93 results about "Optimal learning" patented technology

A personalized web page is generated by an Electronic Publishing System (EPS) based on predetermined profiles representative of the user's optimum mode of learning. The user's optimum mode of learning is based upon the Theory of Multiple Intelligences for seven (7) different modes of learning. An algorithm calculates user profiles based upon the learning theory. Alternatively, the user fills out a questionnaire provided by the system to determine the user's optimum mode of learning. The questionnaire elicits answers used to calculate the user's optimum mode of learning. An algorithm calculates the user's profile and encodes the profile as a vector of weights for the seven modes of learning. Document templates are created to define the structure of information to be presented to the user. When a user requests information, the user profile is obtained from the cookie or server and the information is presented to the user based on his / her profile which displays the information in the user's optimum learning mode.

A method of learning that uses a learning assessment process to determine an optimal way in which to deliver an educational course that best suits the optimal learning style of a specific individual. The method of learning may, in a preferred embodiment, be used to provide and administer a web-based personal financial management course.

The invention discloses a facial expression recognition method based on a convolutional neural network. The method includes facial expression image data set preprocessing, construction of an improvedconvolutional neural network, weight optimization and training, and classification processing of facial expressions. The method introduces continuous convolution into a conventional convolutional neural network to obtain the improved convolutional neural network, the improved convolutional neural network adopts a small-scale convolution kernel to perform feature extraction, so that extracted facial expression features are more precise, two continuous convolution layers also enhance a nonlinear expression capability of the network, in addition, the convolutional neural network and an SOM neuralnetwork are cascaded to form a pretraining network to perform pre-learning, neurons with an optimal learning result are used for initializing the improved convolutional neural network, and the methodprovided by the invention can effectively improve facial expression image recognition precision.

The invention relates to a traffic flow prediction method based on a quick learning neural network with double optimal learning rates. The method comprises the following steps of: normalizing m continuous traffic flow historical data which serve as the input of a prediction network; initializing weights and stretch and shift factors of a wavelet basis function by using a three-layer neural network, wherein the shift factors and transfer factors of the wavelet basis function employ a first learning rate, and network weights employ a second learning rate; providing a learning rate array, and performing network training of the double optimal learning rates; and outputting values of a current moment to first (m-1) periods, which serve as the input of a trained network, performing reverse normalization, and thus obtaining a prediction value of a traffic flow at a next moment of the current moment. The method has the advantages that the first learning rate and the second learning rate employ the optimal learning rates during network training at each time, quick network training can be realized, and high-accuracy prediction of the traffic flow is realized.

The embodiment of the invention discloses an optimum study scheme pushing method and system. The method comprises the steps that subject knowledge and skill maps are structured; all knowledge points in the subject knowledge and skill maps are subjected to scoring according to knowledge point scoring methods to obtain the score of each knowledge point; the optimum knowledge point is obtained according to the score of each knowledge point; the questions corresponding to the optimum knowledge point are obtained from question banks to form an optimum study scheme, and the optimum study scheme is pushed to a user. According to the optimum study scheme pushing method and system, through the construction of the subject knowledge and skill maps, the problems of standardization, visualization and priority of the knowledge points are solved, and the optimum study scheme is formed through the fact that the significance of each knowledge point in the subject knowledge and skill maps is accessed; the formed optimum study scheme is pushed to a student and is used for solving the defect that study can not be carried out according to the real studying condition of the student and the problem that the study of the strudent in traditional computer aided education is lack of pertinency, and therefore the student can obtain the largest study benefit in limiting time.

The invention relates to an adaptive learning path planning system based on reinforcement learning. The system comprises an environmental simulation module, a strategy training module and a path planning module. In the whole process, the ability value of the student at each moment is obtained according to the improved project reflection principle; based on a Markov decision process, a complex learning environment is simulated, a reinforcement learning algorithm is reasonably applied to be combined with a student historical learning track offline training path planning strategy, and finally, alearning path is adaptively planned for the student online according to the trained strategy. Finally, based on the idea of reinforcement learning, the complex scene learned on the online education platform is constructed in the framework of the Markov decision process, the purpose of improving the efficient obtaining capability is achieved, continuous recommendation of learning resources is provided for students, the optimal learning path is planned, and therefore the learning effect and learning efficiency of the learner are improved.

