Abstract: A method for contextual biasing for speech recognition includes obtaining a base automatic speech recognition (ASR) model trained on non-biased data and a sub-model trained on biased data representative of a particular domain. The method includes receiving a speech recognition request including audio data characterizing an utterance captured in streaming audio. The method further includes determining whether the speech recognition request includes a contextual indicator indicating the particular domain. When the speech recognition request does not include the contextual indicator, the method includes generating, using the base ASR model, a first speech recognition result of the utterance by processing the audio data.

Abstract: A method includes receiving audio data corresponding to an utterance spoken by a user and processing, using a first recognition model, the audio data to generate a non-contextual candidate hypothesis as output from the first recognition model. The non-contextual candidate hypothesis has a corresponding likelihood score assigned by the first recognition model. The method also includes generating, using a second recognition model configured to receive personal context information, a contextual candidate hypothesis that includes a personal named entity. The method also includes scoring, based on the personal context information and the corresponding likelihood score assigned to the non-contextual candidate hypothesis, the contextual candidate hypothesis relative to the non-contextual candidate hypotheses.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a machine learning model to generate embeddings of inputs to the machine learning model, the machine learning model having an encoder that generates the embeddings from the inputs and a decoder that generates outputs from the generated embeddings, wherein the embedding is partitioned into a sequence of embedding partitions that each includes one or more dimensions of the embedding, the operations comprising: for a first embedding partition in the sequence of embedding partitions: performing initial training to train the encoder and a decoder replica corresponding to the first embedding partition; for each particular embedding partition that is after the first embedding partition in the sequence of embedding partitions: performing incremental training to train the encoder and a decoder replica corresponding to the particular partition.

Abstract: A system includes a neural network that includes a Mixture of Experts (MoE) subnetwork between a first neural network layer and a second neural network layer. The MoE subnetwork includes multiple expert neural networks. Each expert neural network is configured to process a first layer output generated by the first neural network layer to generate a respective expert output. The MoE subnetwork further includes a gating subsystem that selects, based on the first layer output, one or more of the expert neural networks and determine a respective weight for each selected expert neural network, provides the first layer output as input to each of the selected expert neural networks, combines the expert outputs generated by the selected expert neural networks in accordance with the weights for the selected expert neural networks to generate an MoE output, and provides the MoE output as input to the second neural network layer.

Abstract: The present disclosure provides for a heatshield that can be actively cooled during a rework process. The heatshield may include a backer plate, a metal plate, and/or a package pedestal. The backer plate may include one or more air inlet ports configured to be connected to an air compressor. Air inlet ducts may extend from the air inlet ports through at least a portion of the backer plate. A plurality of vents may extend from the air inlet ducts to a top surface of the backer plate such that the plurality of vents directs cooling gas forced into the heatshield towards the metal plate and a first BGA. The cooling gas may maintain the solder joint temperature of the first BGA package below the reflow temperature and below the solidus temperature of the solder joints to prevent reflow-related solder joint defects from occurring in the first BGA package during rework of a second BGA package.

Abstract: A method includes obtaining a first plurality of feature vectors associated with a first image and a second plurality of feature vectors associated with a second image. The method also includes generating a plurality of transformed feature vectors by transforming each respective feature vector of the first plurality of feature vectors by a kernel matrix trained to define an elliptical inner product space. The method additionally includes generating a cost volume by determining, for each respective transformed feature vector of the plurality of transformed feature vectors, a plurality of inner products, wherein each respective inner product of the plurality of inner products is between the respective transformed feature vector and a corresponding candidate feature vector of a corresponding subset of the second plurality of feature vectors. The method further includes determining, based on the cost volume, a pixel correspondence between the first image and the second image.

Abstract: Example embodiments allow for training of encoders (e.g., artificial neural networks (ANNs)) to facilitate dithering of images that have been subject to quantization in order to reduce the number of colors and/or size of the images. Such a trained encoder generates a dithering image from an input quantized image that can be combined, by addition or by some other process, with the quantized image to result in a dithered output image that exhibits reduced banding or is otherwise aesthetically improved relative to the un-dithered quantized image. The use of a trained encoder to facilitate dithering of quantized images allows the dithering to be performed in a known period of time using a known amount of memory, in contrast to alternative iterative dithering methods. Additionally, the trained encoder can be differentiable, allowing it to be part of a deep learning image processing pipeline or other machine learning pipeline.

Abstract: The specification describes methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a larger neural network from a smaller neural network. One of the described methods includes obtaining data specifying an original neural network and generating a larger neural network from the original neural network. The larger neural network has a larger neural network structure than the original neural network structure. The values of the parameters of the original neural network units and the additional neural network units are initialized so that the larger neural network generates the same outputs from the same inputs as the original neural network, and the larger neural network is trained to determine trained values of the parameters of the original neural network units and the additional neural network units from the initialized values.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium for managing a relationship between content and an environment for provisioning the content. In one aspect, a method includes receiving a request for a content item; and in response to receiving the request: selecting a creative from a plurality of creatives, the creative including a reference to a profile associated with one or more elements; retrieving content data from one or more content feeds bound to the elements; and delivering the creative and the content data to a user device.

Abstract: Key management for encrypted data includes establishing a cache of key decryption keys and periodically evicting the keys from the cache. A pool of key encryption keys also is created and periodically, selected key encryption keys are removed from service. Notably, the rate of removal of the encryption keys differs from the rate of cache eviction for the decryption keys. Thereafter, clear data is encrypted with a cipher to produce cipher text, and the cipher is encrypted with a selected key encryption key from the pool. Finally, in response to an access request for the clear data, an attempt to locate in the cache a key decryption key for the encrypted cipher is made. If attempt fails, the key decryption key is retrieved from remote memory. Finally, the encrypted cipher is decrypted with the located key, and the cipher text decrypted to produce the clear data.

