69 results about "Madaline" patented technology

A process control system integrates the collection and analysis of process control data used to perform certain computationally expensive process control functions, like adaptive model generation and tuning parameter generation, in the same control device in which one or more of the process control routines are implemented, to thereby provide for faster and more efficient support of the process control routines. This system replaces a layered approach using multiple processing devices by integrating an analytical server which performs computationally expensive analyses used by one or more control routines directly into the real-time control device in which the one or more control routines are located. This integration provides the ability to analyze large quantities of data for multiple process loops controlled by a particular device in a fast and efficient manner.

A method of creating and using an adaptive DMC type or other MPC controller includes using a model switching technique to periodically determine a process model, such as a parameterized process model, for a process loop on-line during operation of the process. The method then uses the process model to generate an MPC control model and creates and downloads an MPC controller algorithm to an MPC controller based on the new control model while the MPC controller is operating on-line. This technique, which is generally applicable to single-loop MPC controllers and is particularly useful in MPC controllers with a control horizon of one or two, enables an MPC controller to be adapted during the normal operation of the process, so as to change the process model on which the MPC controller is based to thereby account for process changes. The adaptive MPC controller is not computationally expensive and can therefore be easily implemented within a distributed controller of a process control system, while providing the same or in some cases better control than a PID controller, especially in dead time dominant process loops, and in process loops that are subject to process model mismatch within the process time to steady state.

A system for managing a weapon system is provided. The system includes a number of mobile robotic vehicles (MRVs), a number of squads, each squad having a lead MRV and member MRVs, a number of central control systems, a number of reactive control systems, a central planning control configured to control the plurality of central control systems, a behavior-based reactive control configured to control the plurality of reactive control systems, an intermediated control layer configured to control the central planning control and the behavior-based reactive control, a number of hybrid control models configured to communicate with the intermediate control layer, the hybrid control models including a planning driven model and an adaptation model, a number of synthetic control models configured to communicate with the hybrid control models, and a number of synthetic hybrid control models configured based on combinations of the hybrid control models and the synthetic control models.

A model of a particular biological state can be developed. The model may be used to determine if an unknown biological sample exhibits a particular biological state. This can be done by receiving either a biological sample or data associated with the biological sample. After the data is received, the data may be input into the model. In one embodiment, the acquisition of the data associated with the biological sample is performed at a first location and the imputing of the data into the model is performed at a second location different than the first location. Unless the data maps identically to the model, the data would have an inherent effect on the position of the particular clusters within the discriminatory pattern, if it is allowed to affect the model. The modeling software can keep track of the net effect on the model that each sample received has on the position of the model. If the model has drifted outside of a predetermined tolerance, the model can be updated. Various business relationships may be developed to undertake various steps of the overall method for providing a diagnosis to a patient.

The invention provides a method of predicting output values of a physical system from a set of measured inputs of the system using an adaptive model. At each time when a prediction is made, the model is re-initialized to an initial, off-line model and is then refined to incorporate on-line data using a predetermined number of recent sets of measured inputs and outputs. The model thus always remains “tethered” to the initial, off-line model and if operating conditions remain steady, the model does not become too specific to those operating conditions.

The invention discloses a multi-source domain adaptive model and method based on partial feature alignment, and the method comprises the steps: enabling a feature selection module for partial featureextraction to generate a selection vector of a feature level according to the similarity of each feature dimension of a source domain and a target domain on the basis of a conventional convolutional neural network or residual neural network feature extractor, wherein after the selection vector acts on the initial feature map, part of features highly related to the target domain in the source domain can be screened out. On the basis, the invention further provides three partial feature alignment loss functions for intra-class, inter-domain and inter-class features, so that the purified featuremap has better distinguishability for a classifier, and partial features related to a source domain and a target domain are highlighted. The method is used for the multi-source field self-adaptive classification data set, and compared with an existing multi-source field self-adaptive model, the classification accuracy is higher, and the feature selection effect is better.

