51 results about "Probabilistic classification" patented technology

Disclosed herein are systems, methods, and devices for classifying ophthalmic images according to disease type, state, and stage. The disclosed invention details systems, methods, and devices to perform the aforementioned classification based on weighted-linkage of an ensemble of machine learning models. In some parts, each model is trained on a training data set and tested on a test dataset. In other parts, the models are ranked based on classification performance, and model weights are assigned based on model rank. To classify an ophthalmic image, that image is presented to each model of the ensemble for classification, yielding a probabilistic classification score—of each model. Using the model weights, a weighted-average of the individual model-generated probabilistic scores is computed and used for the classification.

Fluorescence spectral data acquired from tissues in vivo or in vitro is processed in accordance with a multivariate statistical method to achieve the ability to probabilistically classify tissue in a diagnostically useful manner, such as by histopathological classification. The apparatus includes a controllable illumination device for emitting electromagnetic radiation selected to cause tissue to produce a fluorescence intensity spectrum. Also included are an optical system for applying the plurality of radiation wavelengths to a tissue sample, and a fluorescence intensity spectrum detecting device for detecting an intensity of fluorescence spectra emitted by the sample as a result of illumination by the controllable illumination device. The system also include a data processor, connected to the detecting device, for analyzing detected fluorescence spectra to calculate a probability that the sample belongs in a particular classification. The data processor analyzes the detected fluorescence spectra using a multivariate statistical method. The five primary steps involved in the multivariate statistical method are (i) preprocessing of spectral data from each patient to account for inter-patient variation, (ii) partitioning of the preprocessed spectral data from all patients into calibration and prediction sets, (iii) dimension reduction of the preprocessed spectra in the calibration set using principal component analysis, (iv) selection of the diagnostically most useful principal components using a two-sided unpaired student's t-test and (v) development of an optimal classification scheme based on logistic discrimination using the diagnostically useful principal component scores of the calibration set as inputs.

The invention relates to a deep learning based intelligent skin disease auxiliary diagnosis system, which comprises a classifier training unit, a language model unit and an intelligent auxiliary diagnosis unit. The intelligent auxiliary diagnosis unit comprises an image acquisition module, a voice interrogation module, a voice recognition and keyword extraction module, a probability classificationmodel, a RNN condition analysis module and a fusion classifier. The classifier training unit comprises a state diagram training set under a dermatoscope, a state standard database under skin lesion and dermatoscope, a CNN network convolution module and a sampling and classifying module. The language model unit comprises a medical term standard library, a RNN questioning management module, a RNN chief complaint management module and a skin disease medical knowledge base. The auxiliary diagnosis system has advantages that by deep learning for classifying skin lesion images, probable results areinferred, then a pre-installed dermatoscope image and histodiagnosis tag database is retrieved for doctors' reference, and accordingly accuracy in skin disease diagnosis can be greatly improved.

A class is to be characterized of a probabilistic classifier or clustering system that includes probabilistic model parameters. For each of a plurality of candidate words or word combinations, divergence of the class from other classes is computed based on one or more probabilistic model parameters profiling the candidate word or word combination. One or more words or word combinations are selected for characterizing the class as those candidate words or word combinations for which the class has substantial computed divergence from the other classes.

A method and system for classifying semi-structured documents by distinguishing sub-tree structural information as a distinct representative characteristic of a fragment of the document structure identified by a sub-tree node therein. The structural information comprises both an inner structure and an outer structure which individually can be exploited as representative data in a probabilistic classifier for classifying the sub-tree itself or the entire document. Additional representative feature data can also be independently used for classification and comprises the data content of the fragment structurally represented by the sub-tree and additionally with node attributes. The classification values independently generated from each of the different sets of features can then be combined in an assembly classifier to generate an automated classification system.