The invention discloses a rapid area-of-interest detection method based on depth kernelized hashing and mainly aims to solve problems of low detection precision and low efficiency existing in an image area-of-interest detection method in the prior art. The method comprises steps that in a training process, to-be-trained data sets are classified into multiple categories which are inputted to a depth kernelized hashing supervised learning network framework to acquire corresponding hashing codes, fine tuning of network parameters is carried out according to a label information matrix till the optimal learning result is realized; in a test process, inputted test images are pre-processed, and binary codes are acquired through the trained depth kernelized hashing supervised learning network framework; the area-of-interest position of the image is determined according to a decision function and is marked, and area-of-interest detection and identification are accomplished. Through the method, area-of-interest detection analysis performance of the image can be effectively improved, an area-of-interest detection rate and system framework operation efficiency are improved, and the method can be applied to rapid detection and identification of breast image lumps.

The invention provides a lazy learning-based self-adaptive robustness forecast control method of blast furnace molten iron quality, and relates to the technical field of automatic control of blast furnace melting. The lazy learning-based self-adaptive robustness forecast control method comprises the steps of determining controlled quality and control quantity; acquiring historical production input-output measurement data of a blast furnace to construct an initial database; constructing an inquiry regression vector, and determining abnormal data; inquiring a similar leaning subset from the database, selecting an optimal learning subset, and processing the abnormal data; building a forecast model by taking the optimal learning subset; calculating index reference track of the molten iron quality, constructing a forecast control performance index to obtain an optimal control vector; and sending the optimal control vector to a bottom-layer PLC system, adjusting an execution mechanism, acquiring a new group of blast furnace measurement data, pre-processing the data, and updating the database. According to the method provided by the invention, the influence of the input-output interference can be effectively suppressed, the influence of the abnormal data is overcome, the blast furnace molten iron quality is stabilized to a value near to an expected value, and stable running, good quality and high yield of the blast furnace are facilitated.

The present invention discloses a method and system for deep learning automation parameter adjustment based on the Keras. A preset parameter initial value list is employed to perform adjustment of batch size and training epoch of a deep learning model; after an optimal batch size and a training epoch parameter value are found, the weight initialization method is adjusted; after the optimal weight initialization method is obtained, the algorithm learning rate is adjusted according to the optimal batch size and the training epoch; after the adjustment of the algorithm learning rate is completed, an optimal learning rate parameter value is obtained, an activation function is adjusted, and the optimal activation function type of the model is found; and according to the optimal batch size, the optimal training epoch, the optimal weight initialization method, the optimal learning rate and the optimal activation function, the Dropout parameter is adjusted to find out an optimal Dropout parameter value, and the training epoch is subjected to fine tuning to obtain final parameter combination. The method and system for deep learning automation parameter adjustment based on the Keras can greatly reduce the deep learning model parameter adjustment and encoding workload and improve the efficiency of the deep learning development.

The invention discloses a twin convolutional neural network face recognition algorithm introducing a perception model. Firstly, a twin convolutional neural network is introduced as an overall networkstructure model, so that external interference can be effectively reduced, over-fitting is avoided; a sensing model is added to a twin convolutional neural network structure on the basis, the networkwidth is increased, the effect of information cross-channel connection is achieved, the adaptability of the network to the scale is improved, and meanwhile richer feature extraction can be achieved bymeans of the advantage of hardware dense matrix optimization. The whole training process is assisted by a loop learning rate strategy optimization algorithm, so that the optimal learning rate is easyto find, the model convergence can be accelerated, the network performance is improved, and high-precision face recognition under a non-limiting condition is effectively realized. The algorithm is simple in structure, has high robustness for face recognition under the non-limiting condition, can improve the training speed and improve the face recognition accuracy, and is suitable for small-scaledata sets.

The invention relates to a network instruction detection method based on transfer learning and integrated learning. The method comprises the following steps: 1) training and generating multiple independent intrusion detection machine learning judgment models according to standard data; 2) forming an integrated judgment model for the intrusion detection according to the integrated judgment; 3) producing various network traffic data containing attack behaviors by using a network attack tool, and screening the network traffic data by using the integrated judgment model; and 4) taking the screenednetwork traffic data as the attack data, performing transfer learning on the machine learning model, thereby finally obtaining the optimal learning judgment model. Compared with the prior art, the network intrusion detection method disclosed by the invention has the advantages of saving the training time, accelerating the convergence time and high accuracy.

The invention relates to a BNM-based (Bayesian network model-based) fishery forecasting method. The method includes the steps of discretizing historical ocean information datasets of fishery environment; establishing a table of conditional probability between Bayesian network structure charts and Bayesian network nodes; calculating a posterior probability distribution formula of fishery by the Bayesian network structure chart obtained by optimal learning algorithm; and forecasting the fishery by the obtained posterior probability distribution formula. The function of rapidly forecasting for the fishery can be realized by the BNM-based fishery forecasting method.

