Abstract: A multidimensional knowledge transfer model for predicting CTR of a computational ad, the multidimensional knowledge transfer model comprises: a pre-processor for building an ad group node graph based on similarities among ad group nodes, an ad campaign node graph from merging the ad group node graph, and an ad account node graph from merging the ad campaign node graph. The multidimensional knowledge transfer model further comprises a multi-knowledge CTR prediction model for each of the ad account, ad campaign, and ad group layers. The multi-knowledge CTR prediction model predicts the respective node's CTR from the ad account node graph, ad campaign node graph, or ad group node graph, features of the audience group, and features of the node having its CTR predicted appended with the hidden vector of its parent node extracted from the upper layer multi-knowledge CTR prediction model.

Abstract: A 3D building model generator has a server that constructs a 3D texture model for each floor in a building by extruding a 2D floor plan to 3D using user-inputted heights, and adding photos captured by a smartphone or mobile device as textures to image planes for surfaces in the 3D model. The server displays on the smartphone a floor-plan portion, or a virtual 3D display based on the camera orientation with a facing wall highlighted and confirmed by a smartphone user. The user moves mapping points overlaid upon the photo to wall corners to map the photo to the facing surface in the model. The user flags rooms and exits such as doors and elevators as walkable. The server stacks the 3D texture model for the floors and connects them into a walkable network of the rooms and exits that are flagged as walkable. Hashing removes duplicate floorplans.

Abstract: A line-scanning three-dimensional sensing system measures a surface profile of an object. In the system, a dispersion optical module (DOM) performs a forward optical process of chromatically dispersing a polychromatic linear light beam into constituent narrowband linear light beams (CNLLBs) and focusing the CNLLBs on different focal planes to form a rainbow light pattern for illuminating a scanned surface of the object. The illuminated object displays an information-bearing color image (IBCI) containing height information of the scanned surface. The DOM captures the IBCI, and performs a backward optical process of optically condensing the captured IBCI to form an elongated light pattern. The backward optical process is an inverse of the forward one. A slit spatially filters the elongated light pattern to form an output light line. A height profile of the scanned surface is obtained by analyzing a spectral content at each point of the output light line.

Abstract: A method for detecting and tracking target object in a captured video using convolutional neural network (CNN) is provided. The method includes: inputting image data into a detecting model to generate detection results, wherein the detecting model is constructed by the CNN; inputting the image data into tracking models to generate tracking results; performing detection score enhancement operation according to the detection results and the tracking results to obtain enhanced detection results; matching the enhanced detection results and the tracking results by a matching operation; processing matched results and unmatched target detection results and unmatched target tracking results; and selectively updating the tracking models using tracking reliability estimation according to the matched results.

Abstract: A computer-implemented method for compliance checking related to goods in import supply chain logistics is provided, including: adding received legality data elements into a blockchain ledger corresponding to an importation of the goods; establishing a smart contract of the blockchain ledger; adding received sensing data elements corresponding to the smart contract into the blockchain ledger; adding electronic lock (e-Lock) status data corresponding to the smart contract into the blockchain ledger; and adding one or more inspection results corresponding to the smart contract into the blockchain ledger. Furthermore, determining, by the smart contract, whether the goods are within the compliance according to the received legality data elements, the received sensing data elements, the e-Lock status data and the one or more inspection results, wherein in response to determining that the goods are within the compliance, unsealing the e-Locks by sending a control signal to the e-Locks.

Abstract: A Hybrid TEE device allows a Trusted Execution Environment (TEE) by incorporating hardware comprising a Cache Purging Controller, a Memory Isolation Gateway, and a Memory Clean Up into a System on a Chip device, a general purpose computing device, or a special purpose or proprietary computing or electronic device. The addition of the hardware enables a method of protecting the Trusted Execution Environment and thus reducing vulnerability to malicious software or other program code.

Abstract: The present invention provides a method and a system for remote imaging of explosive gases in an area. The method comprises: illuminating the area with a light source having a uniform light intensity distribution over an infrared wavelength range; acquiring images of the illuminated area with an image sensor through gas detection filters having bandpass central wavelength corresponding to absorption curves of target gases respectively; determining existence of the target gases based on the acquired images; predicting distribution of gas concertation for existing target gases respectively by using a non-linear prediction model; and constructing gas distribution images of the area based on the predicted distribution of gas concertation. The present invention can recognize different gases with overlapping absorption curves and provide more accurate prediction of gas concentrations.