A self-adaptive resource management method and a corresponding system are proposed. In the solution of the invention, an authority (105) publishes multiple rules into a corresponding repository (115). Each rule defines a desired target configuration for a category of subjects (110). Some of the rules can be conditioned to an authorization for their application on the subjects (by means of an optional enabling code). Whenever a generic subject tries to apply a conditioned rule, a test is made to verify whether the application of the rule has been authorized for the subject; if not, the subject is considered not compliant with the rule for a lack of authorization. A supervisor (135) authorizes each enabling code in an asynchronous way (with respect to the application of the corresponding rules); for example, the supervisor can authorize the enabling code for specific subjects, for a group of subjects (defined either statically by a list of subjects or dynamically by their characteristics), or for all the subjects. In this way, the application of the rules on the different subjects can be controlled centrally (for example, during a transition from a resource management system based on the enforcement model) without impairing the self-adaptive model.

A method for generating a test-speaker-specific adaptive system for recognising sounds in speech spoken by a test speaker; the method employing:(i) training data comprising speech items spoken by the test speaker; and(ii) an input network component and a speaker adaptive output network, the input network component and speaker adaptive output network having been trained using training data from training speakers;the method comprising:(a) using the training data to train a test-speaker-specific adaptive model component of an adaptive model comprising the input network component, and the test-speaker-specific adaptive model component, and(b) providing the test-speaker-specific adaptive system comprising the input network component, the trained test-speaker-specific adaptive model component, and the speaker-adaptive output network.

A computer system is proposed including an adaptive signal processing model of a kind in which a multiplicative section, such as a VLSI integrated circuit, processes data input to the model, using hidden neurons and randomly-set variables, and an adaptive output layer processes the outputs of the multiplicative section using variable parameters. Controllable switching circuitry is proposed to control which data inputs are fed to which hidden neurons, to reduce the number of hidden neurons required and increase the effective number of data inputs. An algorithm is proposed to selectively disable unnecessary hidden neurons. Normalisation, and a winner-take all stage, may be provided at the hidden layer output.

The invention discloses a medical image segmentation or classification method based on small sample domain self-adaption, and the method comprises the steps: downloading medical image data and clinical image data from a public data set, and taking the processed medical image data as a source domain training data set with a known label; using the processed clinical image data as a to-be-classified or segmented target domain data set, marking a very small amount of clinical image data by a doctor, constructing a small sample domain self-adaption model by a feature extractor and a classifier realized by a convolutional neural network, and training to obtain a trained small sample domain self-adaption model; and inputting a to-be-classified or segmented target domain data set into the small sample domainself-adaption model with the best classification effect to obtain a classification or segmentation result to which the focus of the target domain data set belongs. Different objective functions are adopted according to whether the data contain labels or not, cross-domain migration of the small sample domain self-adaption model is achieved, and the method is applied to classification or segmentation of image data of clinic, pathology, ultrasound and the like.

The invention relates to a multi-view network attack and defense simulation system. The system comprises: a network information environment module used for configuring hardware information and software information and establishing a network information system model through the hardware information and the software information; a background flow module which is used for simulating behavior information of a user in the network environment to obtain a background flow model; an attack and defense module which is used for establishing a network threat model according to the network threat behaviorand establishing a network defense model according to the network defense behavior; a simulation module which is used for constructing a network attack and defense adaptive model according to the network information system model, the background flow model, the network threat model and the network defense model, acquiring preset network attack and defense simulation data, inputting the network attack and defense simulation data into the network attack and defense adaptive model, and outputting network attack and defense element data. By adopting the method, network attack and defense deductioncan be easily realized.

The invention discloses a method for constructing a target detection adaptive model based on a CycleGAN and a pseudo label. The method comprises the following steps: S1, preprocessing a source domaindata set and a target domain data set; S2, converting the source domain data set into an intermediate domain data set close to a target domain data set by using a CycleGAN network, and inputting the intermediate domain data set into a Faster RCNN network for training to obtain a preliminary domain adaptive model Q; re-inputting the target domain data set into the model Q to obtain a target domaindata set with a pseudo tag; and S3, inputting the intermediate domain data set and the target domain data set with the pseudo tag into the model Q in turn for iterative updating and optimization, andfinally obtaining a target detection domain adaptive model based on the CycleGAN and the pseudo tag. According to the method, the confidence is used to improve the target detection total loss functionof the Faster RCNN network to train the obtained target detection domain adaptive model, and the problem of domain drift of target detection due to distribution difference between two domains can besolved.

