105 results about "Mutual learning" patented technology

The invention discloses a multimedia network instructional system which comprises an administrator side, a teacher side and N student sides, wherein the administrator side is used for carrying out instructional resource administration and training business administration; the teacher side is used for performing an instructional activity and carrying out a learning supporting activity; the student sides are used for carrying out course learning and communicating with a teacher and other students; the administrator side feeds back information to the teacher side or monitors the work of the teacher; the teacher side provides or correct an instructional document and an instructional resource for the administrator side; the administrator side implements a service for the student sides, and provides a course constructing module for the student sides; the student sides can mutually learn and communicate with the teacher side, and provide a course feeding-back module for the teacher side; the student sides can learn and communicate with one another, and complete a course implementing module. The multimedia network instructional system can support instructional activities of on-line learning of the students, instruction giving of the teacher and examinations, provides personalized services for users at different levels and with different demands, and provides services of learning other languages by a mother language for the users.

The invention discloses an unsupervised cross-domain pedestrian re-identification method based on clustering, and belongs to the field of computer vision. The method specifically comprises the following steps: (1) establishing a neural network for pre-training; (2) inputting the labeled data in the source domain into a pre-training network, and performing supervised pre-training; (3) building a double-flow mutual learning network framework for fine adjustment; (4) training a double-flow mutual learning framework, and outputting a pedestrian re-identification model suitable for a target domain.Unsupervised cross-domain pedestrian re-identification is realized by utilizing a labeled source domain data set, a label-free target domain data set and a mutual learning network framework; according to the method, a self-attention mechanism and global joint pooling operation are introduced, a new loss function, namely joint flexible optimization loss, is provided, and a clustering method more suitable for open set data is selected, so that the model performance is obviously improved.

The embodiment of the invention provides a label distribution method and a device. The label distribution method comprises the steps that a first router device transmits a first boarder gateway protocol (BGP) updating message including a first virtual private network (VPN) instance identifier, the first router device receives a second BGP updating message including a second VPN substance identifier and first RT, the first router device receives a second BGP updating message including the second VPN instance identifier and the second RT, and the first router device allocates the first VPN instance identifier to the second VPN instance identifier after the first RT and the second RT are confirmed to be identical; and the first router device transmits a third BGP updating message including the first VPN instance identifier to a second router device. According to the technical scheme, through the mutual learning of the VPN instances, the identifier is allocated to the VPN instances, so a layer 3 virtual private network (L3VPN) service can be identified by a service end node to be sourced from which VPN instance, so that the operation administration and maintenance (OAM) problem of the L3VPN can be solved.

The invention discloses a method for spectrum allocation in a distributed cognitive wireless sensor network on the basis of Q study. The average throughput and the average energy efficiency ratio of the maximized distributed cognitive wireless sensor network are taken as objects, nodes in the cognitive wireless sensor network study from one another according to peripheral authorization frequency spectrum state change and allocation strategies of other nodes and finally adapt, the process is mapped into a distributed multi-agent Q study process, and the best spectrum allocation strategy is approximated by executing the best response Q study iterative algorithm in a timing sequence alternation mechanism. The method has the characteristics that the convergence speed is high, the calculation complexity is low, and the average throughput and the average energy efficiency ratio of the cognitive wireless sensor network are improved.

The invention provides a method and system used for sharing notes and based on a network and a cloud server. The method includes the following steps that the cloud server receives and stores at least one note corresponding to text messages and sent by a mobile terminal, the text messages and position messages of the at least one note in a corresponding document of the text messages; the cloud server receives note inquiry requests specific to the text messages of users; the cloud server inquires to obtain corresponding notes according to the note inquiry requests; and the cloud server sends the corresponding notes to the users. By means of the method of an embodiment, the users can share their own notes specific to the text messages through the cloud server and can inquire other notes specific to the same text messages, and therefore the method and system can help the users to study mutually and understand the text messages, especially characters, icons, formulas and the like specific to science and technology documents, can improve reading pleasure of the users, and improves user experience.

