539 results about "Proactive learning" patented technology

Semantic indexing and retrieval of multimedia content requires that the content is sufficiently annotated. However, the great volumes of multimedia data and diversity of labels make annotation a difficult and costly process. Disclosed is an annotation framework in which supervised training with partially labeled data is facilitated using active learning. The system trains a classifier with a small set of labeled data and subsequently updates the classifier by selecting a subset of the available data-set according to optimization criteria. The process results in propagation of labels to unlabeled data and greatly facilitates the user in annotating large amounts of multimedia content.

A document annotation system 10 includes a graphical user interface 22 used by an annotator 30 to annotate documents. An active learning component 24 trains an annotation model and proposes annotations to documents based on the annotation model. A request handler 26, 32, 34, 42 conveys annotation requests from the graphical user interface 22 to the active learning component 24, conveys proposed annotations from the active learning component 24 to the graphical user interface 22, and selectably conveys evaluation requests from the graphical user interface 22 to a domain expert 40. During annotation, at least some low probability proposed annotations are presented to the annotator 30 by the graphical user interface 22. The presented low probability proposed annotations enhance training of the annotation model by the active learning component 24.

Educational gaming systems and methods provide an active learning environment to multiple users, wherein various incentives are provided to the users for enhanced engagement in learning, such as, for example, but not limited to, novel reward mechanisms, serious game elements, user collaboration, social interaction and the like.

The invention relates to the technical field of image information processing, in particular to a multiclass image classification method based on active learning and semi-supervised learning. The method comprises five steps: initial sample selection and classifier model training, BvSB active learning sample selection, CST semi-supervised learning, training sample set and classifier model updating and assorting process iteration. Through the operations of BvSB active learning sample selection, CST semi-supervised learning and SVM classification, the invention has efficient image classification effect under the condition of less manual tagging, does not increase overmuch computation burden, can quickly provide classification effects and also can take consideration of the demand of a classification system on the computation complexity.

A method of quantifying similarities between sequential data streams typically includes providing a pair of sequential data streams, designing a Hidden Markov Model (HMM) of at least a portion of each stream; and computing a quantitative measure of similarity between the streams using the HMMs. For a plurality of sequential data streams, a matrix of quantitative measures of similarity may be created. A spectral analysis may be performed on the matrix of quantitative measure of similarity matrix to define a multi-dimensional diffusion space, and the plurality of sequential data streams may be graphically represented and / or sorted according to the similarities therebetween. In addition, semi-supervised and active learning algorithms may be utilized to learn a user's preferences for data streams and recommend additional data streams that are similar to those preferred by the user. Multi-task learning algorithms may also be applied.

Systems and methods for active learning of statistical machine translation systems through dynamic creation and updating of parallel corpus. The systems and methods provided create accurate parallel corpus entries from a test set of sentences, words, phrases, etc. by calculating confidence scores for particular translations. Translations with high confidence scores are added directly to the corpus and the translations with low confidence scores are presented to human translations for corrections.

The present invention is directed to two fundamental characteristics of a scene image: (1) the scene image often containing complex semantics; and (2) a great number of manual annotation images taking high labor cost. The invention further discloses a scene image annotation method based on an active learning and a multi-label and multi-instance learning. The method comprises: training an initial classification model on the basis of a label image; predicting a label to an unlabeled image; calculating a confidence of the classification model; selecting an unlabeled image with the greatest uncertainty; experts carrying on a manual annotation on the image; updating an image set; and stopping when an algorithm meeting the requirements. An active learning strategy utilized by the method ensures accuracy of the classification model, and significantly reduces the quantity of the scenario image needed to be manually annotated, thereby decreasing the annotation cost. Moreover, according to the method, the image is converted to a multi-label and multi-instance data, complex semantics of the image has a reasonable demonstration, and accuracy of image annotation is improved.

Systems and methods consistent with the present invention improve manual screening processes for a group of documents by implementing an active learning screening process with dynamic information modeling. A classification algorithm is trained to recognize the relationships between concept tags applied to a subset of the documents and true or correct utility or relevance ratings applied to the subset of documents. Once adequately trained using several subsets of documents, the classification algorithm may be applied to the entire group of documents, screening out documents that are not relevant or important.

A body of text comprises a plurality of unknown attributes and a plurality of unknown values. A first classification sub-component labels a first portion of the plurality of unknown values as a first set of values, whereas a second classification sub-component labels a portion of the plurality of unknown attributes as a set of attributes and a second portion of the plurality of unknown values as a second set of values. Learning models implemented by the first and second classification subcomponents are updated based on the set of attributes and the first and second set of values. The first classification sub-component implements at least one supervised classification technique, whereas the second classification sub-component implements an unsupervised and / or semi-supervised classification technique. Active learning may be employed to provide at least one of a corrected attribute and / or corrected value that may be used to update the learning models.