The invention discloses a method and a device for constructing a MADALINE neural network based on sensitivity. The method comprises the following steps of selecting a big enough positive integer m as a hidden layer neuronal number, constructing a three-layer MADALINE neural network, and setting initial network parameters; utilizing a marked sample set to train the neural network until a certain given extremely small threshold value e is converged in a cost function to obtain a classifier through training; calculating the sensitivity of hidden layer neurons, and ranking from small to large according to the sensitivity; eliminating the hidden layer neuron with the minimum sensitivity to obtain the MADALINE neural network of a new structure; training the new MADALINE neural network on the basis of the original parameters by reusing the marked sample set; taking the network structure of the MADALINE neural network which has the minimum hidden layer neuronal number and is capable of converging as a final network structure, wherein the network of the network structure is the classifier for final output. According to the method and the device, the construction efficiency of the neural network can be effectively improved, and the performance of the MADALINE neural network is improved.

The invention relates to a self-adaptive picture classification method in a semi-supervised field based on hierarchical relationship, belongs to the technical field of computer vision processing, and can accurately classify images in a target domain. According to the method, a hierarchical relationship between categories is introduced, and hierarchical relationship information is provided for a prototype by utilizing parent class label and child class label information of all source domains and a small amount of labeled target domain data, so that the prototype distance of the same parent class in a prototype space is relatively short, and a self-adaptive model in the semi-supervised field is helped to obtain a better classification effect. According to the method, the maximum and minimum entropy adversarial learning is carried out on the model by using the gradient inversion layer and using the unsupervised data, so that the prototype vector which has resolution for categories and is not specific to a certain field is extracted, and the classification effect of the model on target domain data is improved. According to the method, the effect on a data set with large domain offset and a large number of categories is ideal, and it is indicated that the method can solve the complex domain offset problem.

The invention belongs to the field of machine learning, recommendation systems and text classification, and particularly relates to a music classification method based on a self-adaptive CNN and a semi-supervised self-training model.The method comprises the following steps: firstly, analyzing the importance of fine classification of music to user selection; extracting music lyric features and emotion features associated with the item through a natural language processing technology, and then extracting and preliminarily classifying the features by using a neural network based on an attention mechanism; and finally, performing fine classification on each song according to a semi-supervised self-training method. According to the method, the self-adaptive CNN model and the semi-supervised self-training model are combined to finely divide the music data, so that the user can accurately search favorite music, the time of the user is saved, and the search accuracy is improved.

The invention provides a method for diagnosing a bearing compound fault in an unknown target domain under a variable working condition. The method comprises the following steps: constructing a fault feature extractor consisting of a deep residual network based on a multi-layer domain adaptive method; inputting a preprocessed bearing vibration signal, and carrying out distribution difference matching on features, extracted through a plurality of residual blocks, of source domain data and target domain data to obtain migratable features; representing the composite fault as a combination of single faults through multi-label learning; and a binary association strategy is used to train corresponding binary classifiers for various single faults, and the features of the single faults are separated from the composite faults and are diagnosed respectively. According to the method, the problems that a traditional diagnosis scheme depends on expert knowledge and is difficult to effectively decouple and recognize the composite fault are solved, accurate diagnosis of the composite fault of the bearing under variable working conditions is achieved, meanwhile, dependence of an existing method on marked data is eliminated, and accurate diagnosis can be conducted on a related but invisible target domain.

The invention discloses an adversarial domain adaptive model training method. An adversarial domain adaptive model comprises a source domain embedding extractor, a speaker discriminator, a target domain embedding extractor and a domain discriminator. The method comprises the following steps of: S10, configuring parameters of a shared part layer between the source domain embedding extractor and thetarget domain embedding extractor; S20, inputting labeled source domain training data into the source domain embedding extractor, and inputting the output of the source domain embedding extractor into the speaker discriminator to obtain speaker loss; and S30, inputting unlabeled target domain training data into the target domain embedding extractor, and inputting the output of the target domain extractor and the output of the source domain embedding extractor into the domain discriminator to obtain Waserstein loss. In the method provided by the invention, the feature extractors of the sourcedomain and the target domain are not completely the same. Therefore, different parameter parts of the feature extractors can solve the conflict problem between a main task and a domain adversarial training task.

