International Business Machines Corporation

Nuance Communications, Inc.

Lisa Amini

Olivier Verscheure

Chitra Venkatramani

Pascal Frossard

Deepak Srinivas Turaga

Peter Hans Westerink

Henrique Andrade

Jeremy Silber

Anshul Sehgal

Joel Leonard Wolf

Ching-Yung Lin

Bugra Gedik

Yoonho Park

Srinivasan Parthasarathy

Daby Mousse Sow

Mitchell Adam Cohen

Justin Wai-chow Wong

Oliver Verscheure

Philip Shilung Yu

Wei Fan

Chuang Gan

Dakuo Wang

Douglas Lee Schales

Reiner Sailer

Nagui Halim

Min Wang

Eric P Bouillet

Josyula R Rao

James R Giles

Archan Misra

Michail Vlachos

Kirsten Weale Hildrum

William Russell Belknap

Martin G Kienzle

Wim De Pauw

Mihai Christodorescu

Marion Lee Blount

Deepak Rajan

Horst Cornelius Samulowitz

Rohit Wagle

John Sidney Davis, Ii

Perry G Hartswick

Abel Nicolas Valente

Udayan Khurana

Brian Christopher Barker

Brian Douglas Pfeifer

Daniel Karl I Weidele

Wietse Z Venema

Philippe L Selo

Edward Sumner Begle

Effrosyni Kokiopoulou

Pierre Vandergheynst

Arunima Chaudhary

Pascal Frossad

Brian K Foo

Gerett Raybourn Yocum

Henrique C M Andrade

Jorge O Lepre

Carolina Maria Spina

A dataset including features and values associated with the features can be received. Each of the features in the dataset can be mapped to a corresponding node in a knowledge graph based on the concept represented by the corresponding node. The knowledge graph can be traversed to find a candidate node connected to at least one mapped node, the candidate node not being mapped to a feature in the dataset. A concept associated with the candidate node can be identified as a new feature. A machine learning model pipeline can use the features in the dataset and the new feature to select a subset of features for training a machine learning model.

Interactive feature engineering in automatic machine learning with domain knowledge

A method for speech recognition, the method includes: extracting time-frequency speech features from a series of reference speech elements in a first series of sampling windows; aligning reference speech elements that are not of equal time span duration; constructing a common subspace for the aligned speech features; determining a first set of coefficient vectors; extracting a time-frequency feature image from a test speech stream spanned by a second sampling window; approximating the extracted image in the common subspace for the aligned extracted time-frequency speech features with a second coefficient vector; computing a similarity measure between the first and the second coefficient vector; determining if the similarity measure is below a predefined threshold; and wherein a match between the reference speech elements and a portion of the test speech stream is made in response to a similarity measure below a predefined threshold. The said reference speech elements correspond to a keyword of interest, wherein Simultaneous Orthogonal Matching Pursuit (SOMP) is used in their alignment.

Method and system for robust pattern matching in continuous speech for spotting a keyword of interest using orthogonal matching pursuit

Techniques for securely and adaptively delivering multimedia content are disclosed in which a set of alternate access units for each time slot is obtained. Then, the encryption stream index of each access unit from the set of alternate access units of the previous time slot are obtained. An encryption stream index is then assigned to each access unit in the set of alternate access units in the current time slot, such that the encryption index increases over time. Thus, the invention overcomes the problem of encrypting a multimedia stream that may have multiple access units for each time slot by selecting the encryption index for each access unit such that the encryption index increases, regardless of which access unit the delivery system (e.g., server) selects for transmission.

Methods and apparatus for secure and adaptive delivery of multimedia content

Performing adaptive cyber-security analytics including a computer implemented method that includes receiving a report on a network activity. A score responsive to the network activity and to a scoring model is computed at a computer. The score indicates a likelihood of a security violation. The score is validated and the scoring model is automatically updated responsive to results of the validating. The network activity is reported as suspicious in response to the score being within a threshold of a security violation value.

Adaptive cyber-security analytics

A system and method for determining data usage based on provenance information, in a stream-processing system, includes progressively setting usage information for output stream data objects (SDOs), determining input SDOs that an output SDO depends on, based on a provenance dependency function; recursively feeding back the usage information for a subset of SDOs that can be discarded; and discarding the subset of SDOs. A system and method for data retention based on usage information, in a stream-processing system, includes managing retention of SDOs by deleting SDOs that are determined to be of null usage; and enhancing retention characteristics of SDOs that are deemed to have usage.

