Abstract: Embodiments are generally directed to compression in machine learning and deep learning processing. An embodiment of an apparatus for compression of untyped data includes a graphical processing unit (GPU) including a data compression pipeline, the data compression pipeline including a data port coupled with one or more shader cores, wherein the data port is to allow transfer of untyped data without format conversion, and a 3D compression/decompression unit to provide for compression of untyped data to be stored to a memory subsystem and decompression of untyped data from the memory subsystem.

Abstract: Embodiments are generally directed to compression in machine learning and deep learning processing. An embodiment of an apparatus for compression of untyped data includes a graphical processing unit (GPU) including a data compression pipeline, the data compression pipeline including a data port coupled with one or more shader cores, wherein the data port is to allow transfer of untyped data without format conversion, and a 3D compression/decompression unit to provide for compression of untyped data to be stored to a memory subsystem and decompression of untyped data from the memory subsystem.

Abstract: Embodiments are generally directed to compression in machine learning and deep learning processing. An embodiment of an apparatus for compression of untyped data includes a graphical processing unit (GPU) including a data compression pipeline, the data compression pipeline including a data port coupled with one or more shader cores, wherein the data port is to allow transfer of untyped data without format conversion, and a 3D compression/decompression unit to provide for compression of untyped data to be stored to a memory subsystem and decompression of untyped data from the memory subsystem.

Abstract: The assessment and ranking of machine learning models is sped up. In one embodiment, a plurality of model definitions are automatically tested and ranked according to their expected generalization ability. Other embodiments include assessing a single change to the model to determine if the change increases or decreases the model's generalization ability. This technique can also be applied to assess input data transformations used to develop the model.

Abstract: An embodiment of a semiconductor package apparatus may include technology to process one or more vectors with a sum of squares operation with a layer of a multi-layer neural network, and determine a fixed-point approximation for the sum of squares operation. Other embodiments are disclosed and claimed.

Abstract: Embodiments are generally directed to compression in machine learning and deep learning processing. An embodiment of an apparatus for compression of untyped data includes a graphical processing unit (GPU) including a data compression pipeline, the data compression pipeline including a data port coupled with one or more shader cores, wherein the data port is to allow transfer of untyped data without format conversion, and a 3D compression/decompression unit to provide for compression of untyped data to be stored to a memory subsystem and decompression of untyped data from the memory subsystem.

Abstract: Embodiments are generally directed to compression in machine learning and deep learning processing. An embodiment of an apparatus for compression of untyped data includes a graphical processing unit (GPU) including a data compression pipeline, the data compression pipeline including a data port coupled with one or more shader cores, wherein the data port is to allow transfer of untyped data without format conversion, and a 3D compression/decompression unit to provide for compression of untyped data to be stored to a memory subsystem and decompression of untyped data from the memory subsystem.

Abstract: According to one aspect of the disclosure, a computing system including at least one networking device with at least one processor and memory. The at least one processor executes instructions stored in the memory to collect data from one or more Internet of Things (IoT) devices via a first communication protocol and provide access to the collected data by one or more user devices via a second and different communication protocol. Further, the one or more IoT devices are communicatively coupled to the at least one networking device to sense the data and transmit the data to the at least one networking device, and the one or more user devices are communicatively coupled to the at least one networking device.

Abstract: Embodiments are generally directed to compression in machine learning and deep learning processing. An embodiment of an apparatus for compression of untyped data includes a graphical processing unit (GPU) including a data compression pipeline, the data compression pipeline including a data port coupled with one or more shader cores, wherein the data port is to allow transfer of untyped data without format conversion, and a 3D compression/decompression unit to provide for compression of untyped data to be stored to a memory subsystem and decompression of untyped data from the memory subsystem.

Abstract: Embodiments are generally directed to compression in machine learning and deep learning processing. An embodiment of an apparatus for compression of untyped data includes a graphical processing unit (GPU) including a data compression pipeline, the data compression pipeline including a data port coupled with one or more shader cores, wherein the data port is to allow transfer of untyped data without format conversion, and a 3D compression/decompression unit to provide for compression of untyped data to be stored to a memory subsystem and decompression of untyped data from the memory subsystem.

Abstract: An apparatus to facilitate compute optimization is disclosed. The apparatus includes a at least one processor to perform operations to implement a neural network and compute logic to accelerate neural network computations.

Abstract: An embodiment of a semiconductor package apparatus may include technology to process one or more vectors with a sum of squares operation with a layer of a multi-layer neural network, and determine a fixed-point approximation for the sum of squares operation. Other embodiments are disclosed and claimed.