The invention discloses a student learning situation analysis system and a use method thereof, comprising a knowledge item bank. The topic exercise module obtains examples from the knowledge item bankfor users to test knowledge points. The analysis module is used to calculate the correct answer rate of examples in the exercises module, and analyze the mastery degree of each knowledge point of users. The learning planning module is connected with the analysis module to plan the best learning path of each knowledge point according to the mastery degree of each knowledge point; Under the controlof the learning planning module, the test module obtains the items from the knowledge bank and tests the learning effect of the feedback learners. The invention enables students to grasp knowledge points more suited to their own level when learning and practicing, formulates suitable learning paths, and improves learning efficiency. Through the analysis module to analyze the user 's knowledge points to grasp the situation, positioning the appropriate starting point and learning path, to improve the learning efficiency of students, to avoid the title is too difficult or too simple, so that thelearning effect is the best.

The invention discloses a power circuit component optimization method based on an orthogonal learning particle swarm, and belongs to the power electronic technology and the field of computational intelligence. An orthogonal learning particle swarm optimization with a mutation strategy is used for carrying out optimization on an optimal component design of a power electronic circuit. Firstly, a method of generating a new optimal learning object based on an orthogonal combination mode is designed, and is used for mining information of a historical optimal solution of a particle individual and information of a globally-optimal solution of a swarm in the orthogonal learning particle swarm optimization, and combining a learning object which can guide particles to develop in a better direction, secondly, a mutation operator which can improve diversity of the orthogonal learning particle swarm optimization is designed, and the defect that the orthogonal learning particle swarm optimization easily falls into local optimum is overcome. All components of the power electronic circuit serve as variables needing to be optimized and are coded into individuals of the orthogonal learning particle swarm optimization, optimization is carried out on values of the components of the power electronic circuit through specific optimization processes such as update of the speed, update of the location, mutation operation and update of the optimal learning object of the orthogonal learning particle swarm optimization, and the power circuit component optimization method based on the orthogonal learning particle swarm has important application value in the existing large-scale circuit design and optimization field.

Embodiments of the invention disclose a robust self-adaptive semi-supervised image classification method and device, equipment and a medium. The method comprises the following steps of: integrating robust self-adaptive embedded label propagation and self-adaptive weight construction into a uniform semi-supervised learning framework, and minimizing an embedded characteristic and embedded label-based reconstruction error; changing an original predicted label set into a preset label space by utilizing robust projection, so as to classify each label in the original predicted label set; decomposingan original data set into a denoising result expression item and a noise fitting error item, and carrying out self-adaptive weight construction and self-adaptive label propagation on the denoising result expression item; and integrating a regressed label approximate error item into the semi-supervised leaning framework, and carrying out combined optimal learning to obtain a projection classifiermatrix. According to the method, the image classification ability and pattern classification prediction ability are effectively enhanced, and benefit is brought to enhance the image classification correctness.

The invention provides a ship course trajectory tracking design method based on self-adaptive fuzzy optimal control. For a ship course nonlinear discrete system, an optimal control problem of the shipcourse discrete nonlinear system is solved by applying a fuzzy optimal learning self-adaptive algorithm, the controller energy consumption is effectively reduced, the steering engine abrasion is lowered, and the course tracking speed and precision are improved.

The invention provides a learning route dynamic planning method and device. The method comprises the following steps: comprehensively evaluating the history learning process, knowledge point mastery degree and learning ability of students; recommending an optimal learning path to the students according to the knowledge graph and the comprehensive evaluation result; enabling a pre-trained AI teacher module to give prediction of mastery situation probability distribution of students for examination questions; and recommending proper after-class exercises to the students according to the prediction result. According to the learning route dynamic planning method and device, the optimal learning route is planned from huge teaching resources according to the ability of students, reasonable questions are pushed, and learning of related knowledge points is completed.

The invention discloses an online learning test question bank management system which comprises a subject question bank used for storing the test questions of the subject knowledge points; a question combination module which is used for obtaining the test questions from the subject question bank and carrying out question combination according to the analysis result of the knowledge mastery degree of the student users by an analysis module; the analysis module which is used for analyzing the mastery degree of the tester to each knowledge point; a learning planning module which is used for planning an optimal learning path of a student user knowledge point; a test module which is used for acquiring the questions from the question combination module for targeted test and sending test results to the score database; a user terminal which is used by the student user to make a test request, answer and feed back a result to the system. The invention also provides an online learning test question bank management method. According to the invention, the mastery condition of each knowledge point by the student user can be statistically analyzed, the question combination for testing the knowledge point is determined, and the knowledge point learning path of the student user is planned.