A system and method for providing decision support to a physician during a medical examination is disclosed. Data is received from a sensor representing a particular medical measurement. The received data includes image data. The received data and context data is analyzed with respect to one or more sets of training models. Probability values for the particular medical measurement and other measurements to be taken are derived based on the analysis and based on identified classes. The received image data is compared with training images. Distance values are determined between the received image data and the training images, and the training images are associated with the identified classes. Absolute value feature sensitivity scores are derived for the particular medical measurement and other measurements to be taken based on the analysis. The probability values, distance values and absolute value feature sensitivity scores are outputted to the user.

A system and method are provided for classifying objects using high resolution radar signals. The method includes determining a probabilistic classifier of an object from a high resolution radar scan, determining a deterministic classifier of the object from the high resolution radar scan, and classifying the object based on the probabilistic classifier and the deterministic classifier.

A probabilistic document categorizer has an associated vocabulary of words and an associated plurality of probabilistic categorizer parameters derived from a collection of documents. A new document is received. The probabilistic categorizer parameters are updated to reflect addition of the new document to the collection of documents based on vocabulary words contained in the new document, a category of the new document, and a collection size parameter indicative of an effective total number of instances of vocabulary words in the collection of documents.

The invention discloses a short text classification method and apparatus. The method comprises the steps of performing word segmentation preprocessing on to-be-classified short texts and obtaining an extended word of each word obtained by word segmentation; obtaining weight values of each word and the extended word of each word according to a pre-constructed lexical item set; according to the weight values, obtaining a probability of each type that a short text belongs to by utilizing a plurality of SVM classification models; and determining the type that the short text belongs to according to a preset probability classification model. According to the short text classification method, the problem of short text characteristic sparsity is solved, the complexity due to the adoption of multiple classification models is effectively lowered, and actual application requirements are better met.

A method and means to utilize machine learning to train a device to generate a confidence level indicator (CLI). The device is a CAD system that has been initially trained using initial machine learning to recommend classifications for image features presented to the device. Probabilistic classification is utilized to incorporate intermediate values given by a human operator to better indicate a level of confidence of the CAD system's recommendations as to what classes should be associated with certain image features.

The present invention provides an information providing method and apparatus, which have advantages of being objective, high in efficiency, wide in application scope, and good in scalability. The method comprises: extracting a sample feature vector and a corresponding sample decision from historical consultancy session data, wherein an element of the sample feature vector is an attribute value extracted from the historical consultancy session data according to a preset attribute, and the sample decision is a user questioning sentence; performing training by using a plurality of sample feature vectors and corresponding sample decisions to obtain a probabilistic classification model; extracting a to-be-tested feature vector from a current client consultancy request, wherein the to-be-tested feature vector is the same as the sample feature vector in format; inputting the to-be-tested feature vector into the probabilistic classification model, then receiving one or more candidate decisions and corresponding probabilities output from the probabilistic classification model, wherein the candidate decisions are candidate user questioning sentences; choosing K candidate decisions of greatest probabilities as forecasted decisions, then providing a client with a standard reply corresponding to the forecasted decisions.

Documents of a set of documents are represented by bag-of-words (BOW) vectors. L labeled topics are provided, each labeled with a word list comprising words of a vocabulary that are representative of the labeled topic and possibly a list of relevant documents. Probabilistic classification of the documents generates for each labeled topic a document vector whose elements store scores of the documents for the labeled topic and a word vector whose elements store scores of the words of the vocabulary for the labeled topic. Non-negative matrix factorization (NMF) is performed to generate a document-topic model that clusters the documents into k topics where k>L. NMF factors representing L topics of the k topics are initialized to the document and word vectors for the L labeled topics. In some embodiments the NMF factors representing the L topics initialized to the document and word vectors are frozen, that is, are not updated by the NMF after the initialization.

Systems and methods are disclosed that facilitate producing probabilistic outputs also referred to as posterior probabilities. The probabilistic outputs include an estimate of classification strength. The present invention intercepts non-probabilistic classifier output and applies a set of kernel models based on a softmax function to derive the desired probabilistic outputs. Such probabilistic outputs can be employed with handwriting recognition where the probability of a handwriting sample classification is combined with language models to make better classification decisions.

