Abstract: Implementations disclosed herein relate to utilizing at least one existing manually engineered policy, for a robotic task, in training an RL policy model that can be used to at least selectively replace a portion of the engineered policy. The RL policy model can be trained for replacing a portion of a robotic task and can be trained based on data from episodes of attempting performance of the robotic task, including episodes in which the portion is performed based on the engineered policy and/or other portion(s) are performed based on the engineered policy. Once trained, the RL policy model can be used, at least selectively and in lieu of utilization of the engineered policy, to perform the portion of robotic task, while other portion(s) of the robotic task are performed utilizing the engineered policy and/or other similarly trained (but distinct) RL policy model(s).

Abstract: A speech recognition model includes an encoder network, a prediction network, and a joint network. The encoder network is configured to receive a sequence of acoustic frames characterizing an input utterance; and generate, at each of a plurality of output steps, a higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames. The prediction network is configured to: receive a sequence of non-blank symbols output by a final Softmax layer; and generate, at each of the plurality of output steps, a dense representation. The joint network is configured to generate, at each of the plurality of output steps based on the higher order feature representation and the dense representation, a probability distribution over possible speech recognition hypotheses. The joint network includes a stack of gating and bilinear pooling to fuse the dense representation and the higher order feature representation.

Abstract: In general, the subject matter described in this specification can be embodied in methods, systems, and program products for providing search results automatically to a user of a computing device. A spoken input provided by a user to a computing device is received. The spoken input is transmitted to a computer server system that is remote from the computing device. Search result information that is responsive to the spoken input is receiving by the computing device and in response to the transmitted spoken input. An alert is provided to the user that the device will connect the user to a target of the search result information if the user does not intervene to stop the connecting of the user. The user is connected to the target of the search result information based on a determination that the user has not intervened to stop the connecting of the user.

Abstract: This disclosure describes a method to calibrate a position of an optical image stabilization (OIS) lensing element 308 based on a temperature reading. The temperature reading is of one or more sensors, such as a Hall Effect sensor, and the position is a deviation from a center position, which is the position of the OIS lensing element 308 when it is not influenced by a force. A center drift coefficient is generated based on the temperature reading. A derived value for the position is adjusted based on the center drift coefficient. Additionally, a scaling sensitivity coefficient is generated based on the temperature reading. The adjusting of the derived value for the position is further based on the scaling sensitivity coefficient. The center drift coefficient and the scaling sensitivity coefficient are further based on maximum and minimum values for the Hall Effect sensor at the temperature reading and a calibration temperature.

Abstract: This document describes systems and techniques directed to touch sensor integration with enlarged active area displays. In aspects, a display includes a cover layer, an array of pixels, and a plurality of transistors that control an electrical activation of one or more pixels of the array of pixels. The plurality of transistors define a smaller area than the array of pixels such that at least one pixel of the array of pixels extends beyond the area defined by the plurality of transistors and above driving circuitry (“extended emitting area”). Variable pixel and/or transistor densities can support the extended emitting area. A touch sensor is integrated between the cover layer and the array of pixels and is operatively coupled to one or more touch trace routings that are, at least partially, disposed between the cover layer and one or more pixels within the extended emitting area.

Abstract: This document discloses technology generally related to an accessory that may wirelessly connect to one or more host devices such that the accessory is able to receive content from two or more of the host devices simultaneously. The accessory may have two or more wireless communication interfaces connected to two or more respective host devices via a type of wireless connection. The accessory may determine a time to simultaneously receive content from and/or transmit content to each of the host devices. The time may be determined by adjusting the start times for reception and/or transmission of content from the two or more host devices, and/or it may be determined based on the type of wireless connection. According to some examples, the accessory may determine a priority for each of the wireless connections using arbitration rules. The accessory may output the received content simultaneously.

Abstract: The various implementations described herein include methods, devices, and systems for managing battery usage and charging In one aspect, a method is performed at a battery charging device that includes one or more processors and memory. The method includes receiving first battery life information for a first battery installed in an electronic device, receiving user demand information for the electronic device, and based on the first battery life information and the user demand information, adjusting a charging rate for a second battery electrically coupled to the battery charging device for charging, the second battery configured for use in the electronic device.