The invention provides a cloud robot collaborative learning method based on hybrid enhanced intelligence. The method comprises the following steps of S1, adopting a meta-learning method of NTP (NeuralTask Programming) for dividing a large task recursion into simple sub-tasks for teaching robots; S2, achieving robot action imitative learning based on a human-computer interaction technology, and teaching the robots about skills; S3, utilizing a self-organizing incremental neural network (SOINN) to gather the scattered robot skills for sharing use. According to the method, multiple robots learnhuman skills separately, then upload learned information into a server and share the information in the server for training and adjustment. Not only can the learning time be greatly shortened, but also the diversity of tasks can be expanded. The robots conduct collaborative learning and transmit experiences to one another through a network (or a cloud robot), that is to say, the robots can learn from one another.

The invention provides an intelligent configurable programming toy of the Internet of Things and belongs to the field of the Internet of Things, artificial intelligence, education and toys. The intelligent configurable programming toy mainly comprises a configuration programmer, a central processer, a memory, a data communication module, a multifunctional sensor and a multi-point actuating mechanism. An open platform is applied, so that devices of the Internet of Things are integrated, entertainment characters of different toys are configured into a whole, a user performs configuration by choosing hardware with hands and even performs programming with the hands, and the toy is arranged into an expected action, from and interaction manner. The operational ability, the learning ability, the cognitive ability, the imaginative ability and the like of the user can be improved. By the aid of the Internet of Things, storage data of the intelligent toy can be shared, users can exchange arrangement methods and contents of the intelligent toy and learn from each other, the forms and actions of the toy are ever-changing, the contents are rich and colorful, other devices of the Internet of Things can be controlled, an interaction function of the toy and the Internet of Things is realized, and the intelligent toy is integrated into people's life, so that the people's life is improved.

The invention provides a learning method between remote controllers, a set-top box and an intelligent television. The method is applied to the intelligent television and comprises the following steps of: obtaining a first control signal of a first remote controller, and obtaining the first mapping relation between the first control signal and a first function; obtaining the second mapping relation between a key identifier of a second remote controller and the first function; establishing the third mapping relation between the key identifier and the first control signal according to the first mapping relation and the second mapping relation; and sending the third mapping relation to the second remote controller. A mapping relation can be established between the control signal from the first remote controller and the key identifier of the second remote controller on the intelligent television, so that the corresponding key on the second remote controller can send the control signal to control the television; and since the learning process is finished by means of the display function of the intelligent television, the learning process is suitable for all people, and the mutual learning efficiency between the remote controllers is improved.

The invention relates to a search engine user information demand satisfaction evaluation method capable of integrating multiple views and semi-supervised learning. The method is divided into the following six stages: preprocessing data, training a subview satisfaction model, distributing a dummy tag for unlabeled data, training a user satisfaction model based on the multiple views and the semi-supervised learning, and carrying out evaluation. Through a semi-supervised learning method, a small quantity of labeled data and a great quantity of unlabeled data are used for improving the performance of an evaluation model, and a multi-view learning method is imported to overcome the problem that a traditional single-view based semi-supervised learning method is always caught in local optimum. The search engine user information demand satisfaction evaluation method has the beneficial effects: (1) under the condition of the small quantity of labeled data, the search engine user information demand satisfaction can be effectively evaluated; (2) the small quantity of labeled data and the great quantity of unlabeled data can be used for improving the evaluation performance of the user satisfaction model; and (3) a search process of the user can be independently described from angles of behaviors and time, and the model can be prevented from being caught into the local optimum through mutual learning.