Active learning-based fraud detection techniques are provided in adaptive authentication systems. An authentication request from an authentication requestor is processed by receiving the authentication request from the authentication requester; comparing current data for the user associated with the user identifier with historical data for the user; generating an adaptive authentication result based on the comparison indicating a likelihood current user data is associated with a fraudulent user; and performing one or more additional authentication operations to improve learning if the request satisfies one or more predefined non-risk based criteria. The predefined non-risk based criteria comprises, for example, (i) the request receiving a riskiness score below a threshold based on current data and wherein the request was expected to have a risk score above a threshold, or (ii) the request being in a bucket having a number of tagged events below a threshold.

An “active learning” method trains a compact classifier for view-based object recognition. The method actively generates its own training data. Specifically, the generation of synthetic training images is controlled within an iterative training process. Valuable and / or informative object views are found in a low-dimensional rendering space and then added iteratively to the training set. In each iteration, new views are generated. A sparse training set is iteratively generated by searching for local minima of a classifier's output in a low-dimensional space of rendering parameters. An initial training set is generated. The classifier is trained using the training set. Local minima are found of the classifier's output in the low-dimensional rendering space. Images are rendered at the local minima. The newly-rendered images are added to the training set. The procedure is repeated so that the classifier is retrained using the modified training set.

The invention discloses a text classification model optimization method based on crowdsourcing feedback and active learning. The method comprises the following steps that: selecting a text dataset, dividing the text dataset into an initial training set and a residual dataset; obtaining a word from the text dataset; constructing the feature set of the text dataset, and carrying out vectorization on the text dataset; and introducing the active learning on a classification model, predicting the sentiment polarity of the text dataset subjected to the vectorization, and combining a crowdsourcing feedback information optimization model to obtain a text classification result. By use of the method, constructing is used for collecting manually annotated reasons, more user information is obtained, the subjective feeling of people is mined, crowdsourcing feedback information is fused into the model in a weight change way, and the text classification model is optimized so as to improve model classification performance. An active learning algorithm is also introduced, and an annotation sample with a highest value is picked up and handed to a crowdsourcing platform to be annotated so as to lower annotation cost. Under a limited budget, annotation accuracy is improved, and the problem that a text classification task containing label data is in shortage is solved.

The invention discloses a method for generating suggested keywords of sponsored search advertisement based on user feedback, which develops semantic relatedness between words and establishes a learning machine to evaluate the relatedness of word pairs by utilizing the feedback information of users on the relatedness of a small quantity of word pairs so as to select words which has high relatednesswith seed keywords describing advertiser products or service concepts as the suggested keywords to be recommended to advertiser users, wherein an active learning method is used for selecting a word pair sample with maximal information quantity to evaluate user relatedness so as to enhance the efficiency and the precision of generating related suggested keywords. The invention can effectively generate hundreds of to thousands of related words to be used as the suggested keywords to be recommended to the advertiser users corresponding to each seed keyword; in addition, the advertiser users carry out sponsored search bidding for the suggested keywords, thus the related click rate of the sponsored advertisement can be effectively enhanced, and the advertisement benefit is increased.

Herein disclosed is a data classifying device whereby support vector machine performs a data classification based on a learning result obtained by performing an active learning method, comprises: a correct answer database adapted to store therein examples and their correct answer classes; a pooling section adapted to pool examples to which correct answer classes are not yet assigned; an SVM learning section adapted to perform learning of the support vector machine by using correct answer examples stored in the correct answer database; an SVM classifying section adapted to store therein the learning result obtained by the SVM learning section and perform the data classification based on the learning result thus stored therein; an active learning-purposed example selecting section adapted to select examples for use in the active learning from the pooling section by using the learning result; and a pooled example increasing section adapted to acquire new examples to which correct answer classes are not yet assigned and pool them in the pooling section such that the number of examples stored in the pooling section is increased. With the data classifying device thus configured, it is possible to reduce time required to improve accuracy in data classifying and to speed up the improvement of the accuracy, thereby providing higher accuracy.

