Abstract: Systems, methods, models, and training data for models are discussed, for determining vehicle positioning, and in particular identifying tailgating. Simulated training images showing vehicles following other vehicles, under various conditions, are generated using a virtual environment. Models are trained to determine following distance between two vehicles. Trained models are used in detection of tailgating, based on determined distance between two vehicles. Results of tailgating are output to warn a driver, or to provide a report on driver behavior. Following distance over time is determined, and simplified following distance data is generated for use at a management device.

Abstract: A technique including capturing, by one or more cameras of a set of cameras disposed about a vehicle, one or more images, wherein a surround view system of the vehicle is configured to render a surround view image using a first hardware accelerator based on the one or more images, determining that a first hardware accelerator is unavailable, and rendering the surround view image using a second hardware accelerator based on the captured one or more images.

Abstract: Embodiments described herein provide A method for training a neural network based model. The methods include receiving a training dataset with a plurality of training samples, and those samples are encoded into representations in feature space. A positive sample is determined from the raining dataset based on a relationship between the given query and the positive sample in feature space. For a given query, a positive sample from the training dataset is selected based on a relationship between the given query and the positive sample in a feature space. One or more negative samples from the training dataset that are within a reconfigurable distance to the positive sample in the feature space are selected, and a loss is computed based on the positive sample and the one or more negative samples. The neural network is trained based on the loss.

Abstract: Embodiments described herein provide a training framework for generative NLP models. Specifically, the training input, e.g., in the form of a sequence of tokens representing a user-agent dialogue, may be randomly masked for a few spans, which can be one or more tokens, one or more words, one or more sentences, or one or more paragraphs. These masked spans are replaced with their embeddings generated from pre-trained large language models are then used for training the NLP model.

Abstract: Systems, methods, and apparatuses of using biometric information to authenticate a first device of a user to a second device are described herein. A method includes storing, by the first device, a first key share of a private key and a first template share of a biometric template of the user. The second device stores a public key, and one or more other devices of the user store other key shares and other template shares. The first device receives a challenge message from the second device, measures biometric features of the user to obtain a measurement vector, and sends the measurement vector and the challenge message to the other devices. The first device receives partial computations, generated using a respective template share, key share, and the challenge message, from the other devices, uses them to generate a signature of the challenge message and send the signature to the second device.

Abstract: Systems, methods, models, and training data for models are discussed, for determining vehicle positioning, and in particular identifying tailgating. Simulated training images showing vehicles following other vehicles, under various conditions, are generated using a virtual environment. Models are trained to determine following distance between two vehicles. Trained models are used in detection of tailgating, based on determined distance between two vehicles. Results of tailgating are output to warn a driver, or to provide a report on driver behavior. Following distance over time is determined, and simplified following distance data is generated for use at a management device.

Abstract: Methods, systems, and devices for imprint recovery management for memory systems are described. In some cases, memory cells may become imprinted, which may refer to conditions where a cell becomes predisposed toward storing one logic state over another, resistant to being written to a different logic state, or both. Imprinted memory cells may be recovered using a recovery or repair process that may be initiated according to various conditions, detections, or inferences. In some examples, a system may be configured to perform imprint recovery operations that are scaled or selected according to a characterized severity of imprint, an operational mode, environmental conditions, and other factors. Imprint management techniques may increase the robustness, accuracy, or efficiency with which a memory system, or components thereof, can operate in the presence of conditions associated with memory cell imprinting.

Abstract: Sulfur species may be processed using an ionic liquid, a deep eutectic solvent, or a combination of both. An example of a method of such processing may include: supplying a first reaction medium including an iodine species and a first catalyst, wherein the first catalyst includes a first deep eutectic solvent, a first ionic liquid, or a first mixture of both; contacting the reaction medium with a first sulfur species, wherein the first sulfur species includes hydrogen sulfide, a sulfur-containing hydrocarbon, or any combination thereof; and reacting the first sulfur species with the reaction medium to produce a second sulfur species, hydrogen iodide, or a combination thereof.

Abstract: Vapor deposition precursor recovery systems and methods are provided. Methods and systems include a precursor container housing a precursor material and a carrier gas container housing a carrier gas. Systems and methods include a condenser assembly having a condenser in fluid connection with an exhaust line, a recycled precursor container, and a cooling circuit. Systems and methods include a vaporizer having one or more inlets in fluid connection with the precursor container and the carrier gas container, and an outlet in fluid connection with a gas distributor and the condenser assembly. Systems and methods include where the condenser is maintained at an internal temperature of greater than or about 10° C. below a boiling point of the precursor material and greater than or about 10° C. above a boiling point of the carrier gas.

Abstract: In one aspect, disclosed is an example method for use by a lighting device including a microphone and a light source, the method including: (i) receiving, by the lighting device via the microphone, an audio signal; (ii) decoding and extracting, by the lighting device, a lighting device control instruction from the received audio signal; and (iii) using, by the lighting device, at least the decoded and extracted lighting device control instruction as a basis to control operation of the light source.

Abstract: A method for probabilistic computing is provided. The method comprises specifying a target distribution for a computational model, wherein the target distribution is defined by a function. A number of coin flips are performed with a number of weighted coinflip devices, wherein weights for the coinflip devices are determined by the function. The number of coinflips are then converted to a random number from the target distribution according to outputs of the weighted coinflip devices, wherein a circuit uses the coin flips as inputs to randomly activate bits in a binary representation of the random number.