The invention discloses a label automatic definition method based on image content, and belongs to the field of image processing and natural language processing, and particularly relates to an automatic image labeling method based on image mutual information analysis. For problems in the prior art that accurate labeling cannot be realized, and the model is influenced by wrong annotation information, the technical scheme is that the method comprises the steps of firstly generating a pre-training model, then crawling pictures in the Internet, putting the crawled pictures into the generated pre-training model to output corresponding labels, assessing the labels output by the pre-training model through a space vector, and finally updating the model and a vector space according to obtained newdata. According to the method, the model can be automatically updated according to the network data, the annotation is more accurate through mutual learning of the annotation information between the images, and the model can be prevented from being influenced by the wrong annotation information through a special label evaluation module.

The invention discloses a network model training system, method and device based on parameter sharing and a medium. The system comprises a first parameter module and a second parameter module, the first parameter module is used for obtaining a first model parameter formed when a first local end trains a network model and synchronizing the first model parameter to a block chain, the second parameter module is used for obtaining a second model parameter requested by a second local end from the block chain and synchronizing the second model parameter to the second local end, and the second modelparameters are used for the second local end to train the network model. According to the present invention, the model parameters are shared through the block chain, and the large-scale application and the mutual learning of a network model are facilitated. Due to the fact that exchange between the network model training set and the test set does not need to be carried out between the local ends,leakage of sensitive data is avoided on the basis of efficient propagation of network model parameters. The method, the system, the device and the medium are widely applied to the technical field of artificial intelligence.

The invention discloses a robot ground classification method and system in a dynamic environment, which can realize real-time and accurate ground type prediction in the dynamic environment by utilizing a vibration signal, and mainly comprises six steps of signal acquisition and segmentation, feature extraction, classifier prediction, classification result correction, classifier updating and classification result output. The vibration view is divided into the time domain and the frequency domain, and a mutual learning mechanism is realized; the classification result is corrected by using the ground sequence time correlation, so that the precision of each domain classifier is improved; meanwhile, two mechanisms of external divergence and internal divergence are introduced, a new pseudo-labeling algorithm is designed, and the algorithm has great advantages in the aspects of extracting key samples and labeling; based on the pseudo-labeling algorithm, the robot can incrementally update theclassifiers of all domains on line, so that accurate and reliable ground classification in a dynamic environment is realized.

The invention discloses a railway foreign matter intrusion security perception and identification system based on distributed optical fiber sensing. On the basis of two distributed optical fiber sensing technologies and video linkage technologies, a phase position optical time domain analysis technology is used for monitoring a vibration intrusion signal generated when illegal personnel, vehiclesand construction intrude a high-speed railway, and information, including the position, the event and the like of the intrusion signal, can be given; a Brillouin optical time domain analysis technology is used for monitoring a high-speed railway fence deformation signal generated when large falling rooks and debris flows intrude the high-speed railway, and information, including the position and the like of the intrusion signal, can be given; and when any one or two systems in the above two optical fiber sensing systems give an alarm, cameras arranged on a high-speed railway tower pole are linked, and the monitoring picture of an intrusion event is called. The above three subsystems synchronous work and mutually learn, and layer-by-layer intrusion monitoring is carried out to jointly realize the early warning and the identification of the high-speed railway foreign matter intrusion.

The invention discloses a local mutual learning method based on attention map migration, and aims to improve the annotation precision of a model attention map so as to improve the interpretability of a model. The method comprises the steps: adopting two lightweight models; in the migration process, measuring the distance between attention maps between the models through a Pearson algorithm, and learning the attention maps of the opposite sides mutually along with the training process. In order to avoid the influence of negative migration, region division is carried out on the attention map, a threshold value is set, and the attention map part with high similarity is selected for migration. Compared with the most advanced method, the algorithm provided by the invention has the advantages that the average decline rate of 28.2% and the average growth rate of 29.5% of visual interpretability confidence coefficient are respectively realized, the algorithm is superior to other methods based on input disturbance and class activation mapping to a great extent, and meanwhile, the most responsive region in the sample picture can be labeled, and is not limited to a visual visualization area.

