Abstract: The different advantageous embodiments provide an apparatus comprising an inflatable portion, a nozzle, and a non-inflatable portion. The inflatable portion has an interior housing and an inflating system. The inflating system is disposed between a number of layers of material. An interior layer of material from the number of layers of material comprises the interior housing. The number of layers of material are flexible and impervious to penetration by air and moisture. The nozzle is connected to the inflatable portion and configured to control fluid flow to and from the inflating system. The non-inflatable portion is connected to an end of the inflatable portion and has a sealing system configured to seal the interior housing of the inflatable portion.

Abstract: A classifier is built to rank documents of different languages found in a query based at least in part on similarity to other documents and the relevance of those other documents to the query. A joint ranking model, e.g., based upon a Boltzmann machine, is used to represent the content similarity among documents, and to help determine joint relevance probability for a set of documents. The relevant documents of one language are thus leveraged to improve the relevance estimation for documents of different languages. In one aspect, a hidden layer of units (neurons) represents clusters (corresponding to relevant topics) among the retrieved documents, with an output layer representing the relevant documents and their features, and edges representing a relationship between clusters and documents.

Abstract: Described is a technology for preloading modules, such as modules that show clinical/medical data maintained at a service, so as to reduce a user's wait time to use a module. The modules for which a user is authenticated are preloaded according to a loading order that is based upon the user's historical usage data. If a user interacts to use a selected module that is not yet loaded, the selected module may be loaded immediately, independent of the order. A background thread preloads the modules according to the order. A normal thread loads the selected module, unless already being preloaded by the background thread; in either situation the loading thread's priority may be temporarily increased to expedite loading. The historical data may be in the form of weight values associated with the modules, with the weight values adjusted based upon module usage.

Abstract: Described is a technology comprising a query processing pipeline in which a SPARQL query is processed into an intermediate LINQ query, which is then processed by a LINQ provider. The LINQ provider decides which instructions correspond to flat database queries, and routes those instructions a database engine (e.g., SQL server) for querying a database. Other instructions are provided to a reasoning engine for processing, e.g., by performing a graph traversal and/or database queries. The pipeline may include a parser that parses the query into an abstract syntax tree, and an optimizer that processes the abstract syntax tree into a LINQ query, including by reordering LINQ instructions and/or associating a flag with each of the instructions that indicates whether to query the database or provide the instruction to a reasoning engine.

Abstract: Described is time-weighted blending of the results of time series algorithms in a manner that changes their relative weights based on the prediction time. The prediction values from each algorithm are mathematically blended into a forecast result corresponding to the desired time of prediction. In this manner, an ARTXP algorithm that provides accurate near term predictions is given more weight than an ARIMA for near term predictions, and less relative weight for long term predictions. An example exponential function to compute the relative weights is described; the function corresponds to a curve having a control variable for the slope and the start of the curve, and constant coefficients, with the weights based (in part) on the prediction time. A user-provided parameter may also affect the relative weights used in the blending result.

Abstract: Described is a technology by which an image is classified (e.g., grouped and/or labeled), based on multi-label multi-instance data learning-based classification according to semantic labels and regions. An image is processed in an integrated framework into multi-label multi-instance data, including region and image labels. The framework determines local association data based on each region of an image. Other multi-label multi-instance data is based on relationships between region labels of the image, relationships between image labels of the image, and relationships between the region and image labels. These data are combined to classify the image. Training is also described.

Abstract: Described is a summarizing a web entity (e.g., a person, place, product or so forth) based upon the entity's appearance in web documents (e.g., on the order of hundreds of millions or billions of webpages). Webpages are separated into blocks, which are then processed according to various features to filter the number of blocks to further process, and rank the most relevant blocks with respect to the entity that remain. A redundancy removal mechanism removes redundant blocks, leaving a set of remaining blocks that are used to provide a summary of information that is relevant to the entity.

Abstract: Described is a technology by which a single physical storage device such as a USB flash memory device is able to boot different computing devices via corresponding different operating systems. The storage device includes a selection mechanism that determines which virtual disk (corresponding to a LUN) is seen by the host as the currently active LUN having sector 0, and therefore is the boot disk. The selection mechanism also may select which (if any) other LUNs are visible to the host. The selection mechanism and accompanying indicator may be operated when the storage device is disconnected, e.g., via manual switches and/or LEDs, buttons and/or a display (e.g., via internal power). Also described is allowing each LUN to have a user-friendly name.

Abstract: Described is a technology in which face alignment data is obtained by processing an image using a component-based discriminative search algorithm. For each facial component, the search is guided by an associated directional classifier that determines how to move the facial component (if at all) to achieve better alignment relative to its corresponding facial component in the image. Also described is training of the classifiers.