Abstract: In one embodiment, a first request is received for a probe link from a client that has downloaded a page. The request is received from an application. A test may be performed to determine a server that could optimally service a content request from the client. After receiving the request, the content router sends a re-direct to the client where the re-direct causes the client to follow the re-direct to the server. The server then sends a re-direct back to the client, which the client follows by sending a second request to the content router. The second request is associated with the client and the round-trip time is calculated for the server. It is then determined if the location for the server should be designated as the content deliverer to the client based on the calculated round-trip time.

Abstract: A technique for selecting a media streaming engine, such as a stream server, to source a media stream to a client involves receiving feedback information from a client that indicates a quality of service experienced by the client and then selecting a media streaming engine, from a set of media streaming engines, in response to the feedback information. In an embodiment, the feedback information is provided in Real-Time Control Protocol (RTCP) messages. For example, the RTCP messages are Receiver Report (RR) messages that contain packet loss and/or inter-arrival jitter information. Feedback information such as packet loss and/or inter-arrival jitter information represents the actual quality of service experienced by a client for a media stream. Because a media streaming engine is selected based on the actual quality of service experienced by a client, the selection process can be configured to optimize the quality of service of the media streaming.

Abstract: A streaming file is constructed with a file header section that includes a file header object, a media data file descriptor, and an index descriptor. A hinting index section includes a first level hinting index with a linear organization corresponding to timing tick key values. A second level hinting index has a non-sequential organization corresponding to such timing tick key values. A special mark in the second level hinting index indicates that the first level hinting index must be consulted for a next timing tick key value. Such mark is positioned in the last of a sequential run of timing tick key values associated with its entries. A data section that can be put in a separate file, and it accepts media data blocks associated in sequential runs of timing tick key values as its entries. Thus hinting is provided for a non-sequential media data file.

Abstract: A system and process are provided to reduce the latency associated with delivery of data for trick functions (e.g., fast-forward or rewind) in data distribution systems. In response to a request for a trick function associated with program data being streamed to a user terminal, first trick data frames for the trick function are retrieved from a first computing apparatus (e.g., a storage server) and in the meantime second trick data frames for the trick function are generated at a second computing apparatus (e.g., streaming server). Until the first trick data frames arrive from the first computing apparatus, the second computing device streams the second trick data frames to the user terminal. The second computing apparatus streams the first trick data frame to the user terminal as they arrive.

Abstract: Systems and methods are disclosed for creating image maps. Some embodiments include a method comprising the steps of: capturing a first image illuminated by natural light and capturing a second image illuminated by infrared light. The second image may be captured at the same time as the first image. The R, G, and B values for each pixel in the first image may be determined. The intensity for each pixel in the first image may be calculated. An IR intensity for each pixel in the second image may be calculated. A depth value may then be estimated for each pixel using the ratio of the IR intensity and the intensity of corresponding pixels in the first and second images.

Abstract: In particular embodiments, method and system for detecting a failure of a primary ad-splicer, conveying a failure information for the failed primary ad-splicer to a redundant ad-splicer, dynamically forwarding one or more pre-spliced packets intended for the failed primary ad-splicer to the redundant ad-splicer, receiving one or more post-spliced packets from the redundant ad-splicer, and transmitting the post-spliced packets towards one or more target receivers are provided.

Abstract: Techniques are provided herein for failover streaming mechanisms. At a first device (e.g., a content router device) that is configured to interface with a plurality of streaming servers for real-time protocol packet streams, communications are configured with a client device and a first of the plurality of streaming servers associated with a streaming session from the first streaming server to the client device so that the first device receives client session control and session feedback messages associated with the streaming session and so that a packet stream associated with the streaming session transmitted by the first streaming server to the client device does not pass through the first device. The first device stores session state information comprising an address of the client device, streaming session identification information and data representing a current state of the streaming session at the client device derived from the client session control and session feedback messages.

Abstract: Systems and methods are disclosed for creating image maps. Some embodiments include a method comprising the steps of: capturing a first image illuminated by natural light and capturing a second image illuminated by infrared light. The second image may be captured at the same time as the first image. The R, G, and B values for each pixel in the first image may be determined. The intensity for each pixel in the first image may be calculated. An IR intensity for each pixel in the second image may be calculated. A depth value may then be estimated for each pixel using the ratio of the IR intensity and the intensity of corresponding pixels in the first and second images.