Siemens Aktiengesellschaft

Siemens Corporation

Amazon Technologies, Inc.

Fabian Moerchen

Dmitriy Fradkin

Klaus Brinker

Claus Neubauer

Edward C Bueche

Toby Ray Latin-Stoermer

Brandon Scott Durham

Panagiota Karanasou

Jonathan Barnett Feinstein

Alok Verma

Amina Shabbeer

Catherine Breslin

Fabian Triefenbach

Bernhard Glomann

Klaus Reiter

Judith Gaspers

Dorin Comaniciu

Bogdan Georgescu

Puneet Sharma

Leo Parker Dirac

Amit Chakraborty

Edo Liberty

Razvan Ioan Ionasec

Ioannis Akrotirianakis

Kai Zhang

Andrey Torzhkov

Gayle M Wittenberg

Zhuang Wang

Bibudh Lahiri

Techniques are provided for training a target language model based at least in part on data associated with a reference language model. For example, language data utilized to train an English language model may be translated and provided as training data to train a German language model to recognize utterances provided in German. By utilizing the techniques herein, the efficiency of training a new language model may be improved due at least in part to replacing labor-intensive operations conventionally performed by specialized personnel with machine-generated data. Additionally, techniques discussed herein provide for reducing the time required for training a new language model by leveraging information associated with utterances of one language to train the new language model associated with a different language.

Techniques for model training for voice features

Techniques are provided for training a language recognition model. For example, a language recognition model may be maintained and associated with a reference language (e.g., English). The language recognition model may be configured to accept as input an utterance in the reference language and to identify a feature to be executed in response to receiving the utterance. New language data (e.g., other utterances) provided in a different language (e.g., German) may be obtained. This new language data may be translated to English and utilized to retrain the model to recognize reference language data as well as language data translated to the reference language. Subsequent utterances (e.g., English utterances, or German utterances translated to English) may be provided to the updated model and a feature may be identified. One or more instructions may be sent to a user device to execute a set of instructions associated with the feature.

A method of building a model for predicting failure of a machine, including parsing () daily machine event logs of one or more machines to extract data for a plurality of features, parsing () service notifications for the one or more machine to extract failure information data, creating () bags from the daily machine event log data and failure information data for multiple instance learning by grouping daily event log data into the bags based on a predetermined predictive interval, labeling each bag with a with a known failure as positive, and bags without known failures as negative, where a bag is a set of feature vectors and an associated label, where each feature vector is an n-tuple of features, transforming () the multiple instance learning bags into a standard classification task form, selecting () a subset of features from the plurality of features, and training () a failure prediction model using the selected subset of features.

Log-based predictive maintenance using multiple-instance learning

Techniques for accent translation are described herein. A plurality of audio samples may be received, and each of the plurality of audio samples may be associated with at least one of a plurality of accents. Audio samples associated with at least a first accent of the plurality of accents may be compared to audio samples associated with at least one other accent of the plurality of accents. A translation model between the first accent and a second accent may be generated. An input audio portion in a first spoken language may be received. It may be determined whether the input audio portion is substantially associated with the first accent, and if so, an output audio portion substantially associated with the second accent in the first spoken language may be outputted based, at least in part, on the translation model.

Accent translation

A method to build a failure predictive model includes: receiving an input of a set of event sequences, where each sequence is labeled as representing a failure or not representing a failure, extracting a single predictive closed pattern from among the input sequences that represents a failure, creating a root node with the single closed pattern, splitting the set of event sequences into a first set that includes the single closed pattern and a second set that excludes the single pattern, and processing each of the first and second sets until at least one child node is created that is labeled as either representing a failure or not representing a failure.

Building a failure-predictive model from message sequences

A method for identifying a plurality of patterns of events from within event log file data includes receiving a query comprising a plurality of patterns, each of the patterns comprising a plurality of events. One or more key events is determined from the plurality of patterns of events. The one or more key events is located within a database of stored event log file data. An event stream comprising the key events and all other events of the event log file data occurring within a predetermined time span from the time of the located one or more events is generated. Each of the plurality of patterns of the received query are searched for from within the event stream.