The invention relates to the technical field of mutual learning of students, in particular to a student interaction explanation board. The overall height and the horizontal angle of the student interaction explanation board can be adjusted to improve adaptability, real objects can be simultaneously fixedly clamped, interaction is facilitated through the real objects, the student interaction explanation board comprises four groups of supporting legs, a supporting board, a writing board, four groups of bases, a fixing board, a rotating shaft, a left connecting board, a left threaded pipe, a left threaded rod, a left clamping board, a right connecting board, a right threaded pipe, a right threaded rod and a right clamping board, the supporting legs comprise lower threaded pipes and lower threaded rods, the top ends of the bases are provided with lower placing grooves, lower ball bearings are arranged in the lower placing grooves, a middle placing groove is formed in the middle of the top en of the fixing board, a middle ball bearing is arranged in the middle placing groove, a left placing groove is formed in the right end of the left connecting board, a left ball bearing is arranged in the left placing groove, a right placing groove is formed in the left end of the right connecting board, and a right ball bearing is arranged in the right placing groove.

The invention discloses an online interactive cooperation system, and belongs to the network communication technology. The system comprises a user information management module used for obtaining login information of a user and distributing corresponding operation authority and identification color through an authority management module according to the login information; a local editor module, used for a user to edit the content of the local editor module; a synchronous editor module, used for editing content by a user with editing permission and synchronously displaying the content to otherusers in real time, and comprising a demonstration mode and a cooperation mode: in the demonstration mode, a demonstrator can demonstrate and explain knowledge points, and others can watch and learn the knowledge points, in the cooperation mode, all people can edit content in the synchronous editor module at the same time, mutual learning and communication can be achieved, a management user can check thinking and practice processes of member users in real time. The problems that interaction is little and manual operation is little in the interaction cooperation process are solved, and therefore the cooperation efficiency is comprehensively improved.

The invention relates to a training and visible light infrared visual tracking method based on an adapter mutual learning model. The method comprises the following steps: S11, constructing a network model; S12, training the whole network model by using the manually labeled visible light thermal infrared data set; S21, obtaining a candidate sample of the current frame; S22, predicting a target position according to the candidate sample; and S23, judging whether the current frame is tracked successfully or not. According to the invention, an adapter mutual learning module is introduced on the basis of multiple adapters (including a mode adapter, a general adapter and an instance adapter) to realize bidirectional cross-modal information transmission, so that the complementary advantages of different feature learning modes in RGBT tracking are fully utilized to further improve the traceability.

The invention discloses a neighbourhood learning meme image segmentation method based on standard cut, mainly solving the problem that mutual learning between pixels is not used in the existing image segmentation method. The neighbourhood learning meme image segmentation method is performed by the steps as follows: firstly, conducting pixel-based two-dimensional random 0 / 1 coding on an image to form initial population and taking the standard cut as an overall measuring standard on the cut result of the image; then cloning individuals in the population according to cloning scale; conducting operations of variation, neighbourhood learning and clone selection in sequence on the cloned individuals; and finally storing an optimum image corresponding to an optimum standard cutting value as the segmentation result according to the standard cutting value in the individuals in the current population, and outputting an optimal image segmentation result according to the requirements of termination conditions. The neighbourhood learning meme image segmentation method is used for segmenting noise-free and noise images, has the advantages of being strong in region consistency, stable in convergence and optimal in overall and can be used for image identification and detection.

The invention discloses an improved sparrow search optimization method for blocking a flow shop scheduling problem, and belongs to the technical field of shop scheduling. The method comprises the steps: employing an integer-real number hybrid coding mode to code a sparrow individual and workpiece production sequence; using a new collaborative initialization strategy to initialize the positions of the sparrows; dividing the sparrow population into a plurality of subgroups for mutual learning; obtaining a good workpiece production sequence solution set, namely a sparrow individual elite solution set, through an improved sparrow individual search strategy; and carrying out local enhancement on the obtained elite solution set, and obtaining an optimal sparrow individual, namely an optimal workpiece machining sequence. The method can solve the problems that the population diversity is reduced and the algorithm is easy to fall into local optimum in the later stage of iteration of the sparrow algorithm; and the processing efficiency of the blockage flow workshop can be effectively improved, and the economic benefits of factories are improved.