A method for generating a test-speaker-specific adaptive system for recognising sounds in speech spoken by a test speaker; the method employing:(i) training data comprising speech items spoken by the test speaker; and(ii) an input network component and a speaker adaptive output network, the input network component and speaker adaptive output network having been trained using training data from training speakers;the method comprising:(a) using the training data to train a test-speaker-specific adaptive model component of an adaptive model comprising the input network component, and the test-speaker-specific adaptive model component, and(b) providing the test-speaker-specific adaptive system comprising the input network component, the trained test-speaker-specific adaptive model component, and the speaker-adaptive output network.

The invention discloses an adaptive model conversion method and system for a spiking neural network. The method comprises the steps: obtaining a to-be-converted deep artificial neural network model, and enabling the to-be-converted deep artificial neural network model to be a deep neural network model under any one of Keras, Pytorch and TensorFlow; if the to-be-converted deep artificial neural network model is a deep neural network model under a Pylot or TensorFlow framework, converting the to-be-converted deep artificial neural network model into a deep artificial neural network model under aKeras framework; converting the deep artificial neural network model under the Keras framework into an impulsive neural network. According to the method disclosed in the invention, the deep artificial neural network model under various intelligent frameworks can be realized, and the adaptive conversion to the novel pulse neural network is completed on the premise of ensuring a small amount of model precision loss.

The invention provides a multi-domain adaptive model training method, system and device and a storage medium, and the method comprises the steps: obtaining source domain data and target domain data, dividing the target domain data into a plurality of groups according to different domains, and each group comprises the target domain data of one domain; the source domain data is input into a source domain classification network, a group of target domain data is input into a target domain classification network for training, the weight of an incremental convolutional layer is updated according to a training result, and the weights of the source domain classification network and the target domain classification network are shared; loading the weight of the incremental convolutional layer to a source domain classification network and a target domain classification network; and selecting another group of target domain data, and carrying out iterative training until the target domain data is trained, so as to obtain a multi-domain adaptive model. The problem that only a single source domain to a single target domain can be used in a traditional adaptive model is solved.

The invention discloses an adaptive speech characteristic model training method of an aviation cabin environment. The method comprises the steps of S1, acquiring a personal adaptive speech characteristic; S2, supplying a personal adaptive speech mark; S3, supplying a basic characteristic model; S4, utilizing a deep neural network (DNN) adaptive algorithm, updating the basic characteristic model according to the personal adaptive speech characteristic and the corresponding personal adaptive speech mark, thereby generating an adaptive model; S5, performing identification testing, and verifying improving capability of the adaptive model to personal speech identification; and S6, performing model packaging for generating a personal characteristic database. The adaptive speech characteristic model training method has advantages of updating the basic characteristic model through the personal adaptive speech characteristic, generating the adaptive model with higher identification capability, and effectively improving identification rate of avionics speech products.

The invention discloses a multi-objective optimization community discovery system and method based on dynamic social network attributes. The system comprises a receiving unit, a data preprocessing unit, a computing unit, a display unit and a main control unit. The method comprises the following steps: abstracting a dynamic network into a dynamic network diagram formed by combining networks of T time steps, and dividing the network of the current time step t into K communities, so that a network division result is kept continuous. The invention provides a multi-objective optimization method tosolve the problem of dynamic network community division. The multi-objective optimization method comprises an initialization part and an iteration part. According to the method, a probability selection method is provided to initialize a scheme and generate an initial scheme set, a structure evolution adaptive model is further provided and comprises community structure quality and community evolution continuity, and Pt quality and evolution continuity between Pt-1 and Pt are guaranteed by using different evaluation functions. The invention provides an objective function dynamic evolution continuity measurement for evaluating community evolution continuity.

