Abstract: The present disclosure describes techniques for implementing instant corruption detection and recovery. A plurality of streams may be created in a storage device. Each of the plurality of streams may contain a sequence of metadata nodes of a same type. Each of the plurality of streams may maintain an initial state, a sequence of delta modifications to the initial state, and an actual state for each of the sequence of metadata nodes. A checking and recovery function associated with a particular stream among the plurality of streams may be determined. The checking and recovery function may comprise checking logic configured to detect corruptions by checking modification operations associated with metadata nodes in the particular stream. The checking and recovery function may further comprise recovery logic configured to perform recoveries from the corruptions. The checking and recovery function associated with the particular stream may be implemented in the storage device.

Abstract: A system includes a first hardware architecture and computational storage. The computational storage includes a second hardware architecture, and memory storing instructions that, when executed by the second hardware architecture, causes the system to perform a first set of operations. The first set of operations include receiving instructions from the first hardware architecture, performing one or more computations on data stored in the second hardware architecture, based on the received instructions, and transmitting a result to the first hardware architecture, based on the one or more performed computations.

Abstract: Example aspects include techniques for implementing real-time and low-latency synthesis of audio. These techniques may include generating a frame by sampling audio input in increments equal to a buffer size of until a threshold corresponding to a frame size used to train a machine learning (ML) model is reached, detecting feature information within the frame, determining, by the ML model, control information for audio reproduction based on the feature information. In addition, the techniques may include generating filtered noise information by inverting the noise magnitude control information using an overlap and add technique, generating, based on the control information, additive harmonic information by combining a plurality of scaled wavetables, and rendering audio output based on the filtered noise information and the additive harmonic information.

Abstract: A method for adaptive mapping for data compression includes determining an input/output (I/O) request pattern, dynamically switching between a segment mapping mode and a flat hash table mapping mode based on the determined I/O request pattern, updating a shared mapping table for the segment mapping mode and the flat hash table mapping mode, and adjusting an entry of the mapping table based on the determined I/O request pattern and a status of the entry.

Abstract: Provided are a video processing method and apparatus, an electronic device, a storage medium and a program product. The method includes steps described below. A to-be-processed video is acquired, and a first face key point of a first target object in the to-be-processed video is recognized, where the first face key point corresponds to a mask key point in a three-dimensional human face mask effect; and the three-dimensional human face mask effect is added to video frames of the to-be-processed video to obtain a target video, where the mask key point in the three-dimensional human face mask effect moves with the corresponding first face key point.

Abstract: The present disclosure describes techniques for controllable music generation. The techniques comprise extracting latent vectors from unlabelled data, the unlabelled data comprising a plurality of music note sequences, the plurality of music note sequences indicating a plurality of pieces of music; clustering the latent vectors into a plurality of classes corresponding to a plurality of music styles; generating a plurality of labelled latent vectors corresponding to the plurality of music styles, each of the plurality labelled latent vectors comprising information indicating features of a corresponding music style; and generating a first music note sequence indicating a first piece of music in a particular music style among the plurality of music styles based at least in part on a particular labelled latent vector among the plurality of labelled latent vectors, the particular labelled latent vector corresponding to the particular music style.

Abstract: A method for accessing blocks of a solid state drive is described. A starting position is received, where the starting position identifies a first block of a contiguous block region within a namespace of the solid state drive. The contiguous block region is to be accessed according to a single input/output operation and the namespace comprises two dimensions of logical address space with respective indices for indexing blocks within a corresponding dimension of the logical address space. A first dimensional identifier that identifies a size of the contiguous block region in a first dimension of the namespace is received. A second dimensional identifier that identifies a size of the contiguous block region in a second dimension of the namespace is received. Blocks of the contiguous block region are accessed in response to the single input/output operation according to the starting position, the first dimensional identifier, and the second dimensional identifier.

Abstract: A method of generating a style image is described. The method includes receiving an input image of a subject. The method further includes encoding the input image using a first encoder of a generative adversarial network (GAN) to obtain a first latent code. The method further includes decoding the first latent code using a first decoder of the GAN to obtain a normalized style image of the subject, wherein the GAN is trained using a loss function according to semantic regions of the input image and the normalized style image.

Abstract: Implementations of the present disclosure relate to methods, devices, and computer program products of extracting a feature for multimedia data that comprises a plurality of medium types. In a method, a first feature is determined for a first medium type in the plurality of medium types by masking a portion in a first medium object with the first medium type. A second feature is determined for a second medium type other than the first medium type in the plurality of medium types. The feature is generated for the multimedia data based on the first and second features. With these implementations, multiple medium types are considered in the feature extraction, and thus the feature may fully reflect various aspects of the multimedia data in an accurate way.

Abstract: Embodiments of the present disclosure relate to a method, device and computer readable storage medium of scene text detection. In the method, a first visual representation of a first image is generated with an image encoding process. A first textual representation of a first text unit in the first image is generated with a text encoding process based on a first plurality of symbols obtained by masking a first symbol of a plurality of symbols in the first text unit. A first prediction of the masked first symbol is determined with a decoding process based on the first visual and textual representations. At least the image encoding process is updating according to at least a first training objective to increase at least similarity of the first prediction and the masked first symbol.

