Abstract: A method includes receiving audio data corresponding to an utterance and generating a pair of positive audio data examples. Here, each positive audio data example includes a respective augmented copy of the received audio data. For each respective positive audio data example, the method includes generating a respective sequence of encoder outputs and projecting the respective sequence of encoder outputs for the positive data example into a contrastive loss space. The method also includes determining a L2 distance between each corresponding encoder output in the projected sequences of encoder outputs for the positive audio data examples and determining a per-utterance consistency loss by averaging the L2 distances. The method also includes generating corresponding speech recognition results for each respective positive audio data example. The method also includes updating parameters of the speech recognition model based on a respective supervised loss term and the per-utterance consistency loss.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for modulating language model biasing. In some implementations, context data is received. A likely context associated with a user is determined based on at least a portion of the context data. One or more language model biasing parameters based at least on the likely context associated with the user is selected. A context confidence score associated with the likely context based on at least a portion of the context data is determined. One or more language model biasing parameters based at least on the context confidence score is adjusted. A baseline language model based at least on the one or more of the adjusted language model biasing parameters is biased. The baseline language model is provided for use by an automated speech recognizer (ASR).

Abstract: Various implementations relate to techniques, for controlling smart devices, that are low latency and/or that provide computational efficiencies (client and/or server) and/or network efficiencies. Those implementations relate to generating and/or utilizing cache entries, of a cache that is stored locally at an assistant client device, in control of various smart devices (e.g., smart lights, smart thermostats, smart plugs, smart appliances, smart routers, etc.). Each of the cache entries includes a mapping of text to one or more corresponding semantic representations.

Abstract: A system may receive an entity identifier for a user. The entity identifier may identify an entity that is associated with a category. The system may identify channel(s) for the category by obtaining data for the user, and analyzing the data to determine a score for each channel. The system may use the scores to select one or more of the channels. The system may provide content related to the selected channel(s) to a client device of the user, and the user may interact with the content via the client device.

Abstract: If a secure element accesses a resource that is separate from the secure element, conducting a secure transaction can be inefficient in terms of power or time. Power usage is inefficient if the resource is never permitted to sleep, and transaction time is inefficient if the resource is permitted to sleep, and the user experiences a delay. To enable dual efficiency, a resource entity is permitted to be powered down. The resource entity is then powered up speculatively by an activation controller. The activation controller predicts an upcoming secure transaction based on sensor output, such as a position fix or a detected electromagnetic field. Based on monitored sensor output, the activation controller issues an activation signal to power up the secure element or the resource entity prior to initiation of the upcoming secure transaction. Thus, power can be conserved without introducing a transaction-processing latency.

Abstract: Generally, the present disclosure is directed to systems and methods that perform adaptive optimization with improved convergence properties. The adaptive optimization techniques described herein are useful in various optimization scenarios, including, for example, training a machine-learned model such as, for example, a neural network. In particular, according to one aspect of the present disclosure, a system implementing the adaptive optimization technique can, over a plurality of iterations, employ an adaptive learning rate while also ensuring that the learning rate is non-increasing.

Abstract: The present disclosure provides computer-implemented methods, systems, and devices for enabling frictionless transactions at a merchant location using audio communication. A central hub device receives transaction notification data describing a transaction with a user computing device. The central hub device accesses one or more location determination signals to estimate a location of the user computing device with the merchant location. The central hub device transmits transaction data to the estimated location of the user computing device using audio-based communications, the transaction data including an audio key that, when detected by the user computing device, causes the user computing device to automatically activate an application for providing payment data. The central hub device receiving transaction payment data from the user computing device. The central hub device executes the transaction by transmitting transaction data to a payment system.

Abstract: Generating audio tracks is provided. The system selects a digital component object having a visual output format. The system determines to convert the digital component object into an audio output format. The system generates text for the digital component object. The system selects, based on context of the digital component object, a digital voice to render the text. The system constructs a baseline audio track of the digital component object with the text rendered by the digital voice. The system generates, based on the digital component object, non-spoken audio cues. The system combines the non-spoken audio cues with the baseline audio form of the digital component object to generate an audio track of the digital component object. The system provides the audio track of the digital component object to the computing device for output via a speaker of the computing device.

