Abstract: An input weight pattern of a machine learning model may be received. The input weight pattern may be pruned to produce an output weight pattern based on a predetermined pruning algorithm. The pruning algorithm may include partitioning the input weight pattern into a plurality of sub-patterns, each row of the input weight pattern including sub-rows of a first number of sub-patterns, and each column of the input weight pattern including sub-columns of a second number of sub-patterns; and pruning sub-columns and sub-rows from the plurality of sub-patterns to achieve predetermined column and row sparsities respectively, with a constraint that at least one sub-row in each row of the input weight pattern is not pruned. The output weight pattern may further be compressed to produce a compact weight pattern. The compact weight pattern has lower memory and computational overheads as compared to the input weight pattern for the machine learning model.

Abstract: The disclosed embodiments describe methods, systems, and apparatuses for determining user intent. A method is disclosed comprising obtaining a session text of a user; calculating, by the processor, a feature vector based on the session text; determining probabilities that the session text belongs to a plurality of intent labels, the probabilities calculated using a multi-level hierarchal intent classification model, the intent labels assigned to levels in the multi-level hierarchal intent classification model; and assigning a user intent to the session text based on the probabilities.

Abstract: A terminal communicates with base stations by receiving a first beacon frame sent by a first base station, the first beacon frame including first location information representing a location of the first base station. In response to one of a determination that a distance between the first location information and second location information is greater than a first predetermined threshold or a determination that a difference between first received signal strength information and second received signal strength information is greater than a second predetermined threshold, the terminal sends first information to a network server, the first information causing the network server to select a base station that transmits downlink data to the terminal. The second location information represents a location of a second base station corresponding to a second beacon frame received before the first beacon frame is received.

Abstract: Embodiments of the present disclosure provide methods and devices for communication by a terminal. The method can include: sending, by the terminal, a first data frame to a relay device in a first mode, the first data frame comprising a first preamble and a first message information set, and the first preamble being used to wake up the relay device; sending, by the terminal, a second data frame to the relay device in the first mode, the second data frame comprising a second preamble and a second information set, the second data frame being sent after the terminal has sent the first data frame and a first time interval has elapsed, and the length of the second preamble being less than that of the first preamble; and receiving, by the terminal, a third data frame sent by the relay device in the first mode, the third data frame being sent to the terminal after being received from the base station by the relay device.

Abstract: A method for content interaction through overlays on a mobile device includes the steps of receiving a content on a display of the mobile device, generating an overlay image on the received content, matching the overlay image only or together with received content to a set of templates for identifying a template associated with the overlay image only or together with received content, identifying a template matched to the overlay image only or together with received content, assigning the overlay template and an action, if any associated with the template, to the received content.

Abstract: The present disclosure provides method and apparatus for motion field storage in video coding. An exemplary method includes: determining whether a first uni-prediction motion vector for a first partition of a block and a second uni-prediction motion vector for a second partition of the block are from a same reference picture list; and in response to the first uni-prediction motion vector and the second uni-prediction motion vector being determined to be from the same reference picture list, storing, in a motion field of the block, one of the first uni-prediction motion vector and the second uni-prediction motion vector for a subblock located in a bi-predicted area of the block.

Abstract: Transcoding bitrate prediction techniques can include receiving a first encoded content. A transcoder bitrate can be estimated based on regression over a video quality estimator of the first encoded content and a second encoded content. The estimated transcoder bitrate can be utilized to transcoding the first encoded content into the second encoded.

Abstract: The present disclosure relates to a system and a method for data protection. In some embodiments, an exemplary method for data encoding includes: receiving a data bulk; performing an erasure coding (EC) encoding on the data bulk to generate one or more EC codewords; distributing a plurality of portions of each EC codeword of the one or more EC codewords across a plurality of solid-state drives (SSDs); performing, at each SSD of the plurality of SSDs, an error correction coding (ECC) encoding on portions of the one or more EC codewords distributed to the SSD to generate an ECC codeword; and storing, in each SSD of the plurality of SSDs, the ECC codeword.

Abstract: A database system as well as a method and an apparatus for querying a database are disclosed herein. In one embodiment, a system comprises a master database, at least one slave database, and a query processing unit. The master database has a data connection with the slave database, and the slave database has a data connection with the query processing unit. The query processing unit is configured to split a query request into query operation instructions having non-overlapping query ranges, send the split query operation instructions to the same or different slave databases to perform the query operations, and return the query results. The disclosed embodiments improve query efficiency.

Abstract: The present disclosure relates to a method for compensating an image. The method comprises estimating transform coefficients of a frequency component for a first image based on the first image, performing a dot multiplication operation between the estimated transform coefficients and a basis function associated with the frequency component to generate a compensation image, and combining the first image and the compensation image to generate a combined image.