This present invention relates to a method of using data redundancy to improve the rolling mill thickness' controlled precision. It considers computer and thickness testing equipment as the thickness controlling equipment. This invention comprises the steps as follow. Intercalate multifunctional testing equipment which is set at the exit of rolling mill as the as thickness testing equipment together to collect data about thickness. Send the data to PLC and process computer, then judge the two group data. If the thickness testing equipment is abnormal, compare the data from multifunctional testing equipment and thickness controlling equipment to get dispersion from which we can acquire the optimal study coefficient. Modify the designed model with the coefficient. This invention enhances model's auto-adapting ability and advances initialization.

The fault diagnosis method for the refrigeration system comprises the following steps of through a simulation water chilling unit fault experiment, carrying out acquisition and processing to obtain training unit data and test unit data; Setting the node number and the layer number of the deep neural network; Establishing a deep neural network model, and determining a topological structure of the deep neural network, the topological structure comprising an input layer number, a weight value and a threshold value of the deep neural network; Determining the number of training steps of the deep neural network; Training in the deep neural network model by applying the training group data to obtain a fault diagnosis model; Training the deep neural network by adopting a small-batch momentum random gradient descent method; Setting a learning rate of the deep neural network; Calculating a loss function C; Optimizing the learning rate by adopting a simulated annealing algorithm; Obtaining an optimal learning rate; Training ending conditions are met, and a trained fault diagnosis model is obtained; And performing fault diagnosis on the test group data in the step S2 by using the trained faultdiagnosis model to obtain a fault diagnosis result.

The invention relates to an RBF neural network-based image watermark embedding and extraction method and device. According to the method, an original watermark image is encrypted, a carrier image is trained by using an optimized RBF neural network, and the encrypted watermark image is embedded into the original watermark image, wherein the optimal learning rate is set in an RBF neural network algorithm to realize self-adapting adjustment of a weight of the algorithm. The encrypted watermark image embedded in the carrier image is extracted through the optimized RBF neural network algorithm and a margin calculation, and the encrypted watermark image is subjected to recovery to obtain the original watermark image. The device comprises a scrambling module, an embedding module and an extraction module. Through the adoption of the method and the device, a contradiction between imperceptibility and robustness of the watermark image is balanced.

The invention discloses a deep belief network parameter optimization method based on an artificial bee colony and a deep belief network parameter optimization system thereof. The method comprises thesteps that a deep belief network model is constructed; and the learning rate of the deep belief network model acts as the problem parameter, the energy function of the deep belief network model acts as the objective function, iterative optimization is performed on the learning rate of the deep belief network model by using the artificial bee colony algorithm to find the optimal learning rate of the deep belief network model and the network parameter under the condition of the minimum energy value, and the optimal learning rate of the deep belief network model and the network parameter under the condition of the minimum energy value act as the initialization parameters of the deep belief network. Deep belief network parameter optimization is performed by using the artificial bee colony algorithm so that the speed of convergence of the learning rate and the network accuracy can be enhanced; and the optimal learning rate and the network parameter under the condition of the minimum energyvalue act as the initialization parameters of the deep belief network so that the fitting degree can be enhanced. The deep belief network parameter optimization method based on the artificial bee colony and the deep belief network parameter optimization system thereof can be widely applied to the field of data mining.

A method is provided herein that can adaptively determine the optional review cycle for different people and for different subjects. The method models the material as a number of learning focuses and the process of material's shifting from short-term to long-term memory as transition through a series of memorization states. Then, for each memorization state, the method performs reviews by evaluating the learning focuses and gathers relevant statistics about the evaluation result to dynamically determine whether the material has now in a next memorization state or when to conduct the next review on what learning focuses.

A method, apparatus and computer program for detecting occurrence of an anomaly. The method can exclude arbitrariness and objectively judge whether a variation of a physical quantity to be detected is abnormal or not even when an external environment is fluctuating. The method includes acquiring multiple primary measurement values from a measurement target. Further, calculating and a reference value for each of the multiple primary measurement values by optimal learning. The method further includes calculating a relationship matrix which indicates mutual relationships between the multiple secondary measurement values. Further the method includes calculating an anomaly score for each of the secondary measurement value which indicates the degree of the measurement target being abnormal. The anomaly score is calculated by comparing the secondary measurement value with a predictive value which is calculated based on the relationship matrix and other secondary measurement values.

