresponding coding vector; by changing the coding vector input to the structure generator and a small batch of training data, the structure generator and the corresponding distillation structure are jointly trained, and a structure generator capable of generating weights for different distillation structures can be acquired.

Abstract: Provided is a patient data visualization method and system for assisting decision making in chronic diseases. According to the present application, a management data model diagram of a patient on a hyperplane is constructed by constructing a chronic disease knowledge graph, and combining static data and dynamic data of the patient, and then the management data model diagram is projected onto a two-dimensional plane. The difference of the Euclidean distance between features of a patient information model on a two-dimensional plane graph from the distance of standard features is compared, and a management plan is generated and recommended in combination with path node concepts and an attribute relationship between the concepts.

Abstract: Disclosed are an automatic compression method and platform for a pre-trained language model based on multilevel knowledge distillation. The method includes the following steps: step 1, constructing multilevel knowledge distillation, and distilling a knowledge structure of a large model at three different levels: a self-attention unit, a hidden layer state and an embedded layer; step 2, training a knowledge distillation network of meta-learning to generate a general compression architecture of a plurality of pre-trained language models; and step 3, searching for an optimal compression structure based on an evolutionary algorithm.

Abstract: The present invention relates to the technical field of network communication, in particular to a method and system for controlling ID identifier network mobility based on a programmable switch. The system includes mobile terminal nodes, mobile access points, programmable switching nodes and control nodes, wherein the control nodes include local control nodes and a global control node, the mobile terminal nodes are connected and communicated with the mobile access points through wireless data links, the mobile access points are connected and communicated with the programmable switching nodes through wired data links, and the programmable switching nodes, the local control nodes and the global control node are connected and communicated in order through control links.

Abstract: The present invention discloses a speech emotion recognition method and system based on fused population information. The method includes the following steps: S1: acquiring a user's audio data; S2: preprocessing the audio data, and obtaining a Mel spectrogram feature; S3: cutting off a front mute segment and a rear mute segment of the Mel spectrogram feature; S4: obtaining population depth feature information through a population classification network; S5: obtaining Mel spectrogram depth feature information through a Mel spectrogram preprocessing network; S6: fusing the population depth feature information and the Mel spectrogram depth feature information through SENet to obtain fused information; and S7: obtaining an emotion recognition result from the fused information through a classification network.

Abstract: Provided is a time series deep survival analysis system combined with active learning. The system includes: a data collection module, an active learning module, and a time series deep survival analysis module; the data collection module is used for obtaining survival data of objects to be analyzed; combined with an active learning method, the active learning module selects a part of right censored data to label a survival time; and the time series deep survival analysis module constructs a time series deep survival analysis neural network model, and takes uncensored data and right censored data as model inputs, so as to obtain survival time prediction results of the objects to be analyzed. The present application can make full use of the right censored data in the survival data and time series features.