Abstract: A three-dimensional (3D) memory device includes a stack structure including a first stack structure and a second stack structure over the first stack structure, and a barrier layer disposed in a first channel hole in the first stack structure.

Abstract: Embodiments of the present disclosure provide a method and an apparatus for video processing, and a storage medium and an electronic apparatus. The method comprises: determining a neural network loop filtering enabled flag of a reconstructed video unit; setting, based on the neural network loop filtering enabled flag, an adaptive loop filtering enabled flag for the reconstructed video unit; and signaling at least one type of the following information: filter information related to neural network loop filter of the reconstructed video unit, and filter information related to adaptive loop filter of the reconstructed video unit.

Abstract: A loop filtering method, a video encoding method, a video decoding method, an electronic device, and a medium are provided. The loop filtering method includes: acquiring a first reconstructed video unit processed by a loop filter; performing pixel classification on the first reconstructed video unit; determining a scale factor corresponding to each class; and performing scale adjustment on the first reconstructed video unit according to the scale factor, so as to obtain a second reconstructed video unit.

Abstract: One embodiment of the invention provides a method for federated learning (FL) comprising training a machine learning (ML) model collaboratively by initiating a round of FL across data parties. Each data party is allocated tokens to utilize during the training. The method further comprises maintaining, for each data party, a corresponding data usage profile indicative of an amount of data the data party consumed during the training and a corresponding participation profile indicative of an amount of data the data party provided during the training. The method further comprises selectively allocating new tokens to the data parties based on each participation profile maintained, selectively allocating additional new tokens to the data parties based on each data usage profile maintained, and reimbursing one or more tokens utilized during the training to the data parties based on one or more measurements of accuracy of the ML model.

Abstract: Training a differential privacy-aware (DP-aware) machine learning model includes transmitting epsilon hyperparameters to federated learning (FL) nodes. A differential privacy-aware (DP-aware) machine learning model is generated based on noise-infused surrogate histograms received from the FL nodes, each noise-infused surrogate histogram based on an epsilon hyperparameter and representing a node-specific dataset. The DP-aware machine learning model is transmitted to the FL nodes. A DP-aware aggregate histogram is generated by merging DP-aware gradients and DP-aware Hessians determined by the FL nodes based on each FL node generating predictions by applying the DP-aware machine learning model to a node-specific dataset therein. A decision tree of the DP-aware machine learning model is expanded by dividing data in one or more decision tree nodes.

Abstract: Systems, devices, computer program products and/or computer-implemented methods of use provided herein relate to federated training and inferencing. A system can comprise a memory that stores computer executable components, and a processor that executes the computer executable components stored in the memory, wherein the computer executable components can comprise a modeling component that trains an inferential model using data from a plurality of parties and comprising horizontally partitioned data and vertically partitioned data, wherein the modeling component employs a random decision tree comprising the data to train the inferential model, and an inference component that responds to a query, employing the inferential model, by generating an inference, wherein first party private data, of the data, originating from a first passive party of the plurality of parties, is not directly shared with other passive parties of the plurality of parties to generate the inference.

Abstract: An IVI system for seamless transfer from discrete graphics to integrated graphics. For example, one embodiment comprises: an SoC comprising an iGPU to process a first rendering workload for an in-vehicle infotainment (IVI) system; and a dGPU coupled to the SoC, the dGPU to process a second rendering workload for a central control display of the IVI system; an HDMI and/or DP switch to couple a first display pipe of the dGPU to the central control display; and logic to detect a failure associated with the dGPU for the central control display and to respond by performing one or both of: triggering the HDMI/DP switch to hot-plug out the first display pipe of the dGPU and hot-plug in a second display pipe corresponding to the iGPU; based on the failure severity/type, either live-migrating the second rendering workload or creating a new instance of the critical workload on the iGPU.

Abstract: An optical lens assembly with an electrically controlled, tunable optical lens with an attached aperture stop provides autofocus or similar functionalities in a camera with a wide field of view (FOV) (that is larger than 100 degrees in diagonal direction). The tunable optical lens is positioned between a first optical lens and a second optical lens in the optical lens assembly with the assembly including any number of negative or positive optical power lenses and/or other optical elements such as polarizers, quarter wave plates, optical filters, and similar ones. An optical profile of the tunable optical lens is modified through a voltage-controlled thin film piezo actuator based on a determined focus distance.

Abstract: Architectures and techniques are provided for improved vibration compensation for a boom lift-robot-mounted (BLMR) system, which can result in improved BLMR operating precision and safety at a reduced cost. Boom lifts typically contain extendable large-scale, variable-stiffness structures, subject to complex nonlinear static deformation and dynamic motion/gust-induced vibrations. Said vibrations can be compensated for at an end effector of a robot via a Jacobian-based vibration compensation via IMU (JVCI) procedure. The output of the JVCI procedure can rely on IMU feedback and can be combined with other vibration-mitigation techniques, including various feedforward techniques.

