Abstract: Processes and systems are directed to training a neural network of an object recognition system. The processes and systems record video streams of people. Sequences of object images are extracted from each video stream, each sequence of object images corresponding to one of the people. A triplet comprising an anchor feature vector and a positive feature vector of the same object and a negative feature vector of a different object of feature vectors are formed for each sequence of object images. The anchor, positive, and negative feature vectors of each triplet are separately input to the neural network to compute corresponding output anchor, positive, and negative vectors. A triplet loss function value computed from the output anchor, positive, and negative vectors. When the triplite loss function value is greater than a threshold, the neural network is retrained using the anchor and positive feature vectors of the sequences of object images.

Abstract: Processes and systems are directed to training a neural network of an object recognition system. The processes and systems record video streams of people. Sequences of object images are extracted from each video stream, each sequence of object images corresponding to one of the people. A triplet comprising an anchor feature vector and a positive feature vector of the same object and a negative feature vector of a different object of feature vectors are formed for each sequence of object images. The anchor, positive, and negative feature vectors of each triplet are separately input to the neural network to compute corresponding output anchor, positive, and negative vectors. A triplet loss function value computed from the output anchor, positive, and negative vectors. When the triplite loss function value is greater than a threshold, the neural network is retrained using the anchor and positive feature vectors of the sequences of object images.

Abstract: A method for combining data values through associative operations. The method includes, with a processor, arranging any number of data values into a plurality of columns according to natural parallelism of the associative operations and reading each column to a register of an individual processor. The processors are directed to combine the data values in the columns in parallel using a first associative operation. The results of the first associative operation for each column are stored in a register of each processor.

Abstract: Methods, database management systems (“DBMS”) and computer-readable media are provided for processing unbounded stream data using a traditional DBMS. Execution of a query that includes a data stream as a data source may be initiated. Tuples may be processed in accordance with the query as the tuples are received through the data stream until an indication is received that execution of the query should cease.

Abstract: A processing method and a processing apparatus for a single-channel convolution layer, and a processing method and apparatus for a multi-channel convolution layer are provided. The processing method for a single-channel convolution layer includes following steps. Data to be processed is divided into a plurality of data blocks, the plurality of data blocks are read by a plurality of graphic processor thread groups into local memories of the plurality of graphic processor thread groups, a plurality of filters are read by the plurality of graphic processor thread groups into the local memories of the plurality of graphic processor thread groups and convolutions of corresponding data points in the plurality of data blocks and the plurality of filters simultaneously are calculated by a plurality of threads in the plurality of graphic processor thread groups.

Abstract: One embodiment is a method that uses MapReduce and Relation Valued Functions (RVFs) with parallel processing to search a database and obtain search results.

Abstract: A processing method and a processing apparatus for a single-channel convolution layer, and a processing method and apparatus for a multi-channel convolution layer are provided. The processing method for a single-channel convolution layer includes following steps. Data to be processed is divided into a plurality of data blocks, the plurality of data blocks are read by a plurality of graphic processor thread groups into local memories of the plurality of graphic processor thread groups, a plurality of filters are read by the plurality of graphic processor thread groups into the local memories of the plurality of graphic processor thread groups and convolutions of corresponding data points in the plurality of data blocks and the plurality of filters simultaneously are calculated by a plurality of threads in the plurality of graphic processor thread groups.

Abstract: A representation of a flow network having vertices connected by arcs is provided. The vertices include a first set of vertices that provide flow to a second set of vertices over arcs connecting the first set and second set of vertices. A balancing procedure in the network is performed that includes redistributing flows on arcs incident on the second set of vertices. The balancing procedure includes selecting a batch of the vertices in the second set, and redistributing flows on arcs incident on the selected batch of vertices. The selecting and redistributing are repeated for other batches of vertices in the second set.

Abstract: The current application discloses a database management system that provides multiple-input, multiple-output-per-input user-defined-function-based operations. The database management system comprises at least one processor and electronic memory, a database-query processor, executed on a computer processor controlled by computer instructions stored in a computer-readable memory, that makes multiple calls to a multiple-input, multiple-output-per-input user-defined-function, in each call transmitting a next input to the multiple-input, multiple-output-per-input user-defined-function, and the multiple-input, multiple-output-per-input user-defined-function, executed on a computer processor controlled by computer instructions stored in a computer-readable memory, that uses three different memory buffers, the contents of which are maintained for three different time periods, to compute and return to the database-query processor multiple outputs in response to at least one of the multiple inputs.

