Abstract: A method and an apparatus for length-aware local tiling in a sparse attention module in a transformer in heterogeneous devices are provided. The method includes that a heterogeneous device including one or more GPUs: divides a transformed sparsity mask into a plurality of first tiles and obtaining one or more effective first tiles from the plurality of first tiles, where each effective first tile includes at least one non-zero element; loads the one or more effective first tiles into a shared memory in the one or more GPUs and loads a plurality of elements in a first matrix corresponding to the one or more effective first tiles into the shared memory; and performs multiplication by a first sampled dense-dense matrix multiplication (SDDMM) kernel in the sparse attention module in the transformer by fetching the one or more effective first tiles and the plurality of elements from the shared memory.

Abstract: A method and an apparatus for automatic speech recognition are provided. The method includes: generating a weight matrix for a layer of a plurality of layers in a neural network; dividing the weight matrix into a plurality of blocks, each block including a plurality of weights; selecting a pre-determined percentage of weights from at least one block for block-wise pruning; and generating a block-wise pruned weight matrix by setting the pre-determined percentage of weights selected from the at least one block to zero. The weight matrix includes a set of weights associated with the layer, the plurality of layers includes a first layer receiving a first input associated with one or more audio feature sequences, and the plurality of layers are executed on one or more processors. The method efficiently accelerates model inference using irregular pruning.

Abstract: A floating point pre-alignment structure for computing-in-memory applications includes a time domain exponent computing block and an input mantissa pre-align block. The time domain exponent computing block is configured to compute a plurality of original input exponents and a plurality of original weight exponents to generate a plurality of flags. Each of the flags is determined by adding one of the original input exponents and one of the original weight exponents. The input mantissa pre-align block is configured to receive a plurality of original input mantissas and shift the original input mantissas according to the flags to generate a plurality of weighted input mantissas, and sparsity of the weighted input mantissas is greater than sparsity of the original input mantissas. Each of the flags has a negative correlation with a sum of the one of the original input exponents and the one of the original weight exponents.

Abstract: The present invention relates to the technical field of smart home, and discloses a method for network configuration, a storage medium, a base station, a cleaning device, and a cleaning system. The method includes: generating network configuration prompt information in response to receiving a second Bluetooth MAC address sent by a second device; obtaining wireless connection information sent by a user terminal through the first Bluetooth connection; connecting to a wireless network according to the wireless connection information, and uploading the second Bluetooth MAC address to a server through the wireless network, in order that the user terminal interacts with the second device. Thus, a network configuration efficiency of connecting the first device and the second device to the wireless network can be improved.

Abstract: Disclosed are a catalyst component and a catalyst for olefin polymerization, and an olefin polymerization method. The catalyst component comprises magnesium, titanium, a halogen and an internal electron donor, wherein the internal electron donor comprises a monocarboxylic acid ester compound and a diether compound, and the molar ratio of the monocarboxylic acid ester compound to the diether compound is (0.0035-0.7):1. By using the catalyst, a polymer having both a high isotactic index and a high melt flow index can be prepared.

Abstract: Methods, systems, computer program products for determining the source of activity during interaction with a user interface are provided. The method comprises selecting one or more input devices from a plurality of available input devices coupled to the user interface and receiving respective measurements for the selected one or more input devices. Based on the received respective measurements, respective feature vectors for the one or more input devices are generated and then inputted to a pre-defined regression model. Then, the source of activity is determined based on a result received from the regression model.

Abstract: The present invention provides methods of using an inhibitor of the IL-6 signaling pathway to inhibit disease progression in a subject with a high-risk myelodysplastic syndrome (MDS) or treat low blast count acute myeloid leukemia (AML). Methods for identifying therapeutics for preventing MDS to AML progression are also provided.

Abstract: A memory unit with time domain edge delay accumulation for computing-in-memory applications is controlled by a first word line and a second word line. The memory unit includes at least one memory cell, at least one edge-delay cell multiplexor and at least one edge-delay cell. The at least one edge-delay cell includes a weight reader and a driver. The weight reader is configured to receive a weight and a multi-bit analog input voltage and generate a multi-bit voltage according to the weight and the multi-bit analog input voltage. The driver is connected to the weight reader and configured to receive an edge-input signal. The driver is configured to generate an edge-output signal having a delay time according to the edge-input signal and the multi-bit voltage. The delay time of the edge-output signal is positively correlated with the multi-bit analog input voltage multiplied by the weight.

Abstract: The present invention relates to the field of elevator braking, in particular to an action state detection method and system for an elevator brake, comprising: acquiring matching pairs of brake linings in the same braking state; constructing normal record data sets based on the matching pairs; grouping the normal record data sets based on braking power corresponding to samples in the normal record data sets to determine a median of elevator running speed corresponding to the samples in each group; dividing the operating state levels of the elevator brake according to the size of the median; constructing a data training set; training a constructed network model to obtain a trained network model; acquiring the operating state levels of an elevator to be detected by the trained network model, evaluating the elevator brake according to the operating state levels, and controlling the elevator.

Abstract: A system for one-touch split-mode conference access includes a server that transmits a first message associated with a conference with a client. A VOIP device displays a notification based on the first message. The VOIP device transmits a second message to transfer an audio component of the conference to the VOIP device. The audio component of the conference is continued on the VOIP device and the video component of the conference is retained on the client.