The invention discloses a hyperspectral image super-resolution method based on mixed attention network fusion. The method comprises the following steps: acquiring a hyperspectral low-resolution image and a corresponding hyperspectral high-resolution image, and forming training data; constructing a hyperspectral image super-resolution basic model; training the hyperspectral image super-resolution basic model by adopting the training data to obtain a final hyperspectral image super-resolution model; obtaining a to-be-processed hyperspectral image; and processing the to-be-processed hyperspectral image by using the hyperspectral image super-resolution model to complete the super-resolution process of the hyperspectral image. According to the invention, the hybrid attention network is adopted to improve the network performance, mutual learning loss is adopted to ensure that each network has mutual supervision and learning capabilities, and finally, an adaptive integration module is adopted to fuse output images of the hybrid attention network. Therefore, the invention is better in effect, higher in reliability and more excellent in performance.

The invention discloses a teaching and learning optimal sliding-mode control method on the basis of personality coefficient adjustment. Correction rules include increasing personality coefficients ofstudents when the students make progress in mutual learning procedures; decreasing the personality coefficients of the students if the students gain to a certain extent in self-learning procedures. The teaching and learning optimal sliding-mode control method has the advantages that an improved teaching and learning algorithm on the basis of personality coefficient adjustment is proposed and can be used for designable parameters of optimal sliding-mode controllers, accordingly, chattering phenomena of sliding-mode control can be weakened, the performance of sliding-mode control systems can beimproved, and the practical feasibility of the teaching and learning optimal sliding-mode control method can be improved; in addition, an incentive measures for each personality coefficient is set inthe improved teaching and learning algorithm pertinently for the condition of reduction of the search speeds of algorithms in late periods, and the personality coefficients of the students can be corrected when the students make progress in learning procedures; the local search capability of teaching and learning algorithms can be enhanced by the aid of the personality coefficients and incentive measures for the personality coefficients, the global convergence can be improved, the convergence speeds of the algorithms in the late periods can be increased, and prematurity phenomena can be effectively prevented.

The embodiment of the invention provides a license plate character recognition method and a device, and electronic equipment. A license plate picture to be recognized is sent to a deep mutual learningnetwork composed of at least two convolutional neural networks, the convolutional neural networks are used for recognizing characters to be recognized in the license plate picture, and a preliminaryoutput result of the characters to be recognized is determined. And a final output result of the to-be-recognized character is obtained according to the preliminary output result of the to-be-recognized character. According to the method, the device and the electronic equipment, the extracted features are further optimized by combining the deep mutual learning network on the basis of extracting the features by utilizing at least two convolutional neural networks, so that the license plate character recognition accuracy is greatly improved, and the license plate character recognition capabilityis improved.

The invention provides a federal mutual learning model training method for non-independent identically distributed data, which comprises the following steps: S1, sending initial global model parameters to an intermediate client, generating intermediate client model parameters by the intermediate client, and S2, generating edge client model parameters by an edge client by using a local data set; S3, enabling the intermediate client and the edge client to update parameters by using a mutual learning method; s4, uploading a probability prediction value output by the intermediate client model to a server, and updating the global model and the intermediate client model by the server by utilizing a distillation technology; and S5, executing the steps S3-S4 until the models meet convergence conditions to obtain a final intermediate client model, a final edge client model and a final global model, and then broadcasting the final global model to all edge clients by the server. The problems of federated learning communication bandwidth limitation and model generation of the non-independent identically distributed data are solved through a grouping mutual learning and knowledge distillation method.

