Microsoft Technology Licensing, LLC

Microsoft Corporation

Thore Graepel

Ralf Herbrich

Andrew Donald Gordon

David Stern

David Shaw

Yoram Bachrach

Thomas Ivan Borchert

Pushmeet Kohli

Andrew Blake

John Michael Winn

Sriram Kottarakurichi Rajamani

Christopher E Miles

Filip Andrzej Radlinski

Margaret Novotny

William Wagner

Lucas Bordeaux

Claudio Vittorio Russo

Aditya Nori

Mohammad Raza

Joaquin Quinonero Candela

Anton Schwaighofer

Jacob Miller

Johannes Borgstroem

Yelisaveta Pesenson

Aryn Shelander

Philipp Hennig

Horst Cornelius Samulowitz

Sheena Stevens

Jurgen Anne Francios Marie Van Gael

Joaquin Quiñonero Candela

Checking and/or completing for data grids is described such as for grids having rows and columns of cells at least some of which contain data values such as numbers or categories. In various embodiments predictive probability distributions are obtained from an inference engine for one or more of the cells and the predictive probability distributions are used for various tasks such as to suggest values to complete blank cells, highlight cells having outlying values, identify potential errors, suggest corrections to potential errors, identify similarities between cells, identify differences between cells, cluster rows of the data grid, and other tasks. In various embodiments a graphical user interface displays a data grid and provides facilities for completing, error checking/correcting, and analyzing data in the data grid.

Checking and/or completion for data grids

Systems and methods for inferring user traits based on indirect questions. Indirect questions may be generated based on one or more triggers. The answers to the indirect questions provide cues to a system as to whether a user has one or more attributes associated with a trait. This information may be used to personalize a computing device.

Inferring cues for use with digital assistant

Predicting user responses to items is useful in many application domains, such as personalized information retrieval and recommendation systems. In an embodiment a contacts service identifies contacts of a target user and predictions are elicited from the contacts about the target user's response to an item. In various examples, the predictions are combined taking into account weights of the contacts to produce a prediction of the target user's response. For example, the response may be one or more of: a numerical rating, a word or phrase describing the targets user's opinion of the item and a word or phrase stating a reason that the target user holds the opinion. In examples, accuracy of the predictions is calculated after observing the target user's actual response. The accuracy may be used to calculate and display scores and rankings of the contact's prediction abilities and to update the weights of the contacts.

Predicting user responses

Online recommendations are tracked through a forwarding service. The forwarding service can provide such statistics to an ad service, which can provide incentives to the recommending user and a consuming user. Example incentives may include an accumulation of points by the recommending user, a discount to the consuming user if a purchase is made in response to the recommendation, etc. To determine how much of an incentive each participant in the recommendation flow receives, a graph is created to model the recommendation flow and incentives are allocated using a cooperative game description based on this graph that associates each participant with a power index that represents that participants share of the incentive.

Pricing in social advertising

There is provided a method and system for modeling a data generating process. The method includes generating a dyadic Bayesian model including a pair of probabilistic functions representing a prior distribution and a sampling distribution, and modeling a data generating process based on the dyadic Bayesian model using observed data. The method includes generating a learner object for the dyadic Bayesian model. The method further includes training the dyadic Bayesian model with the learner object based on the observed data to produce a trained dyadic Bayesian model. The method also includes generating a posterior distribution over parameters based on the trained dyadic Bayesian model. The method also further includes generating a posterior predictive distribution based on the posterior distribution. The method also includes predicting an outcome of observable variables based on the posterior predictive distribution.

Modeling a data generating process using dyadic Bayesian models

A real-time stereo video signal of a captured scene with a physical foreground object and a physical background is received. In real-time, a foreground/background separation algorithm is used on the real-time stereo video signal to identify pixels from the stereo video signal that represent the physical foreground object. A video sequence may be produced by rendering a 3D virtual reality based on the identified pixels of the physical foreground object.

Stereo video for gaming

A recommender system may be used to predict a user behavior that a user will give in relation to an item. In an embodiment such predictions are used to enable items to be recommended to users. For example, products may be recommended to customers, potential friends may be recommended to users of a social networking tool, organizations may be recommended to automated users or other items may be recommended to users. In an embodiment a memory stores a data structure specifying a bi-linear collaborative filtering model of user behaviors. In the embodiment an automated inference process may be applied to the data structure in order to predict a user behavior given information about a user and information about an item. For example, the user information comprises user features as well as a unique user identifier.

Recommending items to users utilizing a bi-linear collaborative filtering model

Matchmaking processes at online game services often result in players having to wait unacceptably long times to receive a match or immediately receiving a poorly matched session. By using a matchmaking process which dynamically adapts a good balance is achieved between the quality of proposed matches (for example, in terms of how balanced, interesting and fun those matches are likely to be) and the waiting time for potential matches. A matchmaking threshold is specified. When a player seeks a match a waiting time is observed, for example, as to how long that player waits until starting a game or dropping out. Information about such waiting times is used to dynamically update the matchmaking threshold. The update is made on the basis of a relationship between information about the observed waiting time and a target waiting time. Further control may be achieved by using separate matchmaking thresholds and target waiting times for different game categories.