Abstract: A system for displaying a virtual image to a user includes a light engine to generate a display light representing the virtual image, a diffractive waveguide, and an incoupler and outcoupler that are each optically coupled to the diffractive waveguide. In operation, the incoupler receives the display light from the light engine and directs the received display light to the diffractive waveguide, and the outcoupler directs at least a portion of the display light from the diffractive waveguide to an eye of the user. The diffractive waveguide is configured to converge a first component light of the generated display light at a first focal distance from the eye of the user, and to converge one or more additional component lights of the generated display light at one or more distinct other focal distances from the eye of the user.

Abstract: The present disclosure provides systems, methods, and computer program products for providing efficient embedding table storage and lookup in machine-learning models.

Abstract: This document describes systems and techniques for protecting the security of information in content selection and distribution. In one aspect, a method includes receiving, by a first computing system of MPC systems, a digital component request including distributed point functions that represent a secret share of a respective point function that indicates whether a user of the client device is a member of a first user group. Selection values are identified. Each selection value corresponds to a respective digital component, a set of contextual signals, and a respective second user group identifier for a respective second user group to which the respective digital component is eligible to be distributed. A determination is made, for each selection value and using the distributed point functions in a secure MPC process, a candidate parameter that indicates whether the second user group identifier matches a user group that includes the user as a member.

Abstract: Methods and apparatus for estimating the fidelity of quantum hardware. In one aspect, a method includes accessing a set of quantum gates; sampling a subset of quantum gates from the set of quantum gates, wherein the subset of quantum gates defines a quantum circuit; applying the quantum circuit to a quantum system and performing measurements on the quantum system to determine output information of the quantum system; calculating output information of the quantum system based on application of the quantum circuit to the quantum system; and estimating a fidelity of the quantum circuit based on the determined output information and the calculated output information of the quantum system.

Abstract: Techniques and apparatuses are described for multiple-input multiple-output transmissions using adaptive phase-changing devices. In aspects, a base station selects one or more adaptive phase-changing devices, APDs, to use in at least one communication path for multiple-input, multiple-output, MIMO, transmissions. The base station can perform a channel characterization process for the at least one communication path using the at least one APD and at least one UE. Based on results of the channel characterization process, the base station configures the at least one APD by which to implement single user-MIMO communication with a UE or multiple user-MIMO communication with multiple UEs. By so doing, the base station may implement MIMO transmissions using APDs to communicate with the at least one UE using same time and frequency resources, which can improve spectral efficiency of a wireless network.

Abstract: The present disclosure provides a closed loop, self-learning system that automatically optimizes what experiences should be presented to each customer. Instead of relying on rules and external targeting, it observes customer reactions to continuously improve performance and adapt to environment changes.

Abstract: A method for identifying malicious software includes receiving and executing a software application, identifying a plurality of uniform resource identifiers the software application interacts with during execution of the software application, and generating a vector representation for the software application using a feed-forward neural network configured to receive the plurality of uniform resource identifiers as feature inputs. The method also includes determining similarity scores for a pool of training applications, each similarity score associated with a corresponding training application and indicating a level of similarity between the vector representation for the software application and a respective vector representation for the corresponding training application.

Abstract: Systems and methods for automatic evaluation of the quality of NLG outputs. In some aspects of the technology, a learned evaluation model may be pretrained first using NLG model pretraining tasks, and then with further pretraining tasks using automatically generated synthetic sentence pairs. In some cases, following pretraining, the evaluation model may be further fine-tuned using a set of human-graded sentence pairs, so that it learns to approximate the grades allocated by the human evaluators.

Abstract: The present disclosure provides systems and method for determining a background noise level. The device may receive audio from two or more microphones. The audio may include a first signal and a second signal, such that each microphone receives its own signal. The time, loudness, frequency of the first and second signals may be compared to determine the source of the audio, such as whether the audio is the user's voice or background noise. Based on the source of the audio, the audio may be suppressed to reduce false estimations when calculating the background noise level.