The invention discloses a big data feature selection method based on an improved bat algorithm. The feature selection is to convert a sample in a high-dimensional space into a low-dimensional space ina mapping or transformation manner, and then delete and select redundant and irrelevant features to further reduce the dimension. The method aims at obtaining feature subsets as small as possible, meanwhile, classification precision is not remarkably reduced, and distribution is not affected. On the basis of analyzing the advantages and disadvantages of the classic feature selection method, a swarm intelligence algorithm is introduced and utilized to optimize feature selection. In view of the fact that the bat algorithm has the advantages of high parallelism, high robustness and high convergence speed, K-based optimization is introduced to overcome the defect that the bat algorithm is prone to being caught in local optimization. And a sub-population division mechanism and a binary difference variation mechanism of the means algorithm. The improved algorithm enhances the ability of mutual learning and efficient information transmission between populations, improves the individual difference and search ability, and avoids premature convergence. And finally, the improved bat algorithm is used for optimizing feature selection and an excellent effect is achieved.

The invention provides an end-to-end long-time speech recognition method. The method comprises the following steps: selecting a corpus as a training data set, and carrying out data preprocessing and feature extraction on voice data in the training data set to generate voice features; constructing an improved RNN-T model fusing an external language model and a long-term speech recognition algorithm, and inputting the speech features into the RNN-T model for training to obtain a trained improved RNN-T model; and taking the trained improved RNN-T model as a teacher model in a mutual learning knowledge distillation algorithm, training a student model in the mutual learning knowledge distillation algorithm by using the mutual learning knowledge distillation algorithm, identifying long-term voice data to be identified by using the trained and verified student model, and outputting a voice identification result. According to the method, the external language model, the long-term speech recognition algorithm module and the RNN-T model are fused, so that the robustness and generalization ability of long-term speech recognition of the model are improved, and the time and space complexity of the algorithm is optimized.

The invention discloses a network multi-language social translation method, a system, a device and a medium. The method comprises the following steps: acquiring voice data of a user in a social network in real time, and identifying the language used by the user; converting the voice data into a text according to the recognized language, and translating the converted text into a target text according to the language used by a social object of the user; and returning the converted text and the target text to the user and the social object of the user. The problem of communication obstacle of people who speak different languages in social communication is solved. The application scenarios are diversified, which can meet the one-to-many communication requirement and effectively improve socialefficiency and help the user and the social object to learn languages from each other.

The invention discloses a distributed database query optimization method based on a multi-ant colony genetic algorithm, and belongs to the internet database technology field. The method comprises thefollowing steps of establishing a distributed database framework; analyzing the query cost of the distributed database, and then updating the ant colony algorithm into a multi-ant colony algorithm; utilizing a smoothing mechanism and a multi-ant-colony mutual learning mechanism to avoid falling into local optimum and premature phenomena, so that the global search capability of the whole algorithmis improved. According to the present invention, the technical problem of improving the query efficiency of the distributed database by adopting the multi-ant-colony algorithm is solved, by introducing the multi-ant-colony algorithm, a learning operator is provided in the algorithm, the sub-ant-colony is mutually learned, the local optimum is prevented, the algorithm performance is improved, and the algorithm can obtain a better global optimum solution.

The invention discloses a text multi-vector representation mutual learning-based spam comment filter method. The method comprises the following steps of: firstly preprocessing a comment text of a training set, and carrying out vectorized representation on a same text by using multiple different text vector representation methods; training multiple different classifiers by adoption of a same classifier training method according to the different vector representation manners; carrying out mutual learning among the classifier by means of a thought of cooperative training, continuously labelling data of an addition set and moving the labelled data into the training set; and finally obtaining a plurality of classifiers, the abilities of which are enhanced. According to the method, the thought of cooperative training is adopted, and automatic labelling can be carried out on residual data through manually labelling a part of data, so that a great deal of manual data labelling work is decreased, the classification ability of the classifiers is enhanced and then the spam comment filter precision is improved.