Temporal pattern matching in large collections of log messages

In a support vector regression approach to forecasting power load in an electrical grid, a feature learning scheme weights each feature in the input data with its correlation with the predicted load, increasing the prediction accuracy. The kernel matrix for the input training data is computed such that features that align better with the target variable are given greater weight. The resulting load forecast may be used to compute commands sent to demand response modules.

Short-term load forecast using support vector regression and feature learning

Systems and methods for determining critical thresholds on a number of events (k) and a window length (t) for properly defining a burst of events in a data stream. A new coverage metric Cis defined and used in the determination, where the coverage metric Cis defined for a particular pair (k,t) as a fraction, with the numerator defined a number of events that occur within some (k,t)-bursty window and the denominator defined as the total number of events (n) that occurred along the entire time span being analyzed. Coverage metric Cis monotonic non-increasing in k and monotonic non-decreasing in t, allowing for a divide-and-conquer search strategy to be used to find the critical threshold pairs (k*, t*).

Critical threshold parameters for defining bursts in event logs

A method and system for generating a patient specific anatomical heart model is disclosed. Volumetric image data, such as computed tomography (CT), echocardiography, or magnetic resonance (MR) image data of a patient's cardiac region is received. Individual models for multiple heart components, such as the left ventricle (LV) endocardium, LV epicardium, right ventricle (RV), left atrium (LA), right atrium (RA), mitral valve, aortic valve, aorta, and pulmonary trunk, are estimated in said volumetric cardiac image data. A multi-component patient specific anatomical heart model is generated by integrating the individual models for each of the heart components. Fluid Structure Interaction (FSI) simulations are performed on the patient specific anatomical model, and patient specific clinical parameters are extracted based on the patient specific heart model and the FSI simulations. Disease progression modeling and risk stratification are performed based on the patient specific clinical parameters.

Method and system for multi-component heart and aorta modeling for decision support in cardiac disease

An approach is provided for condition monitoring from log messages and sensor trends based on time semi-intervals. The approach may be applied to machine condition monitoring. Patterns are mined from symbolic interval data that extends previous approaches by allowing semi-intervals and partially ordered patterns. The semi-interval patterns and semi-interval partial order patterns are less restrictive than patterns using Allen's relations. Combinations and adaptations of efficient algorithms from sequential pattern and itemset mining for discovery of semi-interval patterns are described.

Condition monitoring with automatically generated error templates from log messages and sensor trends based on time semi-intervals

Documents in a high density data stream are clustered. Incoming documents are analyzed to find metadata, such as words in a documents headline or abstract and people, places, and organizations discussed in the document. The metadata is emphasized as compared to other words found in the document. A single feature vector for each document determined based on the emphasized metadata will accordingly take into account the importance of such words and clustering efficacy and efficiency are improved.

Method and apparatus for incorporating metadata in data clustering

Documents from a data stream are clustered by first generating a feature vector for each document. A set of cluster centroids (e.g., feature vectors of their corresponding clusters) are retrieved from a memory based on the feature vector of the document using a locality sensitive hashing function. The centroids may be retrieved by retrieving a set of cluster identifiers from a cluster table, the cluster identifiers each indicative of a respective cluster centroid, and retrieving the cluster centroids corresponding to the retrieved cluster identifiers from a memory. Documents may then be clustered into one or more of the candidate clusters using distance measures from the feature vector of the document to the cluster centroids.

Document clustering that applies a locality sensitive hashing function to a feature vector to obtain a limited set of candidate clusters

Documents from a data stream are clustered by first generating a feature vector for each document. A set of cluster centroids (e.g., feature vectors of their corresponding clusters) are retrieved from a memory based on the feature vector of the document and a relative age of each of the cluster centroids. The centroids may be retrieved by retrieving a set of cluster identifiers from a cluster table, the cluster identifiers each indicative of a respective cluster centroid, and retrieving the cluster centroids corresponding to the retrieved cluster identifiers from a memory. A list of cluster identifiers in the cluster table may be maintained based on the relative age of cluster centroids corresponding to the cluster identifiers. Cluster identifiers that correspond to cluster centroids with a relative age exceeding a predetermined threshold are periodically removed from the list of cluster identifiers.

Growing community of inventors

Rocky Hill, NJ, United States of America

Fabian Moerchen