Microsoft Technology Licensing, LLC

Microsoft Corporation

Sumit Basu

Daniel Morris

Chad Gibson

Lucretia Henrica Vanderwende

Adil Ahmed Sherwani

Ian Simon

Jeremiah Wander

Steven Mark Drucker

Patrick Markus Baudisch

John David Dunagan

Nebojsa Jojic

Danyel Aharon Fisher

Gregory Ray Smith

Maneesh Agrawala

Charles Edward Jacobs

T Scott Saponas

Dan Morris

Junyi Li

Andrew David Wilson

John Carlton Platt

Desney S Tan

Shwetak N Patel

Kentaro Toyama

Nicolas Villar

Gonzalo Alberto Ramos

Jiahe Helen Wang

Michael Gamon

Ashish Kapoor

Hao Lu

Tim Paek

David H Salesin

Sidhant Gupta

Dengyong Zhou

Jonathan Bernard Lester

Matthew Francis Hurst

Gabriel Adam Cohn

Karthik Gopalratnam

Trausti Thor Kristjansson

Ronald Keith Logan

Ronald E Paulsen

Sailaja Malladi

Yi Mao

Igor Labutov

Dmitriy A Belenko

Mauricio Gonzalez De La Fuente

Alan L Ritter

Scott T Saponas

A method for determining a risk of decompensated heart failure in a user includes receiving a first set of data that is fixed with respect to time. A machine-learning model generates one or more initial risk factors based on the first set of data. A second set of data for the user that dynamically updates over time is received from a wearable cardiovascular physiology monitor. The machine-learning model is used to generate dynamic data classifiers based on the one or more initial risk factors. Aggregate risk scores for the user are then indicated based on an evaluation of the second set of data against the dynamic data classifiers. In this way, static electronic medical records may be combined with dynamic, real-time data from wearable cardiovascular physiology monitors to provide an accurate and continuously updating risk of decompensated heart failure for a user.

Machine-learning models for predicting decompensation risk

Questions about a passage of text that includes a sequence of two or more sentences are generated. Each question covers the content of a plurality of sentences in the passage, and includes a context portion of the passage and a question statement that is contextually related to the context portion of the passage. A user is also provided with questions about a passage of text they are reading. Each question is presented to the user, where this presentation includes displaying the context portion of the passage and the question statement that is contextually related to the context portion of the passage.

Generating high-level questions from sentences

A first data window of a pulse waveform signal comprising a first number of samples is analyzed to determine a level of confidence that a pulse sensing device is placed correctly. If an initial level of confidence is met, the user is given positive feedback, and a second data window of a pulse waveform signal comprising a second, larger number of samples is analyzed. If an increased level of confidence is met, the user is given increased positive feedback. If a level of confidence is not met, the user is given negative feedback. If a final level of confidence is met, the user is given feedback that the pulse sensing device is placed correctly.

Wearable pulse sensing device signal quality estimation

According to one aspect of the present disclosure, a method for estimating blood pressure is provided, comprising training a machine learning model on a cohort data set. The cohort data set may include subject-specific contextual data, time-varying features, and blood pressure measurements for a plurality of subjects. The method may include receiving contextual data for a specific subject, wherein the contextual data includes medical history data of the subject. The method may further include personalizing the machine learning model to the subject based on the contextual data. The method may include calibrating the machine learning model to the subject based on a set of time-varying features and blood pressure measurements of the subject. In addition, the method may include using the machine learning model and the time-varying features for the subject to generate a blood pressure estimate.

Blood pressure estimation by wearable computing device

A method and system, are provided herein, for automatic selection of one or more items (e.g., files, emails, etc.) in a user interface, based upon selection data received from a user (hard labels) and possibly also predicted (soft) labels (e.g., from a label prediction algorithm) based on behavior data associated with the selections. More particularly, a user selects and deselects items as positive and negative examples that are used as explicit labels to form an item selection list. The explicit labels are used possibly along with additional, predicted labels to create a list of generated items the user may want to select. This list is returned to the user interface as automatically selected items. In one example, the selection list may be iteratively updated upon receiving subsequent user selection data (e.g., user may select items incorrectly selected) by determining alternative items that the user may be interested in selecting.

Items selection via automatic generalization

A wrist-worn pressure sensing device includes a pressure sensor. The wrist-worn pressure sensing device also includes a first strap that sets the position of the pressure sensor on a wearer's wrist and a second strap that engages with the first strap to adjust the overall length of the strap without moving the set position of the pressure sensor on the wearer's wrist.