Abstract: A method for automatic speech recognition using joint acoustic echo cancellation, speech enhancement, and voice separation includes receiving, at a contextual frontend processing model, input speech features corresponding to a target utterance. The method also includes receiving, at the contextual frontend processing model, at least one of a reference audio signal, a contextual noise signal including noise prior to the target utterance, or a speaker embedding including voice characteristics of a target speaker that spoke the target utterance. The method further includes processing, using the contextual frontend processing model, the input speech features and the at least one of the reference audio signal, the contextual noise signal, or the speaker embedding vector to generate enhanced speech features.

Abstract: Techniques and apparatuses are described for an earbud with low enoise. An electronic device includes: a flexible printed circuit (FPC) having first, second, and third regions, the FPC being shaped to form multiple layers including first, second, and third layers, the second region forming the second layer and being between the first layer formed by the first region and the third layer formed by the third region. The electronic device further includes a multi-layer board (MLB) in the first region. A battery connected to the FPC in the first region and disposed between the first and second regions. A speaker connected to the FPC in the second region and disposed between the second and third regions. Charger pins connected to the FPC in the third region. A microphone connected to the FPC in the third region. An infrared sensor connected to the FPC in the third region.

Abstract: A method of detecting and responding to a visitor to a smart home environment via an electronic greeting system of the smart home environment, including determining that a visitor is approaching an entryway of the smart home environment; initiating a facial recognition operation while the visitor is approaching the entryway; initiating an observation window in response to the determination that a visitor is approaching the entryway; obtaining context information from one or more sensors of the smart home environment during the observation window; and at the end of the time window, initiating a response to the detected approach of the visitor based on the context information and/or an outcome of the facial recognition operation.

Abstract: Implementations are provided for operably coupling multiple robot controllers to a single virtual environment, e.g., to generate training examples for training machine learning model(s). In various implementations, a virtual environment may be simulated that includes an interactive object and a plurality of robot avatars that are controlled independently and contemporaneously by a corresponding plurality of robot controllers that are external from the virtual environment. Sensor data generated from a perspective of each robot avatar of the plurality of robot avatars may be provided to a corresponding robot controller. Joint commands that cause actuation of one or more joints of each robot avatar may be received from the corresponding robot controller. Joint(s) of each robot avatar may be actuated pursuant to corresponding joint commands. The actuating may cause two or more of the robot avatars to act upon the interactive object in the virtual environment.

Abstract: This document describes systems and techniques directed at mitigating display diffraction flares for under-display sensing. In aspects, an equation may be derived that models the effects of a display in producing a diffraction phenomenon at an image plane of a sensing region for an under-display light-sensing device. The equation may be used to determine an arrangement (e.g., an optimized arrangement) of components (e.g., sub-pixels) within the display that minimizes a diffraction efficiency for at least one diffraction order and, thereby, mitigates an intensity and/or a prevalence of optical artifacts in light-sensing data. In implementations, an image intensity point-spread-function is utilized to calculate diffraction efficiencies for respective diffraction orders (e.g., the lowest diffraction orders, the diffraction orders with the greatest brightness).

Abstract: An interconnection for flex circuit boards used, for instance, in a quantum computing system are provided. In one example, the interconnection can include a first flex circuit board having a first side and a second side opposite the first side. The interconnection can include a second flex circuit board having a third side and a fourth side opposite the third side. The first flex circuit board and the second flex circuit board are physically coupled together in an overlap joint in which a portion of the second side for the first flex circuit board overlaps a portion of the third side of the flex circuit board. The interconnection can include a signal pad structure positioned in the overlap joint that electrically couples a first via in the first flex circuit board and a second via in the second flex circuit board.