Abstract: The techniques for dual-level semantic segmentation are provided. Data may be input to a first segmentation network. The input data comprises an image and label information associated with the image. The image may be captured at nighttime and may comprise a plurality of regions. At least one region among the plurality of regions may be determined based at least in part on output of the first segmentation network. The at least one region of the image may be cropped. The cropped at least one region may be input to a second segmentation network. A final output may be produced based on the output of the first segmentation network and output of the second segmentation network.

Abstract: The present disclosure relates to a method and apparatus for knowledge graph construction, storage medium and electronic device. The method for knowledge graph construction, comprises: identifying an entity concept from a title text of a target web page and at least one entity corresponding to the entity concept from a body text of the target web page; constructing a syntax parse tree of the title text based on syntax parse rules of a language to which the title text belongs, and determining, from the syntax parse tree, a modifier for modifying the entity concept; and generating a knowledge graph based on the entity concept, the modifier, and the at least one entity. Through the solution of the present disclosure, knowledge graphs with high accuracy and high recall rates are constructed without structured processing on target web pages.

Abstract: Systems and methods for multi-task joint training of a neural network including an encoder module and a multi-headed attention mechanism are provided. In one aspect, the system includes a processor configured to receive input data including a first set of labels and a second set of labels. Using the encoder module, features are extracted from the input data. Using a multi-headed attention mechanism, training loss metrics are computed. A first training loss metric is computed using the extracted features and the first set of labels, and a second training loss metric is computed using the extracted features and the second set of labels. A first mask is applied to filter the first training loss metric, and a second mask is applied to filter the second training loss metric. A final training loss metric is computed based on the filtered first and second training loss metrics.

Abstract: The present disclosure describes techniques for voice-controlled content creation. The techniques comprise monitoring voice commands spoken by a creator. Recording of a content may be initiated in response to recognizing a first voice command spoken by the creator. Recording of the content may be stopped in response to recognizing a second voice command spoken by the creator. A timestamp associated with the second voice command may be created. A segment may be automatically deleted from the content based on the timestamp. The segment may comprise a recording of the second voice command.

Abstract: The embodiment of the disclosure relates to a touch animation display method and apparatus, device, and medium, wherein the method includes the following steps: identifying a human body key point position on a human body in an existing video frame; according to a predetermined motion mode, acquiring a display position of a target object in the existing video frame; according to the human body key point position and the display position, judging whether a preset touch condition is satisfied; if the touch condition is satisfied, adjusting a motion mode of the target object according to a target key point position. Therefore, by displaying the target object in the video frame, the touch motion during interaction with a user is simulated in real time, the interestingness of video display is improved, and the user's interaction experience is improved.

Abstract: The present disclosure provides a data processing method, a device, a computer apparatus and a storage medium, wherein the method includes: in response to a target disk receiving at least one write request within a preset time period, determining a size threshold value for classifying a data update type according to a size of write data respectively indicated by each write request; determining a data update type corresponding to each write request according to a size of each write request and the size threshold value; dividing write data of the write request to obtain a data block according to a preset data block size, and caching the data block in a cache region of the target disk corresponding to the data update type, the target disk has multiple types of cache regions configured therein, different cache regions are configured to support caching of data with different update frequencies.

Abstract: Systems and methods are provided for improved point querying of a database. The index values are separated from data and retained in cache memory to allow access without requiring a disk input/output (I/O) operation and thereby having less latency resulting from such disk I/O operations. The index values can be compressed using an algorithm such as Crit-Bit-Trie to allow storage of the index values in limited cache memory space. The index values can be selected for storage according to a least recently used approach when cache memory is insufficient to store all index values to maintain a hit rate for the cached portion and reduce the disk I/O operations.

Abstract: Methods and systems for secure computation and communication are provided. The method includes transforming identifications of a first dataset using a first transforming scheme, and transforming attributes of the first dataset using a second transforming scheme. The method also includes dispatching the transformed first dataset, receiving a second dataset, transforming identifications of the received second dataset, dispatching the identifications of the transformed received second dataset, and receiving a set of identifications. The method further includes generating a first intersection of the received set of identifications and the transformed received second dataset, generating a first share based on the first intersection, receiving a second share, and constructing a result based on the first share and the second share.

Abstract: A method for live streaming recommended content from a content request includes live streaming a first content from a first user side, activating a live reaction at the first user side, and providing the content request at a second user side when the live reaction is enabled. The method also includes determining a second content, sending the content request indicating the second content, and receiving a confirmation at the second user side indicating the content request being sent. The method further includes receiving, at the first user side, the content request sent from the second user side, approving the content request at the first user side, and after the content request is approved, live streaming from the first user side, the second content corresponding to the content request.

Abstract: Systems and methods directed to controlling the similarity between stylized portraits and an original photo are described. In examples, an input image is received and encoded using a variational autoencoder to generate a latent vector. The latent vector may be blended with latent vectors that best represent a face in the original user portrait image. The resulting blended latent vector may be provided to a generative adversarial network (GAN) generator to generate a controlled stylized image. In examples, one or more layers of the stylized GAN generator may be swapped with one or more layers of the original GAN generator. Accordingly, a user can interactively determine how much stylization vs. personalization should be included in a resulting stylized portrait.