Microsoft Technology Licensing, LLC

Other

Youssef Hamadi

Lucas Bordeaux

Shahram Izadi

Andrew Blake

Vishwa Vinay

Carsten Rother

Marc Shapiro

Claude-Guy Quimper

Stephanie Ubeda

Georg Ringwelski

Yek Loong Chong

Said Jabbour

An outgoing message monitor is provided. In an embodiment, outgoing messages are monitored to detect potential errors and alerts may be triggered. Using information about a message such as an email, a first classifier classifies the email into an expected class and a second classifier classifies the email into an actual class. On the basis of a comparison of the expected and actual classes an alert may be triggered. In an embodiment, the second classifier uses information derived from text content of the email which may optionally be pre-processed. The first classifier, for example, uses other information about the email such as its intended recipients, information about the presence of attachments, information about whether the email is part of a thread and other information.

Outgoing message monitor

Controlled constraint sharing in parallel problem solvers is described. In embodiments, constraint sharing in a problem solver is controlled by comparing a total number of shared constraints obtained in a time interval to a target, and regulating the subsequent number of shared constraints obtained from other problem solvers accordingly. Embodiments describe how the regulation of the subsequent number of shared constraints can be achieved by controlling the size of constraints shared by other problem solvers. Embodiments describe how an additive increase/multiplicative decrease algorithm can be used to determine the size of constraints to be exchanged. Embodiments also describe how the utility of the shared constraints can be determined, and used to control the size of constraints shared by other problem solvers. In embodiments, the problem solver is a Boolean satisfiability problem solver.

Controlled constraint sharing in parallel problem solvers

It is difficult for managers or other operators to decide how best to modify an available pool of resources used for their workflows. Methods for achieving this by improving the robustness of workflows are described. The more robust a workflow, the better its ability to withstand detrimental changes to its associated pool of resources. It is recognized here that the size of the solution space for the problem of allocating resources to tasks in a workflow provides a useful indicator of robustness of a workflow. In general, the greater the size of the solution space the more robust the workflow. A constraint optimization problem is specified for the problem of how best to modify resource characteristics of a pool of resources in order to maximize a size of a solution space. The solution space is that for the problem of allocating resources from the pool to tasks in a workflow.

Exploitation of workflow solution spaces to account for changes to resources

A user interface is described which displays a shape and enables a user to manipulate that shape by dividing it into a number of regions. Each region has an associated term which is made up from one or more constraints. A mathematical function, such as an objective function, is generated from the resultant graphical representation of the shape, where the function contains a term for each region and where each term is weighted according to the size of the corresponding region.

Graphical acquisition of an objective function

It is difficult to select parameter values for constraint programming problem solvers which will yield good performance. Automated tuning of such problem solvers on a per problem instance basis may be used and this involves learning a function for predicting the runtime of a problem solver depending on parameter values of the problem solver and features of the problem instance being solved. However, it takes time for such prediction functions to be learnt, either during operation of a problem solver or offline, using specified examples. To address this, information about such a prediction function is shared between two or more problem solvers to improve performance. A sharing system may be used to receive prediction function information and send this to problem solvers.

Tuning of problem solvers

It is required to provide an automated process for forming a visually appealing digital collage (also referred to as a tapestry or photomontage) from a plurality of input images. Input images and a collage are labeled. Labels in the collage specify which regions of the input images are used to form those collage regions. An energy function is created for the labeling which contains terms that take into account various criteria to produce a collage. For example, those criteria may relate to the amount of information in the images, similarity of the images, entropy in the images or object class information. A multi-stage optimization process is applied to the energy function which is fast and enables large numbers of input images to be used to produce a collage.

Auto collage

A node participating in a network of nodes that solve a distributed constraint problem. The nodes have mutually constrained sub-problems of the constraint problem. The node receives values or decisions from neighbor nodes. Those values or decisions allow the sub-problems of the neighbor nodes to be solved. Some received values or decisions correspond to a first ordering of the nodes, and some received values or decisions correspond to a second ordering of the nodes. Attempts are made to solve the sub-problem of the node as constrained by the values or decisions of the neighbor nodes. Results of the attempts are communicated to neighbor nodes, where for a given value or decision the node to which the corresponding attempt result is communicated depends on which of the orderings the given value or decision corresponds to. An attempt for one context can be affected by attempts in another context.

Multi-ordered distributed constraint search

A system includes a weighted directional dependency graph defining one or more dependency links in a distributed system having a plurality of nodes. A weight is defined for each of the dependency links. A node ordering module generates a node order based on a cycle-cut set identifying at least one of the dependency links that is part of a cycle in the weighted directional dependency graph to create an output directional dependency graph. A method for ordering a plurality of decision nodes in a distributed system includes generating a weighted directional dependency graph, determining a cycle-cut set identifying one or more dependency links that are part of a cycle in the weighted directional dependency graph, and ordering the plurality of decision nodes based on the cycle-cut set.

Growing community of inventors

Cambridge, United Kingdom

Youssef Hamadi