Alcatel Lucent

Nokia Technologies Oy

Alcatel-lucent USA Inc.

Provenance Asset Group LLC

Fangzhe Chang

Thomas L Wood

Yansong Ren

Dong Hui Liu

Thomas Y Woo

Ramesh Viswanathan

John R Zhang

Jennifer Ren

Yangsong Ren

Systems and methods for training neural networks. One embodiment is a system that includes a memory configured to store samples of training data for a Deep Neural Network (DNN), and a distributor. The distributor identifies a plurality of work servers provisioned for training the DNN by processing the samples via a model of the DNN, receives information indicating Graphics Processing Unit (GPU) processing powers at the work servers, determines differences in the GPU processing powers between the work servers based on the information, and allocates the samples among the work servers based on the differences.

Coordinated heterogeneous processing of training data for deep neural networks

A method for comparing a query video and a target video includes partitioning frames of the query video and frames of the target video into blocks and calculating the mean intensity value for each block. A plurality of query time series is produced for the query video, each query time series representing temporal variation in mean intensity value for blocks from the same location in different frames of the query video. A plurality of target time series is produced for the target video, each target time series representing temporal variation in mean intensity value for blocks from the same location in different frames of the target video the query time series and the target time series are used in determining if alignment exists between the query video and the target video.

Method and apparatus for comparing videos

In a method of comparing a query video and a target video, frames of the query video and frames of the target video are partitioned into blocks. The mean intensity value is calculated for each block. Minima and maxima values are extracted from temporal changes of the mean intensity values for each block. Inclines and declines are obtained from the minima and maxima values. The inclines and declines are used in determining if alignment exists between the query video and the target video.

A method for comparing videos includes detecting local features in frames of a video and determining the positions of detected local features in a first frame compared to their respective positions in a succeeding second frame. For at least some of the detected local features, a movement indicator is generated representing the position of a local feature in the first frame relative to the position of that local feature in the second frame, the movement indicator including an orientation, to obtain a plurality of movement indicators. Movement indicators are grouped by orientation ranges. The number of grouped movement indicators in an orientation range are summed to obtain a value for that orientation range. A time series of values for at least one given orientation range is obtained by determining the positions of detected local features in a first frame compared to their respective positions in a succeeding second frame for a plurality of pairs of frames. For at least one orientation range, a first time series of values for a first video is compared with a second time series of values for a second video to determine the similarity of the first and second videos.

A system for configuring resources in an environment for use by at least one process. In one embodiment, the system includes: (1) a process sorter configured to rank the at least one process based on numbers of resources that steps in the at least one process can use, (2) an optimizer coupled to the process sorter and configured to employ an optimization heuristic to accumulate feasible allocations of resources to the steps based on the ranking of the at least one process, (3) a resource sorter coupled to the optimizer and configured to rank the resources in a non-decreasing order based on numbers of the steps in which the resources can be used, the optimizer further configured to remove one of the resources from consideration based on the ranking of the resources until infeasibility occurs and (4) an environment configuration interface configured to allow the environment to be configured in accordance with remaining ones of the resources.

System and method for optimizing batch resource allocation

The invention includes a method and apparatus for validating system properties exhibited in execution traces. In one embodiment, a method for testing a system under test (SUT) includes determining a system testing result for the SUT using at least one structured term generated by monitoring an execution trace of the SUT using at least one parameterized pattern. A test procedure is executed for the SUT. The test procedure has at least one parameterized pattern associated therewith. An execution trace generated during execution of the test procedure is parsed, where the execution trace includes unstructured information and the execution trace is parsed using the at least one parameterized pattern to identify at least one matching pattern. A system testing result for the SUT is determined using at least one structured term that is generated using the at least one matching pattern. In this manner, behavior of the system under test during the test procedure may be validated.

Growing community of inventors

Edison, NJ, United States of America

Fangzhe Chang