The invention discloses a pipe damage probability prediction method based on a BP neural network, and belongs to the technical field of urban water supply pipe networks. The method comprises the steps: carrying out the early data preparation for the prediction of the pipe damage probability; selecting a to-be-predicted pipe network region, and determining a training data range through the whole pipe network, an administrative area and a peripheral buffering region; selecting an influence factor causing the damage to a pipe, and extracting a training sample, needed by modeling, in the training data range; training a model through employing a BP neural network algorithm and the training sample, and evaluating the stability and prediction precision of the model through employing an ROC curve based on 10-fold cross validation; applying the trained model in the to-be-predicted pipe network region, and obtaining the probability that each pipe will be damaged; classifying the pipe damage probability in ArcGIS through employing a natural break point method, and making a thematic map. The method enlarges the research content of a conventional pipe damage probability prediction model, and provides a new idea for the building of a water supply pipe network asset management scientific method.

The invention discloses a method for classifying hyperspectral images on the basis of a combination of unmixing and adaptive end member extraction. The method is used for solving the technical problem of large errors of an existing method for classifying hyperspectral images on the basis of spectral unmixing. The technical scheme includes that the method comprises steps of roughly classifying the images, and extracting end member sets of various categories by the aid of a confusion matrix; linearly spectrally unmixing training samples in the various categories by the aid of the acquired end member sets, and acquiring optimal classification results by the aid of a probability classifier with an abundance value optimized on the basis of multinomial logistic regression; updating the end member sets of the various categories according to the classification results; iterating the procedure, and continuously optimizing the classifier so that the classification accuracy is improved. The method has the advantage that as shown by test results, the average accuracy of tests on a simulated data set, the average accuracy of tests on data of a true hyperspectral data set AVIRIS Indian Pine and the average accuracy of tests on data of a true hyperspectral data set ROSIS Pavia University are 81.98%, 62.19% and 82.38% respectively.

A system and method for providing decision support to a physician during a medical examination is disclosed. Data is received from a sensor representing a particular medical measurement. The received data includes image data. The received data and context data is analyzed with respect to one or more sets of training models. Probability values for the particular medical measurement and other measurements to be taken are derived based on the analysis and based on identified classes. The received image data is compared with training images. Distance values are determined between the received image data and the training images, and the training images are associated with the identified classes. Absolute value feature sensitivity scores are derived for the particular medical measurement and other measurements to be taken based on the analysis. The probability values, distance values and absolute value feature sensitivity scores are outputted to the user.

The invention provides a multi-label text classification method based on BiGRU and an attention mechanism. The method comprises the following steps: S1, acquiring a plurality of web texts; s2, preprocessing the plurality of web texts; s3, extracting deep information features of the web text by using the pre-trained word vector; s4, adding corresponding weights to the deep information features according to an attention mechanism; s5, performing probability classification of different types of labels on the data obtained in the step S4 by using BiGRU; and S6, outputting the probability of each web text on different types of labels. The method has the advantages that the pre-training word vector is adopted to further shorten the training time, the attention mechanism is adopted to enable theneural network to focus on important information for improving the classification effect, and compared with the prior art, the BiGRU and the attention mechanism are fused to enable the method to obtain the same high accuracy by using less training time.

According to the invention, an apparatus for classifying and sorting input data in a data stream includes a processor having a classifier input control with a first input and second input, an adaptive classifier, a ground truth data input, a ground truth resampling buffer, a source data re-sampling buffer, and an output. The processor is configured for sampling the input data with the input control, comparing one or more classes of the sampled input data with preset data classifications for determining the degree of mis-classification of data patterns, determining a probability proportional to the degree of mis-classification as a criterion for entry into a resampling buffer, entering data patterns causing mis-classification in a resampling buffer with a probability value proportional to the degree of mis-classification, comparing the data patterns to a ground truth source and aligning the data patterns with their associated data pattern labels employing the same decision outcome based on a mis-classification probability as applied to the resampling buffer to form a set of training data, and updating the adaptive classifier to correlate with the training data. These steps are repeated until a sufficient degree of data classification optimization is realized, with the output being an optimized data stream.

