Abstract: Certain aspects of the present disclosure relate to wireless communications, and more particularly, to techniques for xx. A method that may be performed by a user equipment (UE) includes receiving, from a network entity, a first transport block (TB) comprising one or more code blocks (CBs), transmitting, to the network entity, an indication of a preferred code block group (CBG) allocation scheme selected for re-transmission of at least one CB of the one or more CBs based, at least in part, on a decoding status for each of the one or more CBs, and receiving, from the network entity, the at least one CB re-transmitted in accordance with the preferred CBG allocation scheme.

Abstract: Methods, apparatus, systems and articles of manufacture for loss-error-aware quantization of a low-bit neural network are disclosed. An example apparatus includes a network weight partitioner to partition unquantized network weights of a first network model into a first group to be quantized and a second group to be retrained. The example apparatus includes a loss calculator to process network weights to calculate a first loss. The example apparatus includes a weight quantizer to quantize the first group of network weights to generate low-bit second network weights. In the example apparatus, the loss calculator is to determine a difference between the first loss and a second loss. The example apparatus includes a weight updater to update the second group of network weights based on the difference. The example apparatus includes a network model deployer to deploy a low-bit network model including the low-bit second network weights.

Abstract: A power management circuit for a computer system is disclosed. The power management circuit includes a power storage circuit, a first power converter circuit, and a second power converter circuit. The first power converter circuit sources a current to a regulated power supply node via a first inductor under average load conditions using a voltage level of an input power supply node. During high-load transient conditions, the second power converter circuit provides additional current to the regulated power supply node via a second inductor using a voltage generated by the power storage circuit.

Abstract: An example apparatus for mining multi-scale hard examples includes a convolutional neural network to receive a mini-batch of sample candidates and generate basic feature maps. The apparatus also includes a feature extractor and combiner to generate concatenated feature maps based on the basic feature maps and extract the concatenated feature maps for each of a plurality of received candidate boxes. The apparatus further includes a sample scorer and miner to score the candidate samples with multi-task loss scores and select candidate samples with multi-task loss scores exceeding a threshold score.

Abstract: A remote sensing assessment method for soil erosion control degree based on maximum erosion potential includes: data preparation and process, and determining a current soil erosion modulus, a maximum possible soil erosion modulus, and a soil erosion control degree. A soil erosion modulus is determined as the maximum possible soil erosion modulus of a research area under a condition that a most serious water and soil loss possibly occurs; the maximum possible and the current soil erosion moduli are calculated by using a USLE model; and a ratio of a difference between the maximum possible and the current soil erosion moduli to the maximum possible soil erosion modulus is determined as the soil erosion control degree. Therefore, the soil erosion control degree is representative and popularized to measure soil management potential, which improves evaluation efficiency thereof and provides an important reference for further adjusting soil erosion management measures.

Abstract: Methods and apparatus for discrimitive semantic transfer and physics-inspired optimization in deep learning are disclosed. A computation training method for a convolutional neural network (CNN) includes receiving a sequence of training images in the CNN of a first stage to describe objects of a cluttered scene as a semantic segmentation mask. The semantic segmentation mask is received in a semantic segmentation network of a second stage to produce semantic features. Using weights from the first stage as feature extractors and weights from the second stage as classifiers, edges of the cluttered scene are identified using the semantic features.

Abstract: Methods, apparatus, systems and articles of manufacture for loss-error-aware quantization of a low-bit neural network are disclosed. An example apparatus includes a network weight partitioner to partition unquantized network weights of a first network model into a first group to be quantized and a second group to be retrained. The example apparatus includes a loss calculator to process network weights to calculate a first loss. The example apparatus includes a weight quantizer to quantize the first group of network weights to generate low-bit second network weights. In the example apparatus, the loss calculator is to determine a difference between the first loss and a second loss. The example apparatus includes a weight updater to update the second group of network weights based on the difference. The example apparatus includes a network model deployer to deploy a low-bit network model including the low-bit second network weights.

Abstract: Methods and apparatus for discrimitive semantic transfer and physics-inspired optimization in deep learning are disclosed. A computation training method for a convolutional neural network (CNN) includes receiving a sequence of training images in the CNN of a first stage to describe objects of a cluttered scene as a semantic segmentation mask. The semantic segmentation mask is received in a semantic segmentation network of a second stage to produce semantic features. Using weights from the first stage as feature extractors and weights from the second stage as classifiers, edges of the cluttered scene are identified using the semantic features.

Abstract: A remote sensing classification method based on relative entropy includes: determining sample points of n types of ground objects in a study area and determining series remote sensing parameters; extracting, based on the sample points, remote sensing parameter values to form standard time series plots as a first distribution; taking remote sensing parameter values of to-be-classified pixels as a second distribution, determining, based on the second distribution and the first distribution, KL values of the to-be-classified pixels by using a KL-divergence formula, then obtaining n KL layers; and comparing n KL values of each to-be-classified pixel to classify it to be a type of ground objects with a minimum KL value.