Abstract: A method and system for using steganographically-encoded time information as a basis to control when a client carries out a content-related action. The client processes for presentation one or more linear media streams, each defining a respective time sequence of frames of media content, and at least one of the streams being steganographically encoded with at least one watermark at a respective time-point within the linear media stream, the watermark encoding a timestamp of the respective time-point within the linear media stream according to a server clock. The client extracts the timestamp from the watermark and computes a time offset based on a difference between the extracted timestamp and a current time according to a client clock. And the client uses the computed time offset as a basis to determine when to carry out a content-related action in a given one of the one or more linear media streams.

Abstract: Techniques are disclosed for training a feature extraction model. A computing device can receive a training image and generate noised feature vectors using a feature extraction model characterized by first parameters and taking the training image as input. The computing device can determine the noised feature vectors by at least determining a feature vector for individual pixels in the training image and applying noise to each feature vector. The computing device can generate a reconstructed image using a reconstructor model characterized by second parameters and taking the noised feature vectors as input. The computing device can determine a reconstruction loss by comparing the training image with the reconstructed image and a noise loss using the noise applied to each feature vector. The computing device can update the first parameters based on the noise loss.

Abstract: A first forwarding VM may execute in a first availability zone and have a first IP address. Similarly, a second forwarding VM may execute in a second availability zone and have a second IP address. The first and second IP addresses may be recorded with a cloud DNS web service of a cloud provider such that both receive requests from applications directed to a particular DNS name acting as a single endpoint. A service cluster may include a master VM node and a standby VM node. An IPtable in each forwarding VM may forward a request having a port value to a cluster port value associated with the master VM node. Upon a failure of the master VM node, the current standby VM node may be promoted to execute in master mode and the IPtables may be updated to now forward requests having the port value to a cluster port value associated with the newly promoted master VM node (which was previously the standby VM node).

Abstract: A system and method for audio processing using artificial intelligence during a voice interaction is configured to improve early identification of disease and case management of a diagnosed member in a healthcare system. Audio data is generated from a voice call between customer service representative (CSR) of the healthcare system and then processed by one or more of a speech-to-text recognition model, a topic detection model, a speaker recognition model, a tone model, an intent identifier, and a sentiment identifier. Then the output of the audio processing is used to detect a condition and automatically perform an action.

Abstract: In one aspect, a drug delivery device is provided herein including: a monolithic body having a plurality of fluid ducts and at least one outlet duct formed therein; a plurality of drug cartridges attached to the body, the drug cartridges each including a reservoir for accommodating at least one drug wherein, the plurality of fluid ducts is arranged to convey the drugs from the drug cartridges to the at least one outlet duct; and, a needle support spaced from the body, the needle support including a needle configured for insertion into a patient for drug delivery, the needle support including adhesive for releasable securement to a patient, wherein, the needle support is connected to the body by a flexible tether through which passes at least one fluid passageway formed to convey drug from the at least one outlet duct to the needle.

Abstract: A method for controlling a dual fuel engine system includes determining a friction power loss amount of an internal combustion engine of the dual fuel engine system, where the friction power loss amount is based on an engine speed of the internal combustion engine and a friction torque estimate. The method also includes determining an accessory power loss amount of a power of the internal combustion engine, where the accessory power loss amount is based on the engine speed and an accessory torque estimate. The method further includes estimating a net engine power amount based on the accessory power loss amount and a brake power amount of the internal combustion engine, estimating an indicated diesel power, and estimating, based on the estimated net engine power, a first indicated engine power and a first gas power.

Abstract: Resources of a deduplicated storage appliance are monitored to generate a consumption matrix indicating resource utilization. A threshold map associated with a subsystem is accessed. The subsystem is responsible for an operation involving files stored in the appliance. The threshold map specifies resources required by the subsystem for the operation and thresholds below which multi-streaming for the operation is allowed. An assessment is made as to whether the operation should be handled by multi-streaming. The assessment includes determining whether the resources are underutilized or overutilized based on the consumption matrix and the threshold map. When the resources are overutilized, multi-streaming is not triggered. When the resources are underutilized, multi-streaming is triggered. Periodic reassessments are made to determine whether the operation should be handled by multi-streaming.

Abstract: Semantic analysis to resolve ambiguous user input data with respect to a request for data includes identifying tokens based on the text string, identifying a tables in a database, wherein a token indicates a column from a table and a token indicates a column from another table, identifying candidate join paths for joining tables, obtaining respective candidate results using the candidate join paths, outputting combined candidate results including values from the respective candidate results, obtaining second user input data indicating a selected value, identifying the request for data as unambiguous and identifying a selected join path based on the selected value, and, in response to identifying the request for data as unambiguous, outputting data responsive to the request for data using the selected join path.

Abstract: Disclosed is a method that includes a media presentation device (i) generating query fingerprint data representing video frames of a channel being rendered by the media presentation device, and transmitting the generated query fingerprint data from the device over a network to a server, (ii) responsive to transmitting the query fingerprint data to the server, receiving, from the server, channel-icon fingerprint data representing channel icons of multiple channels selected based at least on the server having detected that the query fingerprint data representing the video frames of the channel being rendered by the media presentation device matches the reference fingerprint data representing the frames of the multiple channels of video content, and (iii) responsive to receiving the channel-icon fingerprint data from the server, performing channel-icon fingerprint matching to determine which channel in the multiple channels is the channel being rendered by the media presentation device.