Abstract: Provided in the present application are a single domain antibody targeting the ROR1 protein as well as a chimeric antigen receptor (CAR) and CAR-T cell comprising the single domain antibody. Also provided are uses of the single domain antibody and CAR-T cell in the treatment of tumors expressing the ROR1 protein.

Abstract: A device may include a processor configured to determine a first position of a physical camera in a physical environment, transmit the first position to a second device, obtain metadata of the physical camera, and cause a modification of a virtual camera of the second device based on the metadata. The processor is also configured to obtain, from the second device, at least one virtual image frame captured by the virtual camera placed at a second position in a computer-generated environment, and to obtain, from the physical camera, at least one image frame captured from the physical environment. The processor is also configured to generate a composite image frame including at least a portion of at least one virtual image frame composited with at least a portion of the at least one image frame, and display the composite image frame.

Abstract: Disclosed embodiments may provide techniques for generating a differentiable mesh representation of an image object. A computer-implemented method can include accessing a set of points, in which each point is assigned with a weighted value. The method can also include receiving ground-truth input to generate a differentiable mesh representation associated a target object. The method can also include constructing an initial mesh representation of the target object by generating a plurality of simplices based on the weighted values of the points. The method can also include optimizing the initial mesh representation by iteratively adjusting the existence probability values based on a comparison between the initial mesh representation with the ground-truth input. The method can also include extracting the differentiable mesh representation of the target object by selecting a set of simplices of the plurality of simplices associated with the initial mesh representation.

Abstract: A method for training an image processing model includes: acquiring a to-be-corrected eye image set matching a usage environment of the image processing model; performing style transfer on to-be-corrected eye images in the to-be-corrected eye image set through a style transfer network in the image processing model, to obtain a target eye image; acquiring a training sample matching the usage environment of the image processing model based on the to-be-corrected eye images and the target eye image, the training sample including object eye images matching different gaze positions; and training a correction network in the image processing model through the training sample matching the usage environment of the image processing model, to obtain a model update parameter matching the correction network, and generating a trained image processing model based on the model update parameter.

Abstract: An image noise reduction method including: obtaining a first image to be processed and a second image as a reference image; obtaining a block of the first image by dividing the first image; determining a motion intensity of the block based on the second image; obtaining filter coefficient description information matching the motion intensity based on: a first pixel difference of a first pixel point value of the first image and a second pixel point value of the second image, and a filter coefficient; obtaining a target pixel difference between a third pixel point value of the block and a fourth pixel point value of the second image; determining a target filter coefficient corresponding to the target pixel difference based on the filter coefficient description information; and determining a target noise reduction image of the first image based on the target filter coefficient.

Abstract: Various implementations disclosed herein include devices, systems, and methods that presents playback of application content within a three-dimensional (3D) environment. An exemplary process presents a first set of views that includes application content provided by the application within a 3D environment. The first set of views are provided from a first set of viewpoints during execution of the application. The process records of the execution of the application based on recording program state information and changes to the application content that are determined based on user interactions, and presents a second set of views including a playback of the application content within the 3D environment based on the recording. The second set of views are provided from a second set of viewpoints that are different than the first set of viewpoints.

Abstract: A video frame processing method, performed by a computer device, includes: receiving video frames carrying noise, wherein the video frames include an object; extracting poses of a part of the object; determining deflection degree values of the part based on the poses; generating a control signal based on the deflection degree values; obtaining updated deflection values based on backlash filtering of the control signal; and performing image processing on the video frames based on the updated deflection values.

Abstract: A terminal connector and an electric drive assembly are disclosed. A plurality of conductive members of the terminal block and the filter part are disposed in the housing, and a first connection end and the second connection end are respectively disposed at two ends of the extension body to conduct current. Thus, a plurality of first segments are disposed on the extension body, and the plurality of first segments are arranged at intervals to form an intermediate component disposed inside the magnetic ring. This allows the conductive member to pass through the magnetic ring multiple times via the intermediate component, thereby increasing the coupling strength between the terminal block and the magnetic ring and improving the filtering effect of the magnetic ring on common-mode noise. On the other hand, it avoids the need to set capacitors on the terminal connector. Thereby, the structure of the terminal connector is simplified, and its volume is made more compact.

Abstract: Provided herein are rapid, high quality film sintering processes that include high-throughput continuous sintering of lithium-lanthanum zirconium oxide (lithium-stuffed garnet). The instant disclosure sets forth equipment and processes for making high quality, rapidly-processed ceramic electrolyte films. These processes include high-throughput continuous sintering of lithium-lanthanum zirconium oxide for use as electrolyte films. In certain processes, the film is not in contact with any surface as it sinters (i.e., during the sintering phase).

Abstract: Semiconductor devices, manufacturing methods of the semiconductor devices, and memory systems are provided. In one aspect, a semiconductor device includes a first semiconductor structure that includes a semiconductor body, a capacitor structure, and a connection structure. The semiconductor body extends along a first direction. The capacitor structure is located on a side of the semiconductor body in the first direction and includes a first electrode layer. A first end of the connection structure in the first direction is in contact with the semiconductor body, and at least a part of a second end of the connection structure is located in the first electrode layer.