Abstract: Certain aspects of the present disclosure relate to wireless communications, and more particularly, to techniques for xx. A method that may be performed by a user equipment (UE) includes receiving, from a network entity, a first transport block (TB) comprising one or more code blocks (CBs), transmitting, to the network entity, an indication of a preferred code block group (CBG) allocation scheme selected for re-transmission of at least one CB of the one or more CBs based, at least in part, on a decoding status for each of the one or more CBs, and receiving, from the network entity, the at least one CB re-transmitted in accordance with the preferred CBG allocation scheme.

Abstract: Methods and apparatus for discrimitive semantic transfer and physics-inspired optimization in deep learning are disclosed. A computation training method for a convolutional neural network (CNN) includes receiving a sequence of training images in the CNN of a first stage to describe objects of a cluttered scene as a semantic segmentation mask. The semantic segmentation mask is received in a semantic segmentation network of a second stage to produce semantic features. Using weights from the first stage as feature extractors and weights from the second stage as classifiers, edges of the cluttered scene are identified using the semantic features.

Abstract: The specification provides a method for at least two behavior records that are generated through triggering during running of the applet are obtained; then, a behavior feature of each behavior record is extracted; next, at least one feature combination is formed by using at least two behavior features of at least two successively generated behavior records, where each feature combination includes at least two behavior features, and a sequence of the at least two behavior features in the feature combination is the same as a time sequence of generating behavior records corresponding to the at least two behavior features; and finally, it is determined whether there is a feature combination that includes a predetermined feature combination of a malicious behavior record; and if there is a feature combination that includes the predetermined feature combination of a malicious behavior record, it is determined that the applet conducts a malicious behavior.

Abstract: Some embodiments of this specification disclose a privacy-preserving data risk prevention and control method, apparatus, and device. The method includes: obtaining a processing request for target privacy data; determining privacy data attribute information corresponding to the target privacy data, where the privacy data attribute information includes information of one or more dimensions needed for controlling use of the target privacy data; determining a control rule corresponding to the target privacy data based on the privacy data attribute information corresponding to the target privacy data, where the control rule corresponding to the target privacy data is constructed based on an expert knowledge base; and controlling compliance of the use of the target privacy data based on the determined control rule corresponding to the target privacy data in a process of responding to the processing request.

Abstract: An air conditioning system includes a main circuit having a multi-stage compressor, a condenser, a throttling element and an evaporator connected by pipelines; and a cooling branch, the inlet of which is connected to the main circuit between the condenser and the throttling element, and the outlet of which is connected to at least one of the first-stage suction port and the intermediate-stage suction port of the multi-stage compressor, wherein refrigerant from the cooling branch flows through the drive motor of the multi-stage compressor, and a regulating valve for controlling the opening of the cooling branch is provided on the cooling branch; and a control module that controls the opening of the regulating valve on the cooling branch based on the temperature of the outlet downstream of the drive motor on the cooling branch and the intermediate suction pressure of the intermediate-stage suction port of the multi-stage compressor.

Abstract: Embodiments of this specification provide computer-implemented methods, apparatuses, and computer-readable storage media for interface invocation request processing. In an example interface invocation request processing method, an invocation request for a first interface of an operating system is received from a client application, and the first interface is configured to obtain privacy data. First scenario information is obtained, where the first scenario information is description information of a use scenario of the first interface declared when the client application applies for an invocation permission of the first interface. Current scenario information of the client application is obtained. The invocation request is executed in response to at least that the current scenario information matches the first scenario information.

Abstract: An example apparatus for mining multi-scale hard examples includes a convolutional neural network to receive a mini-batch of sample candidates and generate basic feature maps. The apparatus also includes a feature extractor and combiner to generate concatenated feature maps based on the basic feature maps and extract the concatenated feature maps for each of a plurality of received candidate boxes. The apparatus further includes a sample scorer and miner to score the candidate samples with multi-task loss scores and select candidate samples with multi-task loss scores exceeding a threshold score.

Abstract: A battery includes a battery casing, a first cell, and a second cell. A partition structure is arranged in the battery casing. The first cell is arranged in the battery casing, and at least a part of the first cell is located on a first side of the partition structure. The second cell is arranged in the battery casing, and at least a part of the second cell is located on a second side of the partition structure, and the first side and the second side are oppositely arranged, such that the first cell and the second cell are arranged in a first direction. The partition structure is formed with a communication channel penetrating the first side and the second side.

Abstract: A battery includes a battery casing and an explosion-proof valve. The battery casing includes a first casing member and a second casing member, and the first casing member and the second casing member are welded to form a welded connection region between the first casing member and the second casing member. The welded connection region includes a first region, a second region, and a corner region, and two ends of the corner region are respectively connected to the first region and the second region. The explosion-proof valve is disposed at a corner position of a surface of the first casing member and adjacent to the welded connection region.