A class is to be characterized of a probabilistic classifier or clustering system that includes probabilistic model parameters. For each of a plurality of candidate words or word combinations, divergence of the class from other classes is computed based on one or more probabilistic model parameters profiling the candidate word or word combination. One or more words or word combinations are selected for characterizing the class as those candidate words or word combinations for which the class has substantial computed divergence from the other classes.

The invention discloses a defect detection method and system for a large-resolution image, and the method comprises the following steps of S1, carrying out the transverse and longitudinal sliding on adefect image through employing a sliding window with a certain size, and obtaining a plurality of partition images; S2, carrying out the defect detection on each partition image through a target detection model obtained based on small-resolution image training, obtaining a result set describing a target detection box, wherein the detection result of each target detection box comprises a probability classification value of a defect type contained in the target detection box; S3, calculating the Euclidean distance between two adjacent target detection frames in the result set in sequence, whenthe Euclidean distance is smaller than a window fusion threshold value, fusing the two target detection frames, and taking the target detection frame with the maximum probability classification valueas a defect detection result. According to the present invention, the defect detection of a large-resolution image is realized through a sliding segmentation and window fusion mode, the micro defectscannot be lost, and the overdetection or missing detection is avoided.

A method and means to utilize machine learning to train a device to generate a confidence level indicator (CLI). The device is a CAD system that has been initially trained using initial machine learning to recommend classifications for image features presented to the device. Probabilistic classification is utilized to incorporate intermediate values given by a human operator to better indicate a level of confidence of the CAD system's recommendations as to what classes should be associated with certain image features.

A probabilistic document categorizer has an associated vocabulary of words and an associated plurality of probabilistic categorizer parameters derived from a collection of documents. A new document is received. The probabilistic categorizer parameters are updated to reflect addition of the new document to the collection of documents based on vocabulary words contained in the new document, a category of the new document, and a collection size parameter indicative of an effective total number of instances of vocabulary words in the collection of documents.

The present invention provides a training corpus acquisition method and apparatus, and has the advantages of being high in automation degree and quick in acquisition speed. The method comprises: obtaining a first initial training corpus and a second initial training corpus; performing prediction on an optional training statement by using a probabilistic classification model constructed according to the first initial training corpus, so as to obtain a first prediction result; performing prediction on the optional training statement by using a probabilistic classification model constructed according to the first initial training corpus and the second initial training corpus, so as to obtain a second prediction result; and comparing the first prediction result with the second prediction result; if classification information of the first prediction result is inconsistent with that of the second prediction result, or if the classification information of the first prediction result is consistent with that of the second prediction result and a prediction probability of the first prediction result is less than that of the second prediction result, using the optional training statement and the classification information of the second prediction result as a training corpus and outputting the training corpus.

Techniques for image captioning with weak supervision are described herein. In implementations, weak supervision data regarding a target image is obtained and utilized to provide detail information that supplements global image concepts derived for image captioning. Weak supervision data refers to noisy data that is not closely curated and may include errors. Given a target image, weak supervision data for visually similar images may be collected from sources of weakly annotated images, such as online social networks. Generally, images posted online include ''weak'' annotations in the form of tags, titles, labels, and short descriptions added by users. Weak supervision data for the target image is generated by extracting keywords for visually similar images discovered in the different sources. The keywords included in the weak supervision data are then employed to modulate weights applied for probabilistic classifications during image captioning analysis.

The invention discloses a retinal layering method based on an OCT image. The OCT image of the fundus of the eye is collected as a sample image, the gray-scale normalization processing is carried out on the collected sample images, and then the sample image is flattened. The average filter is used to solve the longitudinal gradient, and then the upper and lower retinal boundaries are determined bycomparison. The calculated features are inputted to the training classifier to obtain the probability that pixels are located at the boundary of different layers. Finally, the optimal boundary and itsposition of the training classifier are optimized by using a graph theory algorithm through the probability. The method of the invention can realize the automatic segmentation of a plurality of layerstructures by using a commonly configured computer, and adopts the multi-level probability classification decision and optimization, and at the same time adopts serialization layering, has fewer original processing steps for the image, and improves detection accuracy certainly.

