Microsoft Technology Licensing, LLC

Ashish Kapoor

Ranveer Chandra

Eric J Horvitz

Desney S Tan

Bongshin Lee

Sudipta Narayan Sinha

Gang Hua

Simon John Baker

Deepak Vasisht

Jie Jerry Liu

Neel Suresh Joshi

James Anthony Fogarty

Dahua Lin

Steven Mark Drucker

Krysta Marie Svore

Amar Phanishayee

Saleema Amin Amershi

Nathan O Wiebe

Huixuan Tang

Simon A J Winder

Debadeepta Dey

Xinxin Jin

Madhusudhan Gumbalapura Sudarshan

Sing Bing Kang

Ross Garrett Cutler

Philip Andrew Chou

Patrice Y Simard

Mary P Czerwinski

Eric P Traut

Nebojsa Jojic

Paul William Garnett

Ratul Mahajan

Sumit Basu

Martin Henri Roetteler

Semiha Ece Kamar Eden

Jin A Suh

Amir Akbarzadeh

Andrey Kolobov

Byungki Byun

Kayur Dushyant Patel

Richard A Caruana

Aakanksha Chowdhery

Blaine S Christian

Marcos De Moura Campos

Shital Shah

Babak Seyed Aghazadeh

Robin Ashok Kothari

Mariya Zheleva

Sudheendra N Vijayanarasimhan

Pradeep U Shenoy

Mike Roberts

Juan Caicedo Rueda

Justin Talbot

Talal Ahmad

Sami Tariq Mian

Michael J Brooks

Sai Hemachandra Vemprala

Jongho Won

Daochen Wang

Aarthi Meenakshi Sundaram

Dorsa Sadigh

Jie Liu

Nathan Wiebe

Simon AJ Winder

A computing device is provided, including a processor configured to transmit, to a quantum coprocessor, instructions to encode a Markov decision process (MDP) model as a quantum oracle. The processor may be further configured to train a reinforcement learning model at least in part by transmitting a plurality of superposition queries to the quantum oracle encoded at the quantum coprocessor. Training the reinforcement learning model may further include receiving, from the quantum coprocessor, one or more measurement results in response to the plurality of superposition queries. Training the reinforcement learning model may further include updating a policy function of the reinforcement learning model based at least in part on the one or more measurement results.

Reinforcement learning with quantum oracle

A computer implemented method for controlling a system moving through an environment includes receiving a stream of event data from an event camera, the stream of event data representing a pixel location, a time stamp, and a polarity for each event detected by the event camera. A compressed representation of the stream of data is generated. The compressed representation is provided to a neural network model trained on prior compressed representations using reinforcement learning to learn actions for controlling the system. A control action is generated via the neural network model to control the movement of the system.

Event camera based navigation control

A method of training a machine learning system. The method comprises collecting a first simulation dataset derived from a computer simulating a hypothetical scenario with a first simulation configuration having a first degree of fidelity. The method further comprises collecting a second simulation dataset derived from a computer simulating the hypothetical scenario with a second simulation configuration having a second degree of fidelity different than the first degree of fidelity. The method further comprises building a multi-fidelity training dataset including training data from both the first simulation dataset and the second simulation dataset according to an interleaving protocol.

Multi-fidelity simulated data for machine learning

A method described herein includes receiving a digital image, wherein the digital image includes a first element that corresponds to a first domain and a second element that corresponds to a second domain. The method also includes automatically assigning a label to the first element in the digital image based at least in part upon a computed probability that the label corresponds to the first element, wherein the probability is computed through utilization of a first model that is configured to infer labels for elements in the first domain and a second model that is configured to infer labels for elements in the second domain. The first model receives data that identifies learned relationships between elements in the first domain and elements in the second domain, and the probability is computed by the first model based at least in part upon the learned relationships.

Image tagging based upon cross domain context

Boltzmann machines are trained using an objective function that is evaluated by sampling quantum states that approximate a Gibbs state. Classical processing is used to produce the objective function, and the approximate Gibbs state is based on weights and biases that are refined using the sample results. In some examples, amplitude estimation is used. A combined classical/quantum computer produces suitable weights and biases for classification of shapes and other applications.