The invention discloses a weak supervised text classification method and device based on active learning. The method comprises steps of firstly, extracting a first sample serving as a cluster center of a sample cluster from an unlabeled sample set; forming an initial training set based on the first samples, training a reference model by using the initial training set to obtain an initial classification model, and forming the initial training set by using the first samples, thereby not only reducing the number of training samples, but also ensuring the accuracy of the classification model at the initial stage; repeatedly utilizing the classification model to obtain the initial classification and confidence coefficient of the remaining samples in the sample set, so that manual labeling is not needed; extracting a second sample from the remaining samples according to the confidence coefficient, and performing data enhancement processing on the second sample to update the training set, thereby improving the generalization capability and robustness of the model; and finally, training the classification model by using the updated target training set until the classification model meets apreset condition, thereby realizing multi-round active training of the classification model.

A method for training a neuron network using a processor in communication with a memory includes determining features of a signal using the neuron network, determining an uncertainty measure of the features for classifying the signal, reconstructing the signal from the features using a decoder neuron network to produce a reconstructed signal, comparing the reconstructed signal with the signal to produce a reconstruction error, combining the uncertainty measure with the reconstruction error to produce a rank of the signal for a necessity of a manual labeling, labeling the signal according to the rank to produce the labeled signal; and training the neuron network and the decoder neuron network using the labeled signal.

Methods and systems for performing active learning for defect classifiers are provided. One system includes one or more computer subsystems configured for performing active learning for training a defect classifier. The active learning includes applying an acquisition function to data points for the specimen. The acquisition function selects one or more of the data points based on uncertainty estimations associated with the data points. The active learning also includes acquiring labels for the selected one or more data points and generating a set of labeled data that includes the selected one or more data points and the acquired labels. The computer subsystem(s) are also configured for training the defect classifier using the set of labeled data. The defect classifier is configured for classifying defects detected on the specimen using the images generated by the imaging subsystem.

A method trains a multi-class classifier by iteratively performing the following steps until a termination condition is reached. The probabilities of class membership for unlabeled data obtained from an active pool of unlabeled data are estimated. A difference between a largest probability and a second largest probability is determined. The unlabeled data with the lowest difference is selected, labeled and then added to a training data set for training the classifier.

An active learning system and method are disclosed for generating a visual representation of a set of unlabeled elements to be labeled according to class. The representation shows the unlabeled elements as data points in a space and each class as a class point in the space. The position of each of the data points in the space reflects the uncertainty of a model regarding the classification of the respective element. The color of each data point also reflects the uncertainty of the model regarding the classification of the element and may be a mixture of the colors used for the class points.

The invention provides an image recognition and recommendation method based on neural network depth learning. The method obtains pictures and classification from an image database, inputs to a convolution neural network, trains the neural network through repeated forward and backward propagation, improves image recognition accuracy, and extracts a 20-layer neural network model. By using this model, the object recognition and classification is carried out by collecting static pictures. Results are recognized, and by combining with the personalized characteristics of the input, the input probability of interest is analyzed. By using the machine learning model based on the effective recognition and classification of the material cloud database, and using the recommendation system algorithm, the predicted content material is pushed to the image inputter for cognitive learning. The method of the invention has the advantages of high image recognition rate, multiple recognition types and accurate content recommendation, and can be applied to the electronic products of a computer with a digital camera, a mobile phone, a tablet and an embedded system, so that people can photograph and recognize the objects seen in the eyes and actively learn the knowledge of recognizing the objects.

The invention discloses a multiclass image classification method based on a semi-supervised extreme learning machine. A training sample set is divided into an already marked sample set L, an unmarked sample set U and a test set T; training subsets of N differences are constructed by carrying out replaceable resampling on samples in the U; semi-supervised extreme learning machine models are respectively trained with the already marked training sample L on each subset, wherein all together there are N classifiers; the sum of output of corresponding nodes of N extreme learning machines is solved and an average is taken; after normalization processing on the output, uncertainty evaluation is carryed out on the samples by use of the active learning technology of a best mark and second best mark algorithm BvSB, most uncertain samples are taken from the U for manual marking and are transferred to the L; and a classifier model is re-updated until iteration ends. Through the method provided by the invention, the problems of low classification correction rate and low learning speed existing in image classification in related arts are solved, and a certain foundation is laid for accurate, rapid and stable image classification.

One embodiment of the present invention sets forth a technique for processing training data for a machine learning model. The technique includes training the machine learning model using training data comprising a set of features and a set of original labels associated with the set of features. The technique also includes generating multiple groupings of the training data based on internal representations of the training data in the machine learning model. The technique further includes replacing, in a first subset of groupings of the training data, a first subset of the original labels with updated labels based at least on occurrences of values for the original labels in the first subset of groupings.