Adaptive matchmaking for games

Machine learning techniques may be used to train computing devices to understand a variety of documents (e.g., text files, web pages, articles, spreadsheets, etc.). Machine learning techniques may be used to address the issue that computing devices may lack the human intellect used to understand such documents, such as their semantic meaning. Accordingly, a topic model may be trained by sequentially processing documents and/or their features (e.g., document author, geographical location of author, creation date, social network information of author, and/or document metadata). Additionally, as provided herein, the topic model may be used to predict probabilities that words, features, documents, and/or document corpora, for example, are indicative of particular topics.

Topic models

Managing a portfolio of experts is described where the experts may be for example, automated experts or human experts. In an embodiment a selection engine selects an expert from a portfolio of experts and assigns the expert to a specified task. For example, the selection engine has a Bayesian machine learning system which is iteratively updated each time an experts performance on a task is observed. For example, sparsely active binary task and expert feature vectors are input to the selection engine which maps those feature vectors to a multi-dimensional trait space using a mapping learnt by the machine learning system. In examples, an inner product of the mapped vectors gives an estimate of a probability distribution over expert performance. In an embodiment the experts are automated problem solvers and the task is a hard combinatorial problem such as a constraint satisfaction problem or combinatorial auction.

Managing a portfolio of experts

Event prediction in dynamic environments is described. In an embodiment a prediction engine may use the learnt information to predict events in order to control a system such as for internet advertising, email filtering, fraud detection or other applications. In an example one or more variables exists for pre-specified features describing or associated with events and each variable is considered to have an associated weight and time stamp. For example, belief about each weight is represented using a probability distribution and a dynamics process is used to modify the probability distribution in a manner dependent on the time stamp for that weight. For example, the uncertainty about the associated variable's influence on prediction of future events is increased. Examples of different schedules for applying the dynamics process are given.

Event prediction in dynamic environments

Reputation systems have been used to promote trust between participants in activities including online activities such as online market places. Existing online market places provide a reputation system which is a simple cumulative registry of user ratings on a given market place member. However, this simple system is open to abuse in situations where, for example, many positive ratings are given in a fraudulent manner. By modeling both reputation of participants and required reputation of participants an improved reputation system is provided. The required reputation may be thought of as a threshold, referred to herein as a required threshold, which may be used in determining how to update an indication of the reputation of a participant in the activity. The reputation system is able to learn information about required reputation and reputation of participants using an update process which is robust to participants who consistently give feedback of a particular type.

Reputation system

Machine learning using relational databases is described. In an embodiment a model of a probabilistic relational database is formed by augmenting relation schemas of a relational database with probabilistic attributes. In an example, the model comprises constraints introduced by linking the probabilistic attributes using factor statements. For example, a compiler translates the model into a factor graph data structure which may be passed to an inference engine to carry out machine learning. For example, this enables machine learning to be integrated with the data and it is not necessary to pre-process or reformat large scale data sets for a particular problem domain. In an embodiment a machine learning system for estimating skills of players in an online gaming environment is provided. In another example, a machine learning system for data mining of medical data is provided. In some examples, missing attribute values are filled using machine learning results.

Machine learning using relational databases

A scalable clustering system is described. In an embodiment the clustering system is operable for extremely large scale applications where millions of items having tens of millions of features are clustered. In an embodiment the clustering system uses a probabilistic cluster model which models uncertainty in the data set where the data set may be for example, advertisements which are subscribed to keywords, text documents containing text keywords, images having associated features or other items. In an embodiment the clustering system is used to generate additional features for associating with a given item. For example, additional keywords are suggested which an advertiser may like to subscribe to. The additional features that are generated have associated probability values which may be used to rank those features in some embodiments. User feedback about the generated features is received and used to revise the feature generation process in some examples.

Scalable clustering

Players in a gaming environment, particularly, electronic on-line gaming environments, may be scored relative to each other or to a predetermined scoring system. The scoring of each player may be based on the outcomes of games between players who compete against each other in one or more teams of one or more players. Each player's score may be represented as a distribution over potential scores which may indicate a confidence level in the distribution representing the player's score. The score distribution for each player may be modeled with a Gaussian distribution and may be determined through a Bayesian inference algorithm. The scoring may be used to track a player's progress and/or standing within the gaming environment, used in a leaderboard indication of rank, and/or may be used to match players with each other in a future game. The matching of one or more teams in a potential game may be evaluated using a match quality threshold which indicates a measure of expected match quality that can be related to the probability distribution over game outcomes.

Growing community of inventors

Cambridge, United Kingdom

Thore Graepel