Quantum deep learning

Various technologies described herein pertain to generating control inputs for a cyber-physical system. A prediction concerning a phenomenon can be generated, utilizing a classifier, based on sensor data acquired by a sensor. The prediction can include a probability distribution over a set of possible values of the phenomenon, where the phenomenon pertains to the cyber-physical system or an environment in which the cyber-physical system operates. Control inputs for the cyber-physical system that satisfy constraints that maintain safe operation of the cyber-physical system in the environment can be synthesized. The constraints can be based on the prediction that includes the probability distribution over the set of possible values of the phenomenon. Further, the cyber-physical system can be caused to operate in the environment based on the control inputs.

Control of cyber-physical systems under uncertainty

Nearest neighbor distances are obtained by coherent majority voting based on a plurality of available distance estimates produced using amplitude estimation without measurement in a quantum computer. In some examples, distances are Euclidean distances or are based on inner products of a target vector with vectors from a training set of vectors. Distances such as mean square distances and distances from a data centroid can also be obtained.

Method and system for computing distance measures on a quantum computer

Examples of the present disclosure relate to generating optimal scanning trajectories for 3D scenes. In an example, a moveable camera may gather information about a scene. During an initial pass, an initial trajectory may be used to gather an initial dataset. In order to generate an optimal trajectory, a reconstruction of the scene may be generated based on the initial data set. Surface points and a camera position graph may be generated based on the reconstruction. A subgradient may be determined, wherein the subgradient provides an additive approximation for the marginal reward associated with each camera position node in the camera position graph. The subgradient may be used to generate an optimal trajectory based on the marginal reward of each camera position node. The optimal trajectory may then be used by to gather additional data, which may be iteratively analyzed and used to further refine and optimize subsequent trajectories.

Optimal scanning trajectories for 3D scenes

A device can include control channel receiver circuitry to receive airborne vehicle control channel packets, decode circuitry to determine contents of the airborne vehicle control channel packets, transceiver circuitry to provide uplink to and receive downlink data from an airborne vehicle, processing circuitry, and a program for execution by the processing circuitry to perform operations comprising determining, based on data from the receiver circuitry, a received signal strength (RSS) of a signal from each of a plurality of airborne vehicles, determining, for each of the airborne vehicles and based on decoded data from the decode circuitry, a length of time the airborne vehicle will be within transmission range of the transceiver circuitry, determining, for each of the airborne vehicles and based on the determined RSS, determined length of time, and a determined bit-rate, an association metric, and causing association with the airborne vehicle associated with the greatest association metric.

Network access from airborne vehicle

A gateway that may be implemented in a local network and that communicates with a cloud network to provide efficient services in a weakly connected setting is disclosed. The gateway may be configured to enable services that efficiently utilize resources in both of the gateway and the cloud network, and provide a desired quality of service while operating in a weakly connected setting. The gateway may provide data collection and processing, local network services, and enable cloud services that utilize data collected and processed by the gateway. The local network may include one or more sensors and/or video cameras that provide data to the gateway. In a further implementation, the gateway may determine an allocation of one or more tasks of a service between the gateway and a cloud network by determining the allocation of the one or more service tasks based on desired service latency.

IoT gateway for weakly connected settings

An apparatus for generating precision maps of an area is disclosed. The apparatus receives sensor data, where the sensor data includes sensor readings each indicating a level of a parameter in one of a plurality of first portions of an area, and video data representing an aerial view of the area. The sensor data may be received from sensors that are each deployed in one of the first portions of the area. The video data may be received from an aerial vehicle. An orthomosaic may be generated from the video data, and the orthomosaic and the sensor data used to generate a predication model. The prediction model may then be used to extrapolate the sensor data to determine a level of the parameter in each of a plurality of second portions of the area. A precision map of the area may be generated using the extrapolated sensor readings.

Generating real-time sensor maps from videos and in-ground sensor data

Disclosed herein are technologies directed to a feature ideator. The feature ideator can initiate a classifier that analyzes a training set of data in a classification process. The feature ideator can generate one or more suggested features relating to errors generated during the classification process. The feature ideator can generate an output to cause the errors to be rendered in a format that provides for an interaction with a user. A user can review the summary of the errors or the individual errors and select one or more features to increase the accuracy of the classifier.

Error-driven feature ideation in machine learning

Growing community of inventors

Kirkland, WA, United States of America

Ashish Kapoor