Abstract: A scheduling algorithm for scheduling training of deep neural network (DNN) weights on processing units identifies a next job to provisionally assign a processing unit (PU) based on a doubling heuristic. The doubling heuristic makes use of an estimated number of training sets needed to complete training of weights for a given job and/or a training speed function which indicates how fast the weights are converging. The scheduling algorithm solves a problem of efficiently assigning PUs when multiple DNN weight data structures must be trained efficiently. In some embodiments, the training of the weights uses a ring-based message passing architecture. In some embodiments, performance using a nested loop approach or nested loop fashion is provided. In inner iterations of the nested loop, PUs are scheduled and jobs are launched or re-started. In outer iterations of the nested loop, jobs are stopped, parameters are updated and the inner iteration is re-entered.

Abstract: An apparatus for video searching, includes a memory storing instructions, and a processor configured to execute the instructions to split a video into scenes, obtain, from the scenes into which the video is split, one or more textual descriptors describing each of the scenes, encode the obtained one or more textual descriptors describing each of the scenes into a video scene vector of each of the scenes, encode a user query into a query vector having a same semantic representation as that of the video scene vector of each of the scenes into which the one or more textual descriptors describing each of the scenes are encoded, and identify whether the video scene vector of at least one among the scenes corresponds to the query vector into which the user query is encoded.

Abstract: An apparatus for vision and language-assisted smartphone task automation, includes a processor configured to execute instructions to identify whether an utterance of a user refers to one among at least one existing automation script, and based on the utterance being identified to not refer to the one among the at least one existing automation script, capture one or more screenshots and one or more coordinates of a display, the one or more screenshots and the one or more coordinates corresponding to one or more interactions of the user with the display while the user is demonstrating a new automation script, perform object detection and optical character recognition on the captured one or more screenshots and the captured one or more coordinates to obtain user interface (UI) elements corresponding to the one or more interactions, and obtain the new automation script, based on the obtained UI elements.

Abstract: An apparatus for video searching, includes a memory storing instructions, and a processor configured to execute the instructions to split a video into scenes, obtain, from the scenes into which the video is split, one or more textual descriptors describing each of the scenes, encode the obtained one or more textual descriptors describing each of the scenes into a video scene vector of each of the scenes, encode a user query into a query vector having a same semantic representation as that of the video scene vector of each of the scenes into which the one or more textual descriptors describing each of the scenes are encoded, and identify whether the video scene vector of at least one among the scenes corresponds to the query vector into which the user query is encoded.

Abstract: An apparatus for vision and language-assisted smartphone task automation, includes a processor configured to execute instructions to identify whether an utterance of a user refers to one among at least one existing automation script, and based on the utterance being identified to not refer to the one among the at least one existing automation script, capture one or more screenshots and one or more coordinates of a display, the one or more screenshots and the one or more coordinates corresponding to one or more interactions of the user with the display while the user is demonstrating a new automation script, perform object detection and optical character recognition on the captured one or more screenshots and the captured one or more coordinates to obtain user interface (UI) elements corresponding to the one or more interactions, and obtain the new automation script, based on the obtained UI elements.

Abstract: A scheduling algorithm for scheduling training of deep neural network (DNN) weights on processing units identifies a next job to provisionally assign a processing unit (PU) based on a doubling heuristic. The doubling heuristic makes use of an estimated number of training sets needed to complete training of weights for a given job and/or a training speed function which indicates how fast the weights are converging. The scheduling algorithm solves a problem of efficiently assigning PUs when multiple DNN weight data structures must be trained efficiently. In some embodiments, the training of the weights uses a ring-based message passing architecture. In some embodiments, performance using a nested loop approach or nested loop fashion is provided. In inner iterations of the nested loop, PUs are scheduled and jobs are launched or re-started. In outer iterations of the nested loop, jobs are stopped, parameters are updated and the inner iteration is re-entered.

Abstract: Techniques for selecting a new compression technique or altering the currently instantiated compression technique employed over a data stream in data collection system are provided. At least one change to at least one external contextual condition is detected. One or more modifications to a defined compression technique is generated, either independently or in conjunction with one or more modifications to the event processing operators that operate on the data stream and whose output provides the values that are compressed by the defined compression technique, in response to the at least one change to the at least one external contextual condition. The defined compression technique of the at least one client device is altered in accordance with the one or more modifications to form a modified compression technique through which the unmodified or processed data stream will be compressed before being sent to the server.

Abstract: A cloud infrastructure that communicates with computing devices is provided. The computing devices install filters on other computing devices that they wish to receive items from including pictures, messages, and documents. The filters include criteria that are evaluated on the computing devices, rather than at a server, to determine if an item may be sent to another computing device. The computing devices may then send items that match the criteria to the cloud infrastructure, and the items may be stored and queued for delivery to other computing devices. The items may be encrypted before being provided to the cloud infrastructure, and decrypted when received by the computing devices.

Abstract: A cloud infrastructure that communicates with computing devices is provided. The computing devices install filters on other computing devices that they wish to receive items from including pictures, messages, and documents. The filters include criteria that are evaluated on the computing devices, rather than at a server, to determine if an item may be sent to another computing device. The computing devices may then send items that match the criteria to the cloud infrastructure, and the items may be stored and queued for delivery to other computing devices. The items may be encrypted before being provided to the cloud infrastructure, and decrypted when received by the computing devices.

Abstract: Tracks associated with a first user are identified by a computing device. Each track may include location identifiers. The identified tracks are clustered to generate a composite track for the first user by the computing device. At least one track that is similar to the composite track is identified by the computing device. The at least one track may be associated with a user other than the first user. Information related to the identified at least one track that is similar to the composite track is provided by the computing device through a network.

Abstract: Techniques for selecting a new compression technique or altering the currently instantiated compression technique employed over a data stream in data collection system are provided. At least one change to at least one external contextual condition is detected. One or more modifications to a defined compression technique is generated, either independently or in conjunction with one or more modifications to the event processing operators that operate on the data stream and whose output provides the values that are compressed by the defined compression technique, in response to the at least one change to the at least one external contextual condition. The defined compression technique of the at least one client device is altered in accordance with the one or more modifications to form a modified compression technique through which the unmodified or processed data stream will be compressed before being sent to the server.