A computer-assisted educational system. Educational courses are provided, comprising individual lessons. The lessons are stored at one, or more, servers, and are transmitted to remote computers over a network, such as an internet. Students utilize the remote computers to participate in the lessons. At any given time, the remote computer displays the lessons available to a student, so that, for a given course, the student is not held in lock-step with the other students, but may study the lesson's of the student's own choice. This freedom is subject to any requirements of certain lessons which require prerequisite lessons be completed. The prerequisites must be completed first.

Systems and methods for motion attribute recognition are defined using data stream pre-processing to orient, align and segment motion data before using non-parametric classification recognition to search a motion data exemplar database or using parametric classification recognition to find attributes by comparing pre-processed motion data with support vector machines. Results from the non-parametric classification recognition and the parametric classification recognition may be fused to produce a single result. Active learning and metric learning are used to improve searches of the database and comparisons to the support vector machines.

The invention discloses a multi-label classifier constructing method based on cost-sensitive active learning. Specific label information of each sample is directly used, an initial classifier is trained on a marked set, then, learning is performed on a non-marked set through a current classifier, a certain number of sample-label pairs of non-marked samples largest in misclassification cost are selected, real label values of the selected sample-label pairs are marked and added into the marked set, and a training sample set and the classifiers are updated. According to the method, target misclassification cost can be achieved with a small number of iteration times, and learning efficiency is greatly improved; sampling granularity is shrunk to the sample-label pairs, cost for marking samples is greatly reduced, and more remarkable effect is achieved in multi-label classification with a large label number.

An active learning framework is operative to identify informative questions that should be added to existing question-answer (Q&A) pairs that comprise a training dataset for a learning model. In this approach, the question—answer pairs (to be labeled as “true” or “false”) are automatically selected from a larger pool of unlabeled data. A spatial-directed clustering algorithm partitions the relevant question-answer space of unlabeled data. A margin-induced loss function is then used to rank a question. For each question selected, a label is then obtained, preferably by assigning a prediction for each associated question-answer pair using a current model that has been trained on labeled question-answer pairs. After the questions are labeled, an additional re-sampling is performed to assure high quality of the training data. Preferably, and with respect to a particular question, this additional re-sampling is based on a distance measure between correct and incorrect answers.

The invention discloses a convolutional neural network and active learning-based ceramic tile surface defect recognition method. The method comprises the following steps of: (1) obtaining and preprocessing an image; (2) establishing a training set; (3) establishing and training a convolutional neural network; (4) carrying out active learning; (5) carrying out model iteration; and (6) carrying outonline detection. Compared with the prior art, the method has the advantages that (1) features of ceramic tile surface defects are automatically extracted by using the convolutional neural network, sothat good priori knowledges are not required in the aspect of defect feature extraction, and multiple defect types in one to-be-detected image can be recognized; and (2) active learning is imported in convolutional neural network training, so that labeling cost of samples is effectively decreased and the model convergence is accelerated.

The invention provides a classifier training method, a device, an electronic device and a computer-readable medium, belonging to the technical field of the Internet. A method for train that classifierinclude training labeled samples in a plurality of data feature to obtain a primary classifier; Inputting an unlabeled sample from the plurality of data features to the primary classifier for training to output a label of the unlabeled sample, and calculating a probability that the unlabeled sample labels the label; According to the probability of the label, a part of the unlabeled samples is selected for expert labeling, and the samples labeled by the expert are retrained to obtain a new classifier. This method is based on active learning to update the classifier. Combined with a large number of data and artificial experience, the classifier model can make the classification results more accurate and improve the classification effect. There are only a small number of labeled samples anda small number of manual labeling in the training samples, which can quickly process the classification of a large number of unlabeled samples and improve the efficiency.

The invention discloses a four-classifier cooperative training method combining active learning, relates to cooperative training combining active learning, and belongs to the technical field of machine learning. Four classifiers and active learning are used in the implementation process to further improve the cooperative training method. The four classifiers are adopted for cooperative training, and samples consistent in the judgment of three classifiers are directly added into a training set, so excessive noise is prevented from being introduced at the same time of improving the confidence of the unmarked samples; because active learning is performed on the difficultly distinguished samples, the learning effect is improved, and meanwhile, the identification function of each classifier is modified by properly processing the samples; and because the samples inconsistent in the judgment of the three classifiers are used as the difficultly distinguished samples, the condition requirement for active learning is high, and the implementation is simple. The method is applied in the fields of web page classification, image processing, face identification, intrusion monitoring and the like.