The invention provides a robot reinforcement learning method based on an adaptive model. According to the robot reinforcement learning method, the step of model self-adaption is added while an environment model is learned normally, and therefore the effect of improving the accuracy of virtual data can be achieved. Specifically, when the model is of a neural network structure, the model adaptivelyimproves the accuracy of the model on virtual data by reducing the feature distribution of real data and virtual data on a network hidden layer. Experiments prove that in the fields of robot control and the like, compared with a previous model-based reinforcement learning method, the method has higher sampling efficiency and final performance.

The invention relates to the field of artificial intelligence, and discloses a vehicle damage feature detection method and device, computer equipment and a storage medium. The vehicle damage feature detection method comprises the steps of obtaining a to-be-detected vehicle damage image and inputting the to-be-detected vehicle damage image into an unsupervised domain adaptive network model; extracting vehicle features through a migration learning model based on a pytorch, and generating a local feature map and a global feature map; outputting a migration feature vector group according to the vehicle features, obtaining a first self-adaptive feature vector group through a strong local feature self-adaptive model, and obtaining a second self-adaptive feature vector group through a weak globalfeature self-adaptive model; and performing regularization processing on the migration feature vector group, the first adaptive feature vector group and the second adaptive feature vector group to obtain an identification result. According to the vehicle damage feature detection method, the damage type and the damage area in the to-be-detected vehicle damage image can be automatically identified.The invention also relates to a blockchain technology. The unsupervised domain adaptive network model in the invention can be stored in the blockchain.

The present application discloses a detection model training method and apparatus. The method includes determining an initial training model; determining a training sample; determining whether a lesion target is present in a first user body organ image through the initial detection model according to a feature of the each first user body organ image, to obtain a detection result; and determining a domain that each user body organ image in the training sample belongs to through the adaptive model according to a feature of the each user body organ image, to obtain a domain classification result; calculating, a loss function value related to the initial training model according to the detection result, the domain classification result, a first identifier, a second identifier, and a third identifier; and adjusting a parameter value in the initial training model according to the loss function value, to obtain a final detection model.

The invention discloses an anti-shielding self-adaptive target tracking method and system, a scale correlation filter is trained while a position correlation filter is trained, scale self-adaptive transformation can be realized, if the transformation does not exist, the size of a target frame is not changed in the training process and is the same as the size of a rectangular frame which is determined manually at the beginning, and the target frame is not changed in the training process. However, after scale transformation, the size of the target frame can be automatically changed along with the distance of the target, the target frame becomes smaller when the target moves farther from the camera, and the target frame becomes larger when the target moves closer, so that the accuracy and robustness of the whole algorithm are improved. According to the target tracking method, an adaptive model updating strategy is adopted, and whether the target is shielded or lost or not is detected by calculating a PSR value, so that a search area is expanded, the problem that tracking cannot be continued once the target is shielded or lost due to target movement and the like in a traditional target tracking method is solved, and the target tracking efficiency is improved. And the continuity and reliability of target tracking are improved.

The invention discloses an adaptive radar sea clutter prediction system, which comprises a radar, a database and an upper computer, wherein the radar, the database and the upper computer are connectedin sequence; the radar illuminates a detected sea area and radar sea clutter data are stored to the database; the upper computer carries out modeling and prediction on the sea clutter data in the database; and the upper computer comprises a data preprocessing module, a fuzzy wavelet neural network modeling module, a sea clutter prediction module, a discriminant model updating module and a resultdisplay module. The invention also provides a radar sea clutter prediction method based on a fuzzy wavelet neural network. The radar sea clutter prediction system and the method have the advantages ofadaptive structure parameters, automatic model updating and strong anti-noise interference ability.

The invention provides an intelligent fault diagnosis method based on an asymmetric domain adversarial self-adaptive model. The problems that existing unsupervised domain adaptive adversarial training is difficult, and target domain unlabeled data demand is large are solved. Firstly, a vibration signal acquired by a sensor is converted into a time-frequency map, and domain invariant fault features are obtained from the time-frequency map by using step-by-step domain confrontation training; and meanwhile, the lightweight network architecture and the Wasserstein i system are integrated into the adversarial network, so that the training cost of a network model is greatly reduced, the dependence on target domain data is reduced, and the adversarial migration model still keeps good performance in a small sample scene.