Abstract: The present document describes an apparatus for reducing fabric dimpling in electronic devices and associated methods. The apparatus is used during assembly to prevent fabric, which is stretched over a perforated part (e.g., speaker housing), from dimpling into the holes (e.g., perforations) of the perforated part. The apparatus includes protrusions (e.g., pins), which act as a negative of the holes in the perforated part, to support the fabric as it is stretched over the perforated part. In particular, the protrusions are inserted through the perforations via an interior surface of the perforated part such that the protrusions are “proud” (slightly projecting from a surface) with respect to an exterior surface of the perforated part. The proudness of the protrusions may vary based on a degree of curvature of the perforated part at a location corresponding to a respective protrusion.

Abstract: A method including: receiving, by a computing device, a request from a user device for access to a hosted virtual machine; dedicating, by the computing device, a port to forward a cast of a particular hosted virtual machine instance to the user device; establishing a connection between the user device and the particular hosted virtual machine instance through the dedicated port; receiving, by the computing device and from the user device, instructions to execute an application on the particular hosted virtual machine instance; logging external calls made by the particular hosted virtual machine instance; and transmitting, by the computer device, the log of external calls to be stored on a server, the logs being synced by the server with the user device in substantially real time.

Abstract: A system includes a neural network implemented by one or more computers, in which the neural network includes an image depth prediction neural network and a camera motion estimation neural network. The neural network is configured to receive a sequence of images. The neural network is configured to process each image in the sequence of images using the image depth prediction neural network to generate, for each image, a respective depth output that characterizes a depth of the image, and to process a subset of images in the sequence of images using the camera motion estimation neural network to generate a camera motion output that characterizes the motion of a camera between the images in the subset. The image depth prediction neural network and the camera motion estimation neural network have been jointly trained using an unsupervised learning technique.

Abstract: A method includes receiving a build request containing build step instructions from a user. The build step instructions specify a usage of containers within memory hardware for building an output container. The containers include at least one private container having private contents and/or at least one public container having public contents. The method also includes authenticating the user initiating the build request and determining whether the user is authorized to access the private containers. When the user is authenticated and authorized to access the private containers, the method includes obtaining the containers specified by the build step instructions from the memory hardware, executing the build step instructions to build the output container while using the received containers, and outputting the built output container.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for adjusting neural network resource usage. One of the methods includes receiving a network input for processing by a task neural network, the task neural network comprising a plurality of neural network layers; receiving a usage input specifying a respective weight for each of one or more usage factors, wherein each usage factor impacts how many computational resources are used by the task neural network during the processing of the network input; and processing the network input using the task neural network in accordance with the usage input to generate a network output for the network input, comprising: selecting, based at least on the usage input, a proper subset of the plurality of neural network layers to be active while processing the network input, and processing the network input using only the selected neural network layers.

Abstract: This document describes techniques and systems for authentication management through IMU and radar. The techniques and systems use inertial sensor data from an inertial measurement unit (IMU) and/or radar data to manage authentication for a computing device. By so doing, the techniques conserve power, improve accuracy, or reduce latency relative to many common techniques and systems for computing-device authentication.

Abstract: The present disclosure provides systems for applying a compression load on at least part of an application specific integrated circuit (“ASIC”) ball grid array (“BGA”) package during the rework or secondary reflow process. The compression-loading assembly may include a top plate and a compression plate. The compression plate may exert a compression load on at least part of the ASIC using one or more compression mechanisms. The compression mechanisms may each include a bolt and a spring. The bolt may releasably couple the top plate to the compression plate and allow for adjustments to the compression load. The spring may be positioned on the bolt between the top plate and the compression plate and, therefore, may exert a force in a direction away from the top plate and toward the compression plate. The compression load may retain the solder joint and may prevent the solder separation defect during the reflow process.

Abstract: This document describes techniques and apparatuses directed to the mitigation of physical impact-induced mechanical stress damage to printed circuit boards through the utilization of a conductive shield track having a varied width (dynamic width). In an aspect, disclosed is a device that includes a printed circuit board, an electrical component on the printed circuit board in a shielded area, a conductive shield track on the printed circuit board, a component shield having a sidewall and a sidewall base, and solder disposed between the sidewall base and the conductive shield track to couple the component shield to the ground plane of the PCB to form a shielded compartment over the shielded area.

Abstract: A computer-implemented method for executing a default action on a touchscreen device is provided. The method includes receiving a touch input from a user on a touchscreen device and determining a context associated with the touch input. The context is associated with one or more actions including a default action. The method also includes determining that the received touch input comprises a default gesture, and performing the default action associated with the determined context. The default gesture may be a two-finger double-tap gesture. Systems and machine-readable media are also provided.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for predicting likelihoods of conditions being satisfied using recurrent neural networks. One of the systems is configured to process a temporal sequence comprising a respective input at each of a plurality of time steps and comprises: one or more recurrent neural network layers; one or more logistic regression nodes, wherein each of the logistic regression nodes corresponds to a respective condition from a predetermined set of conditions, and wherein each of the logistic regression nodes is configured to, for each of the plurality of time steps: receive the network internal state for the time step; and process the network internal state for the time step in accordance with current values of a set of parameters of the logistic regression node to generate a future condition score for the corresponding condition for the time step.