Abstract: Systems and methods of ranking color correction processes are disclosed. An example method includes processing subimages of an image using a plurality of color correction processes. The method also includes ranking the plurality of results of color correction processes across the subimages. The method also includes applying color correction to the image based on the ranking of the color correction processes.

Abstract: A color constancy method and system include dividing an image into a plurality of sub-images and applying a plurality of color constancy algorithms to each of the sub-images. The outputs of each of the color constancy algorithms are analyzed for each of the sub-images to determine which of the color constancy algorithms give inconsistent results across the sub-images. The influence of the outputs of the algorithms providing inconsistent results is adjusted to decrease their influence (e.g. effect or weight) with respect to the outputs of algorithms providing consistent results. The outputs from the plurality of color constancy algorithms are combined based upon the adjustment of the outputs.

Abstract: A device for classifying a document comprises a module to generate a data tree structure and configured to assign terms to a first plurality of nodes of the data tree structure, where each of the first plurality of nodes is assigned a weight. In assigning the weights of the first plurality of nodes, a first generation of combinations of possible weights assignable as the weights of the first plurality of nodes is obtained, and a second generation of combinations of possible weights assignable as the weights of the first plurality of nodes is obtained by performing the genetic algorithms in the first generation of combinations of possible weights. The device determines whether the document is in a document class based at least the weights of the first plurality of nodes.

Abstract: A method is disclosed that includes computing, using at least one uniformly fine-grain data parallel computing unit, a mean-square error regression within a regression clustering algorithm. The mean-square error regression is represented in the form of at least one summation of a vector-vector multiplication. A computer program product and a computer system are also disclosed.

Abstract: Methods, and apparatus to cause performance of such methods, for managing memory. The methods include requesting a particular unit of data from a first level of memory. If the particular unit of data is not available from the first level of memory, the methods further include determining whether a free unit of data exists in the first level of memory, evicting a unit of data from the first level of memory if a free unit of data does not exist in the first level of memory, and requesting the particular unit of data from a second level of memory. If the particular unit of data is not available from the second level of memory, the methods further include reading the particular unit of data from a third level of memory. The methods still further include writing the particular unit of data to the first level of memory.

Abstract: A method is provided for optimizing computer processes executing on a graphics processing unit (GPU) and a central processing unit (CPU). Process data is subdivided into sequentially processed data and parallel processed data. The parallel processed data is subdivided into a plurality of data blocks assigned to a plurality of processing cores of the GPU. The data blocks on the GPU are processed with other data blocks in parallel on the plurality of processing cores. Sequentially processed data is processed on the CPU. Result data processed on the CPU is returned.

Abstract: A method for combining data values through associative operations. The method includes, with a processor, arranging any number of data values into a plurality of columns according to natural parallelism of the associative operations and reading each column to a register of an individual processor. The processors are directed to combine the data values in the columns in parallel using a first associative operation. The results of the first associative operation for each column are stored in a register of each processor.

Abstract: A method is disclosed that includes computing, using at least one uniformly fine-grain data parallel computing unit, a mean-square error regression within a regression clustering algorithm. The mean-square error regression is represented in the form of at least one summation of a vector-vector multiplication. A computer program product and a computer system are also disclosed.

Abstract: Methods, database management systems (“DBMS”) and computer-readable media are provided for processing unbounded stream data using a traditional DBMS. Execution of a query that includes a data stream as a data source may be initiated. Tuples may be processed in accordance with the query as the tuples are received through the data stream until an indication is received that execution of the query should cease.

Abstract: A device for classifying a document comprises a module to generate a data tree structure and configured to assign terms to a first plurality of nodes of the data tree structure, where each of the first plurality of nodes is assigned a weight. In assigning the weights of the first plurality of nodes, a first generation of combinations of possible weights assignable as the weights of the first plurality of nodes is obtained, and a second generation of combinations of possible weights assignable as the weights of the first plurality of nodes is obtained by performing the genetic algorithms in the first generation of combinations of possible weights. The device determines whether the document is in a document class based at least the weights of the first plurality of nodes.

Abstract: A color constancy method and system include dividing an image into a plurality of sub-images and applying a plurality of color constancy algorithms to each of the sub-images. The outputs of each of the color constancy algorithms are analyzed for each of the sub-images to determine which of the color constancy algorithms give inconsistent results across the sub-images. The influence of the outputs of the algorithms providing inconsistent results is adjusted to decrease their influence (e.g. effect or weight) with respect to the outputs of algorithms providing consistent results. The outputs from the plurality of color constancy algorithms are combined based upon the adjustment of the outputs.