Abstract: Techniques are disclosed that enable clarifying whether a user query corresponds to a candidate intent when an intent score (indicating the probability the user query corresponds to the candidate intent) fails to satisfy a threshold likelihood value but is “close” to satisfying the threshold likelihood value. For example, the intent score can fail to satisfy the threshold likelihood value but can satisfy an additional threshold likelihood value. Various implementations include generating the candidate intent and corresponding intent score by processing a natural language user query using a natural language understanding (NLU) model.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for optimizing the execution of the operations of a neural network. One of the methods includes obtaining data representing a graph characterizing a plurality of operations of a neural network, wherein each node of the graph characterizes an operation of the neural network and each edge of the graph characterizes data dependency between the operations; processing the data representing the graph using a graph embedding neural network to generate an embedding of the graph; and processing the embedding of the graph using a policy neural network to generate a task output, wherein the task output comprises, for each of the plurality of operations of the neural network, a respective decision for a particular optimization task.

Abstract: Implementations disclosed herein are directed to federated learning of machine learning (“ML”) model(s) based on gradient(s) generated at corresponding client devices and a remote system. Processor(s) of the corresponding client devices can process client data generated locally at the corresponding client devices using corresponding on-device ML model(s) to generate corresponding predicted outputs, generate corresponding client gradients based on the corresponding predicted outputs, and transmit the corresponding client gradients to the remote system. Processor(s) of the remote system can process remote data obtained from remote database(s) using global ML model(s) to generate additional corresponding predicted outputs, generate corresponding remote gradients based on the additional corresponding predicted outputs. Further, the remote system can utilize the corresponding client gradients and the corresponding remote gradients to update the global ML model(s) or weights thereof.

Abstract: This document describes techniques for suggesting actions based on machine learning. These techniques determine a task that a user desires to perform, and presents a user interface through which to perform the task. To determine this task, the techniques can analyze content displayed on the user device or analyze contexts of the user and user device. With this determined task, the techniques determine an action that may assist the user in performing the task. This action is further determined to be performable through analysis of functionalities of an application, which may or may not be executing or installed on the user device. With some subset of the application's functionalities determined, the techniques presents the subset of functionalities via the user interface. By so doing, the techniques enable a user to complete a task more easily, quickly, or using fewer computing resources.

Abstract: Implementations set forth herein relate to pre-emptively initializing an automated assistant in a vehicle according to certain indications, in order to reduce latency while also seeking to preserve computational resources. In some implementations, data for effectuating one or more features of an automated assistant can be loaded into memory of a computing device based on vehicle interaction data. For example, the vehicle interaction data can characterize instances in which the user, from within their vehicle, invoked the automated assistant within a threshold period of time of an application completing an operation. Based on the vehicle interaction data, subsequent instances of the operation being completed while the user is in the vehicle can cause data to be loaded into memory in order to pre-emptively prepare the automated assistant to be utilized by the user.

Abstract: A method, apparatus, and computer-readable medium for determining a tolerance specification of a clock synchronization between a touch display device (TDD) and a stylus. In the method, an accumulated timing error between the TDD and the stylus is determined based on crystal inaccuracies of the TDD and the stylus, connection interval of Bluetooth communication between the TDD and the stylus, and a number of consecutive failed transmissions of the Bluetooth communication. A total processing delay is determined based on a first processing delay and a second processing delay. The first processing delay is a time delay of sending a synchronization signal from touch driving circuitry of the TDD to Bluetooth circuitry of the TDD. The second processing delay is a time delay of sending a Bluetooth packet from the TDD to the stylus. The tolerance specification is determined based on the accumulated timing error and the total processing delay.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium. In one aspect, a method includes the actions of receiving a request to perform computations for a neural network on a hardware circuit having a matrix computation unit, the request specifying a transpose operation to be performed on a first neural network matrix; and generating instructions that when executed by the hardware circuit cause the hardware circuit to transpose the first neural network matrix by performing first operations, wherein the first operations include repeatedly performing the following second operations: for a current subdivision of the first neural network matrix that divides the first neural network matrix into one or more current submatrices, updating the first neural network matrix by swapping an upper right quadrant and a lower left quadrant of each current submatrix, and subdividing each current submatrix into respective new submatrices to update the current subdivision.