Systems and methods for using provenance information for data retention in stream-processing

Disclosed are a method and a computer program storage product for dynamically stabilizing a stream processing system. The method includes receiving at least one computing resource allocation target. A plurality of downstream processing elements and an upstream processing element are associated with at least one input buffer. Each of the downstream processing elements consumes data packets produced by the upstream processing element received on an output stream associated with the upstream processing element. A fastest input rate among each downstream processing element in the plurality of downstream processing elements is identified. An output rate of the upstream processing element is set to the fastest input rate that has been determined for the plurality of downstream processing elements.

Dynamic stabilization for a stream processing system

A computer-implemented method for configuring a classifier structure includes receiving a classifier structure for identifying a feature of input data, adding a plurality of operating points to a parent binary classifier of the classifier structure, and setting the operating points to minimize a total cost based on resources available to the classifier structure. Each operating point controls how data is classified by the parent binary classifier for distribution to the child binary classifiers.

Configuring individual classifiers with multiple operating points for cascaded classifier topologies under resource constraints

A keyed delay is used in the processing of streaming data to decrease the processing performed and the output provided. A first event, within a particular window, having a particular key starts a delay condition. Arriving events with the same key replace the previous arrival for that key until the delay condition is satisfied. In response thereto, the latest event with that key is output.

Processing of streaming data with a keyed delay

Techniques for dynamically modifying inter-connections between components in an application are provided. The techniques include receiving a data producer profile for each output port within a software application to be executed on one or more processors, receiving a data subscription profile for each input port of each component of the application, establishing connections between the output ports and the input ports of the components in the application based on a comparison of each data producer profile and each data subscription profile, executing the application on one or more processors to process streams of data, receiving either or both of a new data producer profile or a new data subscription profile during the execution of the application, and establishing at least one new connection between an output port and an input port based upon a revised comparison of the received data profiles that include the new data profile.

Dynamically composing data stream processing applications

A technique for identifying a defect in an object produced by a controllable process. A first type of data generated as a result of production of the object by the controllable process is obtained. A second type of data generated as a result of production of the object by the controllable process is obtained. The first type of data and the second type of data are jointly analyzed. A defect is identified in the object based on the joint analysis of the first type of data and the second type of data. By way of example, the controllable process comprises a semiconductor manufacturing process such as a silicon wafer manufacturing process and the object produced by the semiconductor manufacturing process comprises a processed wafer. The first type of data comprises tool trace data and the second type of data comprises wafer image data. The tool trace data is generated by a photolithographic tool.

Multi-modal data analysis for defect identification

Disclosed are a method, upstream processing node, and computer readable medium for dynamically stabilizing a stream processing system. The method includes receiving at least one computing resource allocation target. The method further includes determining that an input data flow rate of at least one upstream processing element varies. The computing resource is dynamically allocated to the upstream processing element in response to the input rate of the upstream processing element varying. Data flow is dynamically controlled between the upstream processing element and at least one downstream processing element.

A technique for analyzing two or more data streams respectively generated from two or more components of a controllable process includes the following steps. In a first step, a statistical analysis is performed on each of the two or more data streams to generate first analysis results in the form of respective statistical results for the two or more data streams. In a second step, at least a portion of the statistical results from at least one of the two or more data streams is combined with at least a portion of the statistical results from at least another one of the two or more data streams to yield second analysis results. The controllable process is adjustable based on at least one of the first analysis results and the second analysis results.

Hierarchical and incremental multivariate analysis for process control

Systems and methods for adaptive signal sampling and sample quantization for resource-constrained stream processing. Exemplary embodiments include a signal sampling and signal quantization method for a data stream from a ECG sensor into a pervasive device, the method including adaptively sampling the data stream, compressing data samples from the data stream, reducing the number of samples via quantization, defining a utility function for signal reconstruction, and formulating optimization on the reconstructed signal that is jointly sampled and quantized.

Systems and methods for adaptive signal sampling and sample quantization for resource-constrained stream processing

A system and method for dynamically building applications for stream processing includes providing processing elements with a flow specification describing each input and a stream description describing each output such that the flow specification indicates a stream or streams which are to be received based on processing information and the stream descriptions indicate the processing information. Processing elements that can be reused are identified by determining equivalence between the processing elements. Processing elements that are new and are not reusable are instantiated in a flow graph. An application is dynamically composed, using the instantiated processing elements by routing available streams to the instantiated processing elements in accordance with the flow specifications.

Method and system for dynamic application composition in streaming systems

A publisher-subscriber system includes a broker configured to receive and distribute at least one data stream from publishers to subscribers in accordance with subscriptions. An annotator/classifier is configured to annotate or classify the data stream by employing one or more additional stream-level attributes to create an annotated data stream wherein the annotated data stream applies the stream-level attribute to all messages therein. Subscribers can subscribe to the annotated data stream created by the annotator/classifier.

Method and system for supporting dynamic stream attributes

Growing community of inventors

Yorktown Heights, NY, United States of America

Lisa Amini