Abstract: The present disclosure provides systems and methods for wrap-around motion compensation. One exemplary method comprises: receiving a wrap-around motion compensation flag; determining whether a wrap-around motion compensation is enabled based on the wrap-around motion compensation flag; in response to a determination that the wrap-around motion compensation is enabled, receiving data indicating a difference between a width of the picture and an offset used for determining a horizontal wrap-around position; and performing a motion compensation according to the wrap-around motion compensation flag and the difference.

Abstract: Embodiments of the disclosure provide methods and apparatuses for video searches and methods and apparatuses for index construction. In one embodiment, the method comprises: upon receiving a search request input by a user to search for a target video, processing, based on a pre-configured algorithm, multimodal search data for the target video included in the search request; providing a processing result of the multimodal search data with regard to a corresponding pre-constructed index to search to obtain the target video.

Abstract: A model processing method and apparatus, a device, and a computer-readable storage medium. The method comprises: obtaining a first computational graph corresponding to a model to be trained and a parallelization strategy of said model, the parallelization strategy of said model comprising at least one of pipeline parallelism, model parallelism, data parallelism, and operator splitting (S601); adding parallelization information in the first computational graph according to the parallelization strategy of said model to obtain a second computational graph (S602); determining a distributed computational graph according to the second computational graph and computing resources (S603); and training said model according to the distributed computational graph (S604).

Abstract: One embodiment described herein provides a distributed computing system. The distributed computing system can include a compute cluster comprising one or more compute nodes and a storage cluster comprising a plurality of storage nodes. A respective compute node can be configured to: receive a request for a computation task; obtain path information associated with data required by the computation task; identify at least one storage node based on the obtained path information; send at least one computation instruction associated with the computation task to the identified storage node; and receive computation results from the identified storage node subsequently to the identified storage node performing the computation task.

Abstract: The present disclosure relates to methods and apparatuses for execution of a neural network. An exemplary method can be implemented by a processing unit. The processing unit can include a command parser configured to dispatch commands and computing tasks and at least one core communicatively coupled with the command parser and configured to process the dispatched computing task. Each core can include a convolution unit, a pooling unit, at least one operation unit and a sequencer communicatively coupled with the convolution unit, the pooling unit, and the at least one operation unit and configured to distribute instructions of the dispatched computing task to the convolution unit, the pooling unit, and the at least one operation unit for execution.

Abstract: Methods for performing palette coding of video data may include: determining whether a luma component of a coding unit (CU) and chroma components of the CU are coded jointly or separately in a palette mode; and in response to the luma component and the chroma components being coded separately in the palette mode, determining a maximum separate palette-table size for the CU; determining a maximum separate palette-predictor size for the CU; and predicting the CU based on the maximum separate palette-table size and the maximum separate palette-predictor size.

Abstract: A method for managing a region tag comprises: acquiring region identifier information of a geographic region; acquiring an attribute evaluation tag of the geographic region by a user in the geographic region and a time identifier of the attribute evaluation tag; integrating the region identifier information, the attribute evaluation tag, and the time identifier thereof into a region tag of the geographic region; clustering region tags to generate a region time period tag of the geographic region for a time period based on that each region tag comprises the region identifier information of the geographic region and a time identifier within the time period, wherein the region time period tag of the geographic region for the time period corresponds to a function attribute of the geographic region for the time period; and outputting region time period tags of the geographic region for time periods, respectively.

Abstract: The present invention discloses a debugging unit and a processor. The debugging unit includes; a register, adapted to sample input data under control of a clock signal; and a dock control unit, adapted to generate a control signal based on a clock enable signal to control the clock signal, so that the register is controlled to sample the input data in a validity period of the clock signal when the control signal is valid. The present invention also discloses a corresponding system-on-chip and an intelligent device including the system-on-chip.

Abstract: Methods and apparatuses video processing include: in response to receiving a video sequence, encoding first flag data in a parameter set associated with the video sequence, wherein the first flag data represents whether gradual decoding refresh (GDR) is enabled or disabled for the video sequence; when the first flag data represents that the GDR is disabled for the video sequence, encoding a picture header associated with a picture in the video sequence to indicate that the picture is a non-GDR picture; and encoding the non-GDR picture.

Abstract: Improved convolutional layers for neural networks can obtain an input feature map comprising groups of channels. Each group of channels can include one or more channels having a predetermined size. The predetermined sizes can differ between the groups. The convolutional layer can generate, for each one of the groups of channels, an output channel. Generation of the output channel can include resizing the channels in the remaining groups of channels to match the predetermined size of the each one of the groups of channels. Generation can further include combining the channels in the each one of the groups with the resized channels and applying the combined channels to a convolutional sub-layer to generate the output channel.