The invention discloses a knowledge point system teaching method. The knowledge point system teaching method comprises the following steps: S1, constructing a knowledge point system, namely, aiming atthe learning subjects, the learning content, the learning objectives or the learning stage, starting from the integral knowledge content and the knowledge structure, constructing the complete knowledge point system; S2, constructing a relation graph among the knowledge points, and marking the optimum learning path of one certain knowledge point; S3, establishing the mapping relation between the learning content and the knowledge points; S4, enabling students to learn the knowledge points according to the established mapping relation; and S5, detecting the knowledge content, and checking the learning achievement. With the teaching method provided by the invention, a personalized learning solution is independently and intelligently provided for the students; according to the mastering levelfor the knowledge points of the students, by utilizing the 'intelligent mapping relation between learning materials and knowledge points', a learning solution suitable for the current learning stageof the students is provided for the students, the most suitable learning materials are provided for the students, and thus the learning efficiency is improved.

The invention discloses a data set classification learning algorithm automatic selection system and method, and belongs to the technical field of machine learning. The method aims at solving the problems that a selection mode of a learning algorithm involved in existing data processing does not have universality, and if attempts are conducted one by one, the calculated amount is too large. The system comprises a training feature selection module for selecting each classification problem data set and processing each classification problem data set to obtain corresponding classification meta-knowledge; a selector module which is used for selecting effective features from the classified meta-knowledge as meta-features, forming a selector training set and training a meta-knowledge training selector; an algorithm selection module which is used for processing the to-be-processed data set to obtain to-be-processed meta-features, analyzing by adopting a meta-knowledge training selector to obtain an optimal learning algorithm of the to-be-processed data set; and a knowledge base module which is used for obtaining an algorithm selection training set comprising a one-to-one correspondence relationship between different classification problem data sets and corresponding learning algorithms. The method can predict the optimal learning algorithm for the data set.

The invention discloses a three-freedom-degree space mechanical arm motion planning method based on a learning generalization mechanism. Self motion characteristics obtained through training of a mechanical arm are adopted to build a motion sample database, through intelligent screening, an optimal learning sample is selected for learning, the mechanical arm generalizes motion characteristics of anew space target, and finally, according to the generalized space characteristics, joint motion of the mechanical arm is planned; the mechanical arm can carry out screening learning on the training motion of the mechanical arm so as to generalize motion of the new target, and capacity of cognition and learning generalization of the mechanical arm can be achieved; and the mechanical arm has the obstacle avoidance characteristic and has the good environment adaptation.

The invention discloses a strong convection weather duration forecasting method based on integrated learning. The method comprises the steps of S1, selecting a data source, wherein ground meteorological station data of a forecast area and two sounding station data closest to the forecast area are selected; S2, carrying out preprocessing data, wherein errors and missing data are eliminated, the duration of each strong convection weather is selected as output according to the calculated relevant strong convection forecast parameters as input, and the time is considered as 0 when no strong convection weather occurs on the current day, and normalization processing is performed on the forecast parameters, namely the input; and S3, performing selection of machine learning algorithms, wherein a Knearest neighbor algorithm, a polynomial regression algorithm, a decision tree algorithm and a neural network algorithm are selected. According to the method disclosed by the invention, various meteorological elements of the current day of the strong convection weather are mainly used for speculation of the possible duration of the strong convection weather, and through a multi-machine learning algorithm comparison strategy, the target task is trained and tested, and the optimal learning algorithm is selected and used in an actual forecasting task.

The invention discloses a large wind turbine variable pitch system identification method based on an optimized RBF neural network. The method comprises the following steps that firstly, dynamic optimization improvement is carried out on a network structure by adopting an output sensitivity method on the basis of the traditional neural network identification algorithm technology, simulation software is adopted to control simulation to obtain experimental data by adopting a Bladed wind turbine from a great Britain company named Grarrad Hassan Partners, the wind speed v and the pitch angle beta are used as input signals, and the power generation power P serves as an output signal. Further, according to the system identification principle, a model and related measurement information are used for building an identification system framework. Secondly, the RBF is used for identifying the algorithm due to the strong nonlinear mapping capability of the neural network, under the excitation of asystem input signal, the identification system infinitely and approximately outputs the actual power output of the system. Finally, the problem that the network learning speed rate is difficult to select is solved, a gradient descent method and an optimization algorithm are provided, and the optimal learning speed rate of the network structure is derived. The method has high self-adaptive capacityand anti-interference capability, and has a certain practical value.