Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training neural networks through contrastive learning. In particular, the contrastive learning is modified to use a relative margin to adjust a training pair's contribution to optimization.

Abstract: Methods, systems and apparatus for determining an error-mitigated expectation value of a target observable with respect to a noisy quantum state. In one aspect a method includes obtaining multiple copies of the noisy quantum state; performing measurements on tensor products of M copies of the noisy quantum state to compute an expectation value of the target observable with respect to an entangled quantum state, wherein M?1 and eigenvalues corresponding to non-dominant eigenvectors of the noisy quantum state in the spectral decomposition of the entangled quantum state are suppressed exponentially in M; and using the computed expectation value of the target observable with respect to an entangled quantum state to determine the error-mitigated expectation value of the target observable with respect to the noisy quantum state.

Abstract: Speech processing techniques are disclosed that enable determining a text representation of alphanumeric sequences in captured audio data. Various implementations include determining a contextual biasing finite state transducer (FST) based on contextual information corresponding to the captured audio data. Additional or alternative implementations include modifying probabilities of one or more candidate recognitions of the alphanumeric sequence using the contextual biasing FST.

Abstract: Implementations are directed to providing a voice bot development platform that enables a third-party developer to train a voice bot based on training instance(s). The training instance(s) can each include training input and training output. The training input can include a portion of a corresponding conversation and a prior context of the corresponding conversation. The training output can include a corresponding ground truth response to the portion of the corresponding conversation. Subsequent to training, the voice bot can be deployed for conducting conversations on behalf of a third-party. In some implementations, the voice bot is further trained based on a corresponding feature emphasis input that attentions the voice bot to a particular feature of the portion of the corresponding conversation. In some additional or alternative implementations, the voice bot is further trained to interact with third-party system(s) via remote procedure calls (RPCs).

Abstract: Systems and methods are provided for identifying and rendering content relevant to a user's current mental state and context. In an aspect, a system includes a state component that determines a state of a user during a current session of the user with the media system based on navigation of the media system by the user during the current session, media items provided by the media system that are played for watching by the user during the current session, and a manner via which the user interacts with or reacts to the played media items. In an aspect, the state of the user includes a mood of the user. A selection component then selects a media item provided by the media provider based on the state of the user, and a rendering component effectuates rendering of the media item to the user during the current session.

Abstract: The various implementations described herein include methods, devices, and systems for interactive cloud gaming. In one aspect, an interface device with processors and memory storing an application program is used for cloud gaming. The interface device includes a communication module for receiving and/or transmitting messages and media streams between the interface device, a game controller, and a remote gaming server. The interface device further includes an output module configured to provide output of the application to an input of a display device coupled to or integrated with the interface device. The application is configured to implement a first mode in which the interface device processes and displays outputs associated with substantially all messages transmitted by the game controller, and a second mode in which the interface device processes and displays a predefined subset of messages from the game controller and displays gaming media streams from the gaming server.

Abstract: A method for facilitating a Parkinson's Disease (“PD”) assessment of a patient includes capturing first video of a patient performing first test movements while holding the mobile device; capturing second video of the patient performing second test movements while maintaining the mobile device on their person; capturing third video of the patient performing third test movements including standing and walking; capturing one or more IMU readings using an IMU of the mobile device; processing the first video, the second video, and the third video according to (i) a hand landmark model to generate one or more hand biomarkers, (ii) a face landmark model to generate one or more face biomarkers, and (iii) a body landmark model to generate one or more body biomarkers; and determining an assessment score based on a standardized PD assessment by processing the biomarkers.

Abstract: A UE communicating in DC with an MN and an SN receives a configuration related to a C-SN and at least one condition for connecting to the C-SN, determines that a radio connection between the UE and the MN has failed, and transmits a first indication of the failed radio connection to the SN to initiate a recovery procedure. The UE also releases the configuration and/or the at least one condition, or suspending a procedure, for connecting to the C-SN\.

Abstract: A system and method are disclosed for identifying a plurality of past videos posted via a first group of channels, determining, for each channel of a second group, an average number of access requests that a future video to be posted via a respective channel is predicted to receive within a first time interval after posting, determining, for each channel of the second group, a plurality of adjusted average numbers of access requests that the future video to be posted via the respective channel of the second group is predicted to receive within the first time interval, wherein each of the plurality of adjusted average numbers of access requests is determined using an adjustment factor reflecting a number of access requests received by a past video posted via a respective channel of the first group, and determining whether the second group of channels satisfies a predetermined access criterion.

Abstract: Aspects of the technology evaluate a client device query to identify an IP address-based estimated location of a mobile device. One or more candidate wireless stations are selected in accordance with the IP-based estimated location and a location-relevant response to the query is also generated in accordance with the IP-based location. This information is transmitted to the mobile device. The mobile device compares the received information about the candidate wireless stations against one or more actual wireless stations with which the mobile device is in communication. Log information is generated as a result of the comparison. The log information is anonymized and transmitted to the network without user ID or other client-identifiable information included in the log. The received log information is then used to validate the initial estimated location. Aggregation of log information from multiple devices may be used in the validation process.