Sizable wrist-worn pressure sensing device

Generating responses to input utilizing an ontology-crowd-relevance methodology is described. The techniques described herein access a plurality of data items and determine an ontology associated with the plurality of data items. The ontology includes one or more ontological elements. Furthermore, the techniques describe sending, to a plurality of devices, a request to generate response templates based on the one or more ontological elements and receiving, from the plurality of devices, the response templates directed to the one or more ontological elements.

Ontology-crowd-relevance deep response generation

A system and method infer true labels for multiple items. The inferred labels are generated from judgments. Multiple judges select the judgments from a specified choice of labels for each item. The method includes determining a characterization of judge expertise and item difficulties based on the judgments. The method also includes determining, using maximum entropy, a probability distribution over the specified choice of labels for each judge and item, based on the judgments. The method further includes selecting improved labels for the items from the specified choice such that the entropy over the probability distribution is reduced. The improved labels represent an improvement from the judgments toward the true labels. Additionally, the method includes performing iterative procedure to determine the true labels, the characterizations of judge expertise and the labeling difficulties.

Learning with noisy labels from multiple judges

An overwhelming number of articles are available everyday via the internet. Unfortunately, it is impossible to peruse more than a handful, and it is difficult to ascertain an article's social context. The techniques disclosed herein address this problem by harnessing implicit and explicit contextual information from social media. By extracting text surrounding a hyperlink to an article in a post and assessing the article as a function of content surrounding the hyperlink, an article's social context is determined and presented. Additionally, articles that are sufficiently similar in content may be grouped to establish a many-to-one relationship between posts and an article, creating a more accurate assessment.

Providing context for web articles

Assisted clustering systems and methods are described herein that provide a user interface by which a user can easily create clusters and selectively associate data items with such clusters. Information regarding data item-cluster associations made by the user is processed by a recommendation engine to learn a clustering model. The clustering model is then be used to generate recommendations for the user regarding which unassociated data items should be associated with which clusters. In certain embodiments, after the user has determined that the clustering model is performing at a satisfactory level based on the quality of the recommendations, the user can cause the system to automatically cluster a large quantity of remaining unassociated data items. In accordance with further embodiments, a user can specify arbitrary data item types for clustering as well as features of such data types that should be considered in generating the clustering model.

Assisted clustering

Technology is described for diagnosing problem causes in complex environments by using factorization of a plurality of features. An embodiment can include the operation of identifying a plurality of entities having entity weighting parameters. The entities may be computing devices. The plurality of features can be associated with a respective entity having feature weighting parameters, and an instance of the plurality of features can be associated with individual entity instances. A fault label can be applied for an ensemble entity. The plurality of features can be linked using the feature weighting parameter and the entity weighting parameter with a bilinear model. A further operation is estimating weighting values for the entity weighting parameters and the feature weighting parameters for use in a statistical model. The meaningful feature parameters can be found for the statistical model that are likely to be responsible for entity faults.

Diagnosis of problem causes using factorization

After an initial training session, a 'Dynamic Echo Canceller' (DEC) provides echo cancellation where only access to an input signal and a composite output signal are available, and the input signal is subjected to an unknown variable gain function. In one embodiment, the DEC uses echo cancellation to provide a 'clean' copy of a second input signal where only a first input signal and a composite of the first and second input signal is available. An example is a &#x201c;black box&#x201d; amplifier coupled to a microphone and a phone line, with access to only the microphone input and a combined output signal where it is desired to retrieve a clean copy of a remote caller signal from the combined output. The DEC is applicable to many fields, including: signal separation; cancellation of echoes caused by impedance mismatches, periodic electrical noise, acoustic echoes caused by acoustic coupling, etc.

Echo cancellation for channels with unknown time-varying gain

A graphical user interface for facilitating generation of an accompaniment for a recorded audio melody is described. A Hidden Markov Model, trained with blended chord transition matrices and melody observation matrices, is used for providing the accompaniment for the recorded audio melody. The recorded audio melody includes segments. Frequency analysis of the recorded audio melody is performed. The total duration of each fundamental frequency within a segment of the melody is summed. Based on the summing, a probability for each possible chord for each segment is computed. Based on the computed probabilities, a set of chords are selected for the segments. The chords are displayed on a chord chart of the graphical user interface. The graphical user interface facilitates various manipulations using the chords and/or controls, and generation of a new accompaniment for a recorded audio melody based on the manipulations.

Automatic accompaniment for vocal melodies

Relating higher-level descriptive musical metadata to lower-level musical elements to enable creation of a song map, song model, backing track, or the like. The musical elements are queried based on input metadata to create a set of musical elements of varying types such as notes, chords, song structures, and the like. The set of musical elements is provided to a user for selection of particular musical elements The selected musical elements represent the song model.

Metadata-based song creation and editing

Growing community of inventors

Seattle, WA, United States of America

Sumit Basu