Abstract: The present disclosure discloses an adaptive high-precision compression method and system based on a convolutional neural network model, and belongs to the fields of artificial intelligence, computer vision, and image processing. According to the method of the present disclosure, coarse-grained pruning is performed on a neural network model by using a differential evolution algorithm first, and the coarse-grained space is quickly searched through an entropy importance criterion and an objective function with good guidance to obtain a near-optimal neural network structure. Then fine-grained search space is built on the basis of an optimal individual obtained from the coarse-grained search, and fine-grained pruning is performed on the neural network model by a differential evolution algorithm to obtain a network model with an optimal structure. Finally, the performance of the optimal model is restored by using a multi-teacher multi-step knowledge distillation network to reach the precision of an original model.

Abstract: A system, method, and computer program product are configured to extract key value pair (KVP) data from one or more documents; obtain a content index for the one or more documents; and enhance the content index with the KVP data to provide a KVP content index.

Abstract: A device and a method for dynamic crack detection are provided. The device includes a probe configured to acquire permanent magnetic perturbation (PMP) data and motion-induced eddy current (MIEC) data of an inner surface of a pipe, and a controller, in communication with the probe, configured to: receive the PMP data and the MIEC data from the probe; determine the PMP data and the MIEC data as data of a crack in a case where the PMP data and the MIEC data conform to characteristics of a crack defect signal; and determine characteristic data of the crack based on the MIEC data. The PMP data and MIEC data of the inner surface of the pipe are acquired by a probe, which can detect crack defects in gas pipes under high-speed movement, thus improving the efficiency and stability of dynamic crack detection.

Abstract: Constructing a motion vector candidate list for inter-prediction in video coding is provided. A video encoder performs inter prediction for a frame of the video to generate prediction residuals. The inter prediction is performed using motion vectors (MVs) for blocks of the frame. To determine the motion vector of a block, the encoder determines an integer-valued MV for the block based on a reference frame and selects, based on the integer-valued MV, a subset of candidate MVs from a list of MVs including MVs of neighboring blocks. The encoder identifies a MV from the subset of candidate MVs for the block and searches in a neighborhood of the identified MV for a refined MV. The encoder calculates the prediction residuals for the block based on a reference block in the reference frame pointed by the refined MV and encodes the prediction residuals into a bitstream representing the video.

Abstract: An integrated transformer includes: two magnetic cores, wherein each magnetic core includes two magnetic yokes arranged opposite to each other along a first direction and two magnetic columns located between the two magnetic yokes, the two magnetic columns are arranged opposite to each other along a second direction; two short-circuited magnetic blocks, wherein one magnetic yoke in one of the magnetic cores is connected to one magnetic yoke in the other magnetic core through one of the short-circuited magnetic blocks to connect the two magnetic cores in series into a ring; four first windings, wound on four magnetic columns of the two magnetic cores, respectively; and a second winding, wound on the magnetic core or the short-circuited magnetic block, wherein magnetic flux directions generated by the second winding in two magnetic columns of any one of the magnetic cores are the same.

Abstract: A secondary battery including an electrode assembly, a tab assembly, and a packaging pouch. The electrode assembly is accommodated in the packaging pouch. The packaging pouch includes a sealing portion, the tab assembly includes tabs and insulating adhesives, and the tabs are connected to the electrode assembly and run through the sealing portion to extend out of the packaging pouch. The insulating adhesives are disposed on the tabs and connect the tabs and the packaging pouch. The insulating adhesive includes a first insulating adhesive layer, a second insulating adhesive layer, and a third insulating adhesive layer sequentially stacked in a first direction, the first insulating adhesive layer is connected to the tab, and the third insulating adhesive layer is connected to the packaging pouch.

Abstract: Embodiments of the present disclosure relate to feature crossing for machine learning. According to example embodiments of the present disclosure, a method comprises determining a semantic correlation relationship between a plurality of feature categories, the semantic correlation relationship indicating respective degrees of semantic correlation between respective pairs of feature categories among the plurality of feature categories; obtaining at least two features classified in at least two of the plurality of feature categories for machine learning; and performing feature crossing on the at least two features based on the semantic correlation relationship.

Abstract: Entity scoring calibration includes receiving values for a metric by which to calibrate a reputation scoring model. The reputation scoring model is usable to determine a reputation score. The reputation scoring model to be calibrated is based at least in part on a set of model parameters associated with reputation scoring components. It further includes receiving a plurality of feedback items pertaining to one or more entities. It further includes calibrating the reputation scoring model at least in part by adjusting at least some of the model parameters associated with the reputation scoring components such that reputation scores generated by the calibrated reputation scoring model track the values for the metric.

Abstract: A system includes a first sense amplifier having a first data input, a second data input, a clock input configured to receive a first clock signal, a first output, and a second output. The system also includes a second sense amplifier having a first data input, a second data input, a clock input configured to receive a second clock signal, a first output, and a second output. The first data input of the second sense amplifier is coupled to the first data input of the first sense amplifier, and the second data input of the second sense amplifier is coupled to the second data input of the first sense amplifier. The system also includes a set-reset (S-R) latch coupled to the first output and the second output of the first sense amplifier, and coupled to the first output and the second output of the second sense amplifier.

Abstract: The present disclosure provides an in-situ electric stratified remediation device and method for accurate position in remediation depth in sediment. the device includes a telescopic rod, a porous cylinder, a water storage tank, a curing agent storage device, and a depth determining module. The water storage tank is configured to collect sediment pore water and contaminants converged along a surface groove of a columnar electrode; and the fixed determining module is configured to send an indication information in response to that the curing agent storage device is inserted below a mud-water interface, while triggering release of curing agent from a wall of the curing agent shroud.