Abstract: A hierarchical health index evaluation method for an intelligent substation includes: obtaining a basic health index of a device based on static information and dynamic information of the device; obtaining a device correlation based on a communication connection relationship between the device and another device, and correcting the basic health index of the device based on the device correlation to obtain a health index of the device; obtaining a layer-based health index of a layer based on a device health index and a device importance weight of the layer; obtaining a whole-station health index of an intelligent substation based on a layer-based health index of each layer and a sum of device importance weights of each layer; and regulating the intelligent substation based on a health index of each device in the intelligent substation, the layer-based health index of each layer, and the whole-station health index.

Abstract: A method for preparing an electrode material includes: a) producing a microspherical precursor by way of co-precipitation; b) forming an intermediate product by calcining the precursor with a stoichiometric amount of sodium carbonate, lithium carbonate and a structural stabilizer; and c) performing an ion exchange process to the intermediate product under molten LiNO3/LiCl to form a lumpy residue. An electrode for lithium-ion battery includes an electrode material having a general formula of Li[Li1/3(TMxAly)]O2, and lithium-ion battery comprising an electrode such as a cathode having the above electrode material are also addressed.

Abstract: A method, apparatus, and a non-transitory computer-readable storage medium for generating heterogenous platform code. The method may obtain a neural network model. The neural network model may be programed to run on at least one platform. The method may also obtain an initial intermediate representation (IR) code by encoding the neural network model, and obtain a target IR code by adding decorations to the initial IR code based on a target platform. The method may also output an executable code optimized to run on the target platform by decoding the target IR code.

Abstract: A method of cyclic etching, comprising: (A) depositing, prior to cyclically etching a substrate through a mask opening, a pre-etch protection layer conformally over the mask, sidewalls of the mask defining the mask opening; and an exposed portion of the substrate exposed through the mask opening, the pre-etch protection layer deposited to a first thickness; and (B) cyclically etching the substrate by: (i) depositing a protection layer in the opening of the mask, the protection layer deposited to a second thickness that is less than half of the first thickness; (ii) etching through a portion of the protection layer disposed on the substrate and etching the substrate; and (iii) repeating (i) and (ii) until an end point is reached.

Abstract: A client is configured to control one or more functions of another client remotely via a wireless connection. Enabling control of the one or more functions of another client includes receiving a control command from a first client. The control command indicates a function to be performed by a second client. The control command is transmitted to the second client when a binding status indicates a binding between the first client to the second client.

Abstract: An approach is provided that retrieves fingerprint configuration sets corresponding to a received data source and uses the configuration sets to generate fingerprints that correspond to the data source. These fingerprints are compared to a number of fingerprints that are stored in a repository. If a match is found, then the data quality configuration set is retrieved from the repository and used to perform a data quality analysis. On the other hand, if a match is not found, then one of the configuration sets is selected to perform the data quality analysis on the received data source and the repository is updated so that the selected fingerprint configuration set corresponds to the received data source.

Abstract: A process for enzymolysis of chicken includes the follow steps: Step 1, processing chicken into small pieces, transferring the small pieces into a reaction kettle, adding water and preheating; Step 2, stirring the chicken preheated in step 1, raising a temperature to 40 to 60° C., adding protease to perform enzymolysis; and Step 3, heating an enzymatic hydrolysate obtained in step 2 to 85 to 100° C., holding the temperature, and terminating the enzymolysis reaction. The protease is a composite flavourzyme including exo-protease produced by fermentation of Aspergillus oryzae and endo-protease produced by fermentation of Bacillus subtilis.

Abstract: The application relates to a convolutional neural network acceleration method and system based on a Cortex-M processor, and a medium. The method comprises: setting a MCR instruction and a CDP instruction according to common basic operators of a convolutional neural network, the common basic operators comprising a convolution operator, a Relu activation operator, a pooling operator, a table look-up operator and a quantization operator; and configuring an internal register of a convolutional neural network coprocessor through the MCR instruction, and then enabling the common basic operators of the convolutional neural network through the CDP instruction. Through the application, problems of inefficiency, high cost and inflexibility of a cyclic neural network algorithm in the execution of a processor are solved, and the basic operators needed for the cyclic neural network to be executed through a coprocessor instruction set are realized.

Abstract: Provided is a method, computer program product, and system for classifying a set of data items based on format organizations. A processor may determine at least one format organization of a set of data items. The format organization of a data item indicates a symbol type of at least one continuous symbol in the data item and a number of the at least one continuous symbol. The processor may determine at least one candidate data class for the set of data items from a plurality of predetermined data classes based on the at least one format organization. The processor may classify the set of data items into at least one target data class selected from the at least one candidate data class. In this way, the set of data items can be efficiently classified.

Abstract: A computer-implemented method includes: reading a vector of a first table in a database, the vector including counts of a plurality of keywords in the first table, the plurality of keywords including a first keyword and a second keyword; determining a first custom attribute describing the first table, the first custom attribute having a vector including counts of at least a first portion of the plurality of keywords in the first table; determining a multiplier of the first custom attribute, the multiplier being a number of other tables that reference the first custom attribute; and revising the vector of the first table based on the first custom attribute.