Abstract: A method for classifying a glass object via acoustic analysis by a classifying apparatus is provided. The method including: receiving, by a processor, sound data of a knock sound generated by applying a knocking operation on the glass object; determining, by the processor, a type of the glass object by performing a knock-sound analysis to the sound data, wherein the type of the glass object includes an organic glass and an inorganic glass; if the type of the glass object is determined as the inorganic glass, receiving, by the processor, echo data of an echo induced by applying an ultrasonic-echo operation on the glass object; and determining, by the processor, a further type of the glass object by performing an echo-decay analysis to the echo data, wherein the further type of the glass object includes a crystal glass, a borosilicate glass and a soda-lime glass.

Abstract: A method for removing stale context in a service instance, comprising: generating a user message according to UID and request instruction of received request data; routing the user message to first service instance; if target context corresponding to the user ID is saved in a global cache of a database, loading the target context from the global cache; when the loaded target context does not belong to the first service instance, setting the loaded target context to belong to the first service instance, wherein the set target context is saved into a first local cache and the global cache; and instructing to remove one or more invalid target contexts corresponding to the user ID from one or more other service instances, wherein the invalid target contexts are target contexts which do not belong to the first service instance, such that stale contexts which are the invalid target context are removed.

Abstract: A multi-functional, computer-aided gastroscopy system optimized with integrated AI solutions is disclosed. The system makes use of multiple deep-learning neural models to achieve low latency and high-performance requirements for multiple tasks. The optimization is made at three levels: architectural, modular and functional level. At architectural level, the models are designed in such a way that it is able to accomplish HP infection classification and detection of some lesions for one inference in order to reduce computation costs. At modular level, as a sub-model of HP infection classification, the site recognition model is optimized with temporal information. It not only improves the performance of HP infection classification, but also plays important roles for lesion detection and procedure status determination. At functional level, the inference latency is minimized by configuration and resource aware optimization.

Abstract: A movable camera travels along an inspection path for optically inspecting an inspection surface of an object for defect detection. In planning the inspection path, a set of viewpoints on the inspection surface is generated. Each viewpoint is associated with a patch, which is a largest area of the inspection surface within a field of view (FOV) of the camera when the camera is located over the viewpoint for capturing an image of FOV. An effective region of the patch is advantageously predicted by a neural network according to a three-dimensional geometric characterization of the patch such that the predicted effective region is valid for defect detection based on the captured image. A valid area is one whose corresponding area on the captured image is not blurred and is neither underexposed nor overexposed. The inspection path is determined according to respective effective regions associated with the set of viewpoints.

Abstract: A lossless data compressor prevents normalization overruns on-the-fly as symbol occurrence counts are rounded to generate symbol frequencies, allowing an encoding table generator to generate encoding table entries without waiting for the symbol frequency table to finish filling. Rounding errors are accumulated as symbols are normalized and compensated for by reducing a symbol frequency when the symbol frequency is at least 2 and the accumulated errors have exceeded a threshold. The symbol frequency is also reduced when the number of remaining states in the encoding table is insufficient for a number of remaining unprocessed symbols and states for a current encoding table entry. Since error compensation occurs as symbols are being normalized, encoding table generation is not forced to wait for all symbols in the block to be processed, reducing latency. Three pipeline stages can operate on three input blocks: symbol counting, normalization/error compensation/encoding table generation, and data encoding.

Abstract: Used batteries are screened based on a measured Sectional Constant-Current Impulse Ratio (SCCIR). A used battery is partially charged over a small voltage range using a Constant Current (CC) until a voltage target is reached, and the current integrated to obtain the CC charge applied, Qcc. Then the battery continues to be charged using a Constant Voltage (CV) of the voltage target until the charging current falls to a midrange current target before the battery is fully charged. The current is integrated over the CV period to obtain the CV charge applied, Qcv. The measured SCCIR is the ratio of Qcc to (Qcc+Qcv) and is input to a calibration curve function to obtain a modeled State of Health (SOH) value for sorting. The calibration curve function is obtained by aging new batteries to obtain SCCIR and SOH data that are modeled using a neural network.