Abstract: A method includes receiving a text embedding generation query from a user requesting generation of a text embedding for one or more data elements stored at a data warehouse. In response, the method includes selecting, using the text embedding generation query, a text embedding model from a plurality of different text embedding models. The method includes generating, using the selected text embedding model, the text embedding for the one or more data elements and storing the text embeddings at the data warehouse. The method includes receiving a machine learning model training query from the user device requesting training of a machine learning model using the text embeddings. In response to receiving the machine learning model training query, the method includes training the machine learning model using the text embeddings. The method includes providing, to the user device, a notification indicating that training of the machine learning model is complete.

Abstract: Adaptive howling suppression is provided for active noise control (ANC) systems of wearable audio components (WACs), such as earbuds, based on detecting whether the WACs are presently being worn. Upon first detecting howling, embodiments can pre-suppress the howling audio signature by reducing ANC output. Then, embodiments detect whether the WAC is presently on-ear (worn) or off-ear and attempt to fully suppress the howling using different parameters, based on this detection. For example, if on-ear, embodiments use on-ear tuning settings to cycle through a set of suppression stages until ANC output can be restored without howling; if off-ear, embodiments use off-ear tuning settings to cycle through the set of suppression stages until ANC output can be restored without howling. Once ANC output can be restored without howling, the system can be released to a normal idle state.

Abstract: Techniques and apparatuses are described that implement random modulation of charge-pump noise phases to reduce structured noise induced by the charge pump. In an example aspect, a correlated double sampling (CDS) circuit is coupled to a pixel array including at least one pixel circuit. The CDS circuit receives an input signal generated by the at least one pixel circuit from the pixel array and samples a reset component of the input signal during a first sampling time to generate a reset component sample. The first sampling time is at a first offset from a reset control signal and prior to a settling time of the at least one pixel circuit. The CDS circuit samples a signal component of the input signal during a second sampling time to generate a signal component sample and determines an output signal based on the reset component sample and the signal component sample.

Abstract: This document describes techniques directed to active thermal-control of a floodlight and associated floodlights. As described, an example floodlight includes a first heat-transfer subsystem that uses a fully enclosed heat sink to transfer heat from an array of LEDs to a first housing component of the floodlight. The floodlight further includes a second heat-transfer subsystem to transfer heat from one or more PSUs to a second housing component of the floodlight. Described techniques include using thermistors located throughout the floodlight to actively monitor a temperature profile within the floodlight and, if one or more operating-temperature thresholds are violated, reducing power consumption within the floodlight.

Abstract: The present disclosure contemplates a method for synchronizing a large number of generators on an AC bus by synchronizing each generator's output to a nominal output that is generated from a common external source. For example, each generator, using a high speed communication signal, can synchronize to a nominal output provided by a master generator, or centralized command module. In another example, each generator can generate its own nominal output referenced to a common external time signal, such as a global positioning system (GPS) signal, or other reference. By synchronizing independently of bus voltage and frequency, the generators are able to synchronize in parallel, instead of serially.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating a respective neural network output for each of a plurality of inputs, the method comprising, for each of the neural network layers: receiving a plurality of inputs to be processed at the neural network layer; forming one or more batches of inputs from the plurality of inputs, each batch having a number of inputs up to the respective batch size for the neural network layer; selecting a number of the one or more batches of inputs to process, where a count of the inputs in the number of the one or more batches is greater than or equal to the respective associated batch size of a subsequent layer in the sequence; and processing the number of the one or more batches of inputs to generate the respective neural network layer output.

Abstract: A method for automatic hotword threshold tuning includes receiving, from a user device executing a first stage hotword detector configured to detect a hotword in streaming audio, audio data characterizing the detected hotword. The method includes processing, using a second stage hotword detector, the audio data to determine whether the hotword is detected by the second stage hotword detector. When the hotword is not detected, the method includes identifying a false acceptance instance at the first stage hotword detector indicating that the first stage hotword detector incorrectly detected the hotword. The method includes determining whether a false acceptance rate satisfies a false acceptance rate threshold based on a number of false acceptance instances within a false acceptance time period. When the false acceptance rate satisfies the false acceptance rate threshold, the method includes adjusting the hotword detection threshold of the first stage hotword detector.

Abstract: A computer-implemented method includes setting, by a computing device, a maximum quantization parameter (QP) value for encoding an input video frame to a value which is the maximum of: a first QP value corresponding to a first proportion of an application-specified maximum QP value, or a second QP value determined based on a value which is the minimum of: a third QP value determined based on an average value of QP values used to encode a plurality of video frames before the input video frame, or a fourth QP value corresponding to a second proportion of the application-specified maximum QP value. The computer-implemented method further includes using the set maximum QP value as a quality bound for encoding the input video frame.

Abstract: Systems and method for the structure, configuration, or arrangement of the input fields in a user interface. For example, a computer-implemented method includes displaying a structured input interface configured with a plurality of input fields. The structured input interface can be configured for rendering in a graphical user interface. A first input field can correspond to an initial notification element in the graphical user interface for rendering in association with the first input field. The method includes receiving user data. The method includes determining, based on the user data, an updated notification element for rendering in association with the first input field. The method includes updating the structured input interface comprising the first input field and the updated notification element.

Abstract: Systems and methods are provided for creating a cinematic storytelling experience. A user selection of a story is received from a client device. User input to initiate verbal recitation of the selected story is received. Correspondence between the verbal recitation and at least one keyword is detected, and the at least one corresponding action is performed.