Abstract: Described is a technology in which video shots are clustered based upon the location at which the shots were captured. A global energy function is optimized, including a first term that computes clusters so as to be reasonably dense and well connected, to match the possible shots that are captured at a location, e.g., based on similarity scores between pairs of shots. A second term is a temporal prior that encourages subsequent shots to be placed in the same cluster. The shots may be represented as nodes of a minimum spanning tree having edges with weights that are based on the similarity score between the shots represented by their respective nodes. Agglomerative clustering is performed by selecting pairs of available clusters, merging the pairs and keeping the pair with the lowest cost. Clusters are iteratively merged until a stopping criterion or criteria is met (e.g., only a single cluster remains).

Abstract: Described is a technology by which logged events such as in a security event log (e.g., within a defined timeframe) are processed to determine logon times and logoff times, which are then used to determine system usage. Logoff times may correspond to an actual logoff event, a shutdown event, or another login. In one example, logon and logoff times determine the combined times that each user of the system was logged on, e.g., a total session time for each user. Multiple users' times may be combined into a total usage time of all users. Each user's combined session time may be evaluated against the total usage time to determine whether any user is a primary user, e.g., by being logged on more than a threshold percentage of the total usage time. A primary user may be identified, or the system determined to be a shared system without a primary user.

Abstract: Described is noise reduction technology generally for speech input in which a noise-suppression related gain value for the frame is determined based upon a noise level associated with that frame in addition to the signal to noise ratios (SNRs). In one implementation, a noise reduction mechanism is based upon minimum mean square error, Mel-frequency cepstra noise reduction technology. A high gain value (e.g., one) is set to accomplish little or no noise suppression when the noise level is below a threshold low level, and a low gain value set or computed to accomplish large noise suppression above a threshold high noise level. A noise-power dependent function, e.g., a log-linear interpolation, is used to compute the gain between the thresholds. Smoothing may be performed by modifying the gain value based upon a prior frame's gain value. Also described is learning parameters used in noise reduction via a step-adaptive discriminative learning algorithm.

Abstract: Described is a technology by which software instrumentation data collected from user program sessions are analyzed to output an analysis report or the like via example methods and an architecture configured for efficient operation. A client component queries a service for analysis related information. To process the query, the service works with a data manager, and via a high dimensional analysis component may use information processed from the software instrumentation data, such as in the form of one or more inverted indexes and/or raw value files. The service may include a usage analysis component, a feature recognition component that locates features from command sequences, a user recognition component and/or a program reliability component. One or more counterpart components at the client may generate analysis reports or the like based on the query results. The client also may maintain user libraries and feature libraries to facilitate analyses.

Abstract: Described is automatically testing the quality of an audio coupling between juxtaposed first and second digital telephones, e.g., VoIP telephones, such as to quantitatively determine the quality of audio echo cancellers in those digital telephones. An analyzer receives timestamps from a first telephone and second telephone during a calling session, including timestamps for when the second telephone initially provides audio (e.g., speech) to the first telephone, when the first telephone initially detects sound, when the first telephone initially provides audio to the second telephone, and when the second telephone initially detects sound. The analyzer uses the relative timing of the timestamps and the speech recognizer's outcome to determine whether the audio coupling is experiencing interference or echo. When the audio includes speech, a confidence level corresponding to accuracy of speech recognition also may establish the audio coupling's quality.

Abstract: Described is a technology by which a large image is processed into a seamless, radiometric mean image containing meaningful image objects. A large source image is tiled into non-overlapping tiles, which are each segmented into regions and merged via their regions. Source image pixels are separated into tiles comprising rows and columns. Once tiled, the tiles are segmented into regions, with each region having an identifier and other associated information such as size data, boundary data and neighbor data. The tiles are merged by merging columns or rows of tiles into row blocks or column blocks, respectively. The row blocks or column blocks are vertically or horizontally merged, respectively. Merging may proceed by selecting a pair of regions based on merging cost, merging that pair, updating its associated data and data of neighbor regions based on the merge, and repeating for other pairs until merging stop criteria are met.

Abstract: Described is a technology by which a compiler's intermediate representation is generated in a way that includes type checking to ensure safe execution. Typed representation of virtual method invocation uses an exact class name to represent objects of the class but not those of the subclasses, and a combined class type as an encoding of an existential type to represent objects of the class and subclasses of the class. Typed representation of interface method invocation is also provided, as is testing whether the object can be cast to a class, to an interface and/or to a class vector. Vector store checking is also able to be type checked. The medium-level intermediate representation is translated into a low-level intermediate representation, including choosing data representation for types, and converting medium-level intermediate representation operators into instructions in the low-level representation.