Abstract: Methods and apparatus for forming a barrier layer are provided herein. In some embodiments, a method of forming a barrier layer on a substrate includes treating an exposed layer deposited on a substrate and within a feature of the substrate by pulsing a bias power applied to a substrate support supporting the substrate while exposing the layer to a plasma. The exposed layer can be deposited by an atomic layer deposition process, and can be, for example, a tantalum nitride layer. The bias power can be up to 500 watts of RF power at a pulse frequency of about 1 Hz to about 10 kHz. The bias power can be pulsed uniformly or at multiple different levels.

Abstract: One embodiment of the invention provides a method for creating an inverted index. The method comprises receiving documents to be indexed, classifying the documents to determine document types of the documents, and analyzing the documents to identify crucial portions of the documents. The method further comprises generating, based on the analyzing, inverted indexes corresponding to indexing words contained in the documents. Each inverted index corresponding to each indexing word includes a document type tier identifying at least one document type of at least one document containing the indexing word, and the inverted index further includes a crucial list identifying each document which contains the indexing word in a crucial portion of the document if the indexing word occurs in the crucial portions of the documents. The method further comprises selectively caching each inverted index based on whether the inverted index includes a crucial list.

Abstract: One embodiment of the invention provides a method for creating an inverted index. The method comprises receiving documents to be indexed, classifying the documents to determine document types of the documents, and analyzing the documents to identify crucial portions of the documents. The method further comprises generating, based on the analyzing, inverted indexes corresponding to indexing words contained in the documents. Each inverted index corresponding to each indexing word includes a document type tier identifying at least one document type of at least one document containing the indexing word, and the inverted index further includes a crucial list identifying each document which contains the indexing word in a crucial portion of the document if the indexing word occurs in the crucial portions of the documents. The method further comprises selectively caching each inverted index based on whether the inverted index includes a crucial list.

Abstract: A method is applied to the electronic device, and the electronic device includes an audio system capable of emitting an ultrasonic wave. The method includes the electronic device determines that a target application is started, where the target application is an application that reuses the audio system with the ultrasonic wave; and when a system volume value is less than a preset volume value, the electronic device adjusts the system volume value to be greater than or equal to the preset volume value, and adjusts a volume value of the target application based on an adjusted system volume value, so that listening volume emitted by the electronic device remains unchanged.

Abstract: Embodiments of the present disclosure relate to semantic parsing for short text. According to embodiments of the present disclosure, a method is proposed. The method comprises: obtaining a set of sentences associated with a short text, each of the set of sentences containing all of words in the short text; determining a set of syntactic features associated with the set of sentences, each of the set of syntactic features indicating at least one of a constituency relation and a dependency relation of the corresponding sentence; and determining a semantic structure of the short text based on the set of syntactic features.

Abstract: Techniques for optimizing and deploying deep neural network (CNN) machine learning models for inference using static analysis are described. A method includes obtaining a deep neural network (DNN) machine learning (ML) model, generating an intermediate representation for the ML model, the intermediate representation including one or more nodes corresponding to one or more operators utilized by the ML model, identifying, for at least one node of the intermediate representation, an optimized schedule for at least one operator corresponding to the at least one node using a static analysis that is based on a hardware-specific cost model, generating an optimized intermediate representation using the optimized schedule that is optimized for execution on a hardware platform, and generating code corresponding to the ML model based at least in part on the optimized intermediate representation, wherein the code is specific to the hardware platform.

Abstract: A method for implementing formal verification of an optimized multiplier via symbolic computer algebra (SCA)-satisfiability (SAT) synergy includes: systematically recovering, by a reverse engineering algorithm, an adder tree from an optimized multiplier; 2) generating, by a constraint satisfaction algorithm, a reference multiplier only by using an adder based on a constraint condition; and 3) combining, by an SCA-based and SAT-based verification method, complementary advantages of SCA and SAT. In the verification framework, the method introduces a reference multiplier generator for generating a correct reference multiplier. The correct reference multiplier has both a structure similar to a structure of the optimized multiplier and a clear adder boundary. The clear adder boundary allows proving correctness of the correct reference multiplier through SCA-based verification.

Abstract: A cleaning attachment and rotary tool for cleaning are provided. A cleaning attachment includes a housing, a plurality of bristles, and a drive shaft extending from the housing in an axial direction, the drive shaft being coupled to the housing such that rotation of the drive shaft causes rotation of the housing. The housing includes a reservoir, a valve for dispensing the fluid from the reservoir, and at least one resilient element surrounding an axis defined by the drive shaft. Translation of the drive shaft along the axial direction causes compression of the resilient element, and compression of the resilient element causes the valve to open to dispense fluid from the reservoir. A rotary tool for cleaning includes the cleaning attachment and a rotary tool having a motor, power source, and a chuck configured to receive the drive shaft of the cleaning attachment.

Abstract: In an approach, a processor performs document layout analysis on a document generating a plurality of textual regions; extracts characteristics from each of the plurality of textual regions and associates the respective characteristics to the respective textual region as metadata; classifies each of the plurality of textual regions as an optical character recognition (OCR) region, non-OCR valuable region, or non-OCR non-valuable region using a classifier; performs OCR on each OCR region generating an OCR output; identifies associated constant OCR data from a constant OCR data repository for each non-OCR valuable region; merges the associated constant OCR data with the OCR output generating a complete OCR data for the received document; performs data extraction on the complete OCR data to identify data fields and key-value pairs generating extracted data; and determines whether the extracted data is valid based on a set of rules.