Abstract: An apparatus installed in a head-mounted display (HMD) has a coupling prism formed by packing diagonally-reflective (DR) prisms together. Each DR prism has an internal diagonal plane that is at least partially reflective. A captured image of eye or environmental scene received by a DR prism is reflected to an image-leaving end surface thereof. Image-leaving end surfaces of all DR prisms are oriented along a same direction to optically multiplex the captured images to create a multi-channel image. An imaging sensor on the coupling prism images the multi-channel image, avoiding inter-channel interference caused by spillover of captured-image signals while allowing one imaging sensor instead of multiple ones to image the captured images. A micro display displays a visible image to one DR prism, whose internal diagonal plane reflects the visible image along a direction towards an eye. Hence, the apparatus also enables image displaying to a HMD wearer.

Abstract: An iterative multi-image camera orientation estimation comprising: capturing an image of a scene before the camera; detecting line segments in the scene; computing a maximum likelihood (ML) camera orientation by maximizing a likelihood objective by rotating the camera's X-Y-Z coordinate system such that it is being optimally aligned with the line segments in at least two of the frontal, the lateral, and the vertical orthogonal directions; estimating a maximum a-posteriori (MAP) camera orientation that maximizes an a-posteriori objective such that the MAP camera orientation is an optimal value in between the priori camera orientation and the ML camera orientation, and is closer to the one with smaller uncertainty; iterating the multi-image camera orientation estimation with the priori camera orientation and its corresponding priori camera orientation uncertainty set to the computed MAP camera orientation and its corresponding uncertainty respectively until the uncertainty is lower than a threshold.

Abstract: The present invention provides a method for achieving high availability of stream control transport protocol (SCTP) connection in a microservice system. The microservice system comprises at least one access network and a core network. The access network includes at least one base station comprising at least one SCTP peer. The core network includes at least one non-SCTP-capable microservice instance acting as a SCTP service client, multiple SCTP-capable microservice instances, a service node manager and a service delegator.

Abstract: A wearable electronic device including: a first portion and a second portion. The first portion includes a first conductive area on a first inner surface of the first portion and a third conductive area on a first outer surface of the first portion. The second portion includes one or more second conductive areas on a second outer surface of the second portion and a fourth conductive area on a second inner surface of the second portion, wherein the first portion is connected to the second portion. A circuit unit is disposed in a second space of the second portion, wherein the circuit unit is configured to enable an input/output unit of the circuit unit in response to the first conductive area and the second conductive areas are triggered.

Abstract: A data-compression analyzer can rapidly make a binary decision to compress or not compress an input data block or can use a slower neural network to predict the block's compression ratio with a regression model. A Concentration Value (CV) that is the sum of the squares of the frequencies and a Number of Zero (NZ) symbols are calculated from an un-sorted symbol frequency table. A rapid decision to compress is signaled when their product CV*NZ exceeds a horizontal threshold THH. During training, CV*NZ is plotted as a function of compression ratio C % for many training data blocks. Different test values of THH are applied to the plot to determine true and false positive rates that are plotted as a Receiver Operating Characteristic (ROC) curve. The point on the ROC curve having the largest Youden index is selected as the optimum THH for use in future binary decisions.

Abstract: A level shifter circuit uses standard n-channel and p-channel transistors except for a pair of Lateral-Diffusion Metal-Oxide-Semiconductor (LDMOS) transistors that have an added lateral diffusion under the gate between the source and the conduction channel, increasing the breakdown voltage. The source of each LDMOS transistor connects to a drain of a transient differential transistor that has its gate driven by a oneshot that generates a pulse after an input transition. After the pulse ends a holding differential transistor draws a smaller bias current from the LDMOS transistors. The source of each LDMOS transistor connects to the drain and gate of a p-channel sensing transistor that drives gates of mirror transistors generating mirrored currents to cross-coupled n-channel mirror transistors that drive both terminals of a bistable latch that holds the output using a floating ground between driver transistors of a Buck converter switched by the bistable latch.

Abstract: An On-Off Shift Keying (OOK) receiver has non-continuous Automatic Gain Control (AGC) that sets gain at the beginning of each frame and locks the gain setting for the remainder of the frame, preventing OOK data from causing AGC loop instability. An AGC controller initializes to maximum the gain settings for a Low-Noise Amplifier (LNA), Low-Pass Filter (LPF), and a Programmable Gain Amplifier (PGA) in series that power a rectifier generating a voltage output to a 1/0 data decision circuit. A level detector compares the voltage output to two thresholds. When both thresholds are exceeded, the AGC controller steps down gain settings until the voltage output is between the two thresholds, when the gain settings are locked for the remainder of the current frame. A frame detector resets the AGC controller between frames when a long series of 0 data between frames is detected. LNA gain is reduced last.