The invention belongs to the technical field of network topology modeling, and particularly relates to a router ownership detection method and system based on IP connection probability classification,and the method comprises the steps: dividing IP connection into intra-domain connection and inter-domain connection, taking an IP connection type as a probability model hidden variable, and taking the probability that connection features appear in different connection types as a probability model parameter; marking each IP connection initial type in the router-level topology by utilizing router-to-AS mapping in the data set; estimating probability model parameters by utilizing expectation maximization, and taking the connection type with higher probability as the connection type of the feature corresponding to the connection feature vector by utilizing naive Bayesian classification in each iteration until a model convergence condition is reached; and determining the AS to which the routerlocated in the same AS as the trace detection target IP belongs through a voting mechanism. According to the method, the characteristics of the data can be better applied instead of only depending onsubjective experience judgment, the accuracy and effectiveness of router ownership detection can be improved, and the method has a higher application value.

Documents of a set of documents are represented by bag-of-words (BOW) vectors. L labeled topics are provided, each labeled with a word list comprising words of a vocabulary that are representative of the labeled topic and possibly a list of relevant documents. Probabilistic classification of the documents generates for each labeled topic a document vector whose elements store scores of the documents for the labeled topic and a word vector whose elements store scores of the words of the vocabulary for the labeled topic. Non-negative matrix factorization (NMF) is performed to generate a document-topic model that clusters the documents into k topics where k>L. NMF factors representing L topics of the k topics are initialized to the document and word vectors for the L labeled topics. In some embodiments the NMF factors representing the L topics initialized to the document and word vectors are frozen, that is, are not updated by the NMF after the initialization.

The invention discloses an author disambiguation method based on incremental learning. The author disambiguation method comprises the following steps: obtaining a historical citation record, wherein the historical citation information has known clustering labels, and different clustering labels represent different author individuals; judging whether each clustering cluster is a clustering clusterof a first type or a clustering cluster of a second type according to the number of the historical citation records, and for the clustering clusters of the first type with a large number, training a corresponding naive Bayes classifier by using the feature vectors and clustering labels of the historical citation records; and screening out candidate clustering clusters, according to the types of all candidate clustering clusters, carrying out classification processing on the new citation records according to conditions, comprehensively using a naive Bayesian classifier to calculate the affiliated probability for classification, using the synergy person similarity to perform supplementary judgment on the affiliated probability mode classification, and calculating the semantic similarity withthe second type of clustering cluster to solve the problem that the naive Bayesian classifier cannot be used for probability classification. The author disambiguation method is good in author disambiguation effect and low in calculation overhead.

The present ivnetion relates to a power transmission and transformation equipment state prediction method and system. The method comprises: acquiring observation data of to-be-predicted equipment and constructing a characteristic sequence; acquiring a classification maximizing a probability of each characteristic vector in the characteristic sequence from a preset probability classification model; and according to the characteristic vectors and an optimal predictive coefficient corresponding to a probability classification model class of the classification maximizing the probability of each characteristic vector, performing a calculation to obtain predicted data corresponding to the to-be predicted equipment, and outputting the predicted data. The characteristic sequence is constructed according to the observation data of the to-be-predicted equipment, and the probability classification model class constructed by historical observation data of power transmission and transformation equipment is combined to obtain the corresponding optimal predictive coefficient so as to complete data prediction on the to-be-predicted equipment. According to the power transmission and transformation equipment state prediction method and system provided by the present invention, a characteristic changing trend is fully excavated from the historical data of the equipment, and the optimal predictive coefficient is adopted to predict an equipment state, so that a better prediction effect can be obtained, the method and the system can take a good effect of carrying out state monitoring, early-warning diagnosis and the like on the power transmission and transformation equipment, and reliability is improved.