Abstract: A method includes executing an initial instance of a change log process for a distributed system, each instance of the change log process configured to store a transaction history of transactions executed on the distributed system. The method also includes receiving transaction requests for executing corresponding transactions on the distributed system and determining a change log load based on the received transaction requests. The method includes executing at least one subsequent instance of the change log process when the change log load satisfied a threshold load. When multiple instances of the change log process are executing, the method includes ceasing execution of the at least one subsequent instance of the change log process and merging the transaction history of the initial instance of the change log process and the transaction history of the at least one subsequent instance of the change log process.

Abstract: Techniques and apparatuses are described for performing proactive thermal throttling based on forecasted temperature data. In example aspects, a thermal control system controls an internal temperature of a computing device to protect the computing device from being damaged due to overheating, to maintain reliability of the computing device, and to avoid creating a potentially unsafe situation for the user to operate the computing device. The thermal control system performs temperature forecasting to generate forecasted temperature data based at least on measured temperature data. The thermal control system performs proactive thermal throttling by comparing the forecasted temperature data to a target temperature. If the forecasted temperature data meets or exceeds the target temperature, the thermal control system activates one or more cooling techniques to maintain the temperature of the computing device below a temperature limit, such as at the target temperature.

Abstract: A method (500) includes receiving audio data 102 characterizing an utterance (101) including a particular phrase (182) and a subsequent sequence of individual characters (184) that spell the particular phrase. The method also includes processing, using a speech recognition model (142), the audio data to generate an initial transcription (107) including a misrecognition of the particular phrase. The method also includes, in response to detecting a spelling structure (180) in the initial transcription subsequent to the misrecognition of the particular phrase: extracting, from the initial transcription, the sequence of characters; constructing a corrected phrase for the misrecognition of the particular phrase from the extracted sequence of characters; and normalizing the transcription to obtain a final transcription (170) for the utterance by replacing the extracted misrecognition of the particular phrase with the corrected phrase constructed from the extracted sequence of individual characters.

Abstract: Techniques and devices for continual in-line learning for radar-based gesture recognition are described in this document. Through continual in-line learning, a computing device may improve recognition of even the hardest-to-recognize gestures by gradually storing characteristics of ambiguous gestures performed by a user. Specifically, a radar system may detect a first ambiguous gesture that the computing device fails to recognize as a known gesture and a second gesture that the computing device successfully recognizes as the known gesture. The computing device may identify a similarity between the first and the second gesture, and in doing so, store a characteristic of the first gesture to recognize the known gesture more-accurately in a future occurrence.

Abstract: A method includes obtaining training data that includes a plurality of training utterances each paired with a corresponding target output sequence. Each corresponding target output sequence includes a sequence of words and associated timestamps. Each timestamp indicates a timing associated with one of the words in the sequence of words of the target output sequence. For each respective training utterance, the method includes generating a corresponding audio encoding based on the respective training utterance using an audio encoder, generating a respective speech recognition result based on the corresponding audio encoding using a decoder, and determining a corresponding loss based on the corresponding target output sequence and the respective speech recognition result. The method includes training a speech recognition model based on the corresponding loss determined for each respective training utterance of the plurality of training utterances.

Abstract: An ASR model includes a first encoder configured to receive a sequence of acoustic frames and generate a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames. The ASR model also includes a second encoder configured to receive the first higher order feature representation generated by the first encoder at each of the plurality of output steps and generate a second higher order feature representation for a corresponding first higher order feature frame. The ASR model also includes a decoder configured to receive the second higher order feature representation generated by the second encoder at each of the plurality of output steps and generate a first probability distribution over possible speech recognition hypothesis. The ASR model also includes a language model configured to receive the first probability distribution over possible speech hypothesis and generate a rescored probability distribution.

Abstract: This document describes power supplies for compute cores. In one aspect, a power supply system for a compute core includes a primary power converter configured to provide and regulate direct current (DC) power to the compute core over a power rail that electrically couples an output of the primary power converter to the compute core. The power supply system also includes a transient suppressor circuit coupled to the power rail and configured to suppress transient voltage differences between a target supply voltage for the compute core and an actual supply voltage to the compute core.

Abstract: A method includes, for each training sample of a plurality of training samples, processing, using an RNN-T model, a corresponding sequence of acoustic frames to obtain an n-best list of speech recognition hypotheses, and, for each speech recognition hypothesis of the n-best list, determining a corresponding number of word errors relative to a corresponding ground-truth transcription. For a top-ranked hypothesis from the n-best list, the method includes determining a first loss based on the corresponding ground-truth transcription. The method includes identifying, as an oracle hypothesis, the speech recognition hypothesis from the n-best list having the smallest corresponding number of word errors relative to the corresponding ground-truth transcription, and determining a second loss for the oracle hypothesis based on the corresponding ground-truth transcription.

Abstract: An electronic device includes a radar input control system that can perform presence detection, as well as distinguish between touch input, gesture input, and/or potential docking devices. In implementations, the electronic device includes a plurality of radar reception antennas and at least one transmit channel that feeds a plurality of radar transmission antennas. One or more radar reception antennas of the plurality of radar reception antennas may correspond to a discrete input location on a surface of the electronic device. In this way, the electronic device can be implemented without mechanical switches or a touchscreen interface and still receive user input.

Abstract: Implementations described herein relate to using self-evaluation when utilizing a large language model (LLM) to generate a response to a natural language (NL) based input. The LLM can be used to process the NL based input to generate a plurality of responses, and to generate a critique of those responses by comparing the responses to a set of response evaluation criteria. One of the responses can then be selected based on the comparison with the set of response evaluation criteria which can vary from one NL based input to another. If the NL based input was obtained a user of a client device during an inference stage, then the selected response can be rendered for presentation to the user. If the NL based input was obtained during a training stage, then the selected response can be stored as a training instance and optionally in association with additional data.

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: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing images using self-attention based neural networks. One of the methods includes obtaining one or more images comprising a plurality of pixels; determining, for each image of the one or more images, a plurality of image patches of the image, wherein each image patch comprises a different subset of the pixels of the image; processing, for each image of the one or more images, the corresponding plurality of image patches to generate an input sequence comprising a respective input element at each of a plurality of input positions, wherein a plurality of the input elements correspond to respective different image patches; and processing the input sequences using a neural network to generate a network output that characterizes the one or more images, wherein the neural network comprises one or more self-attention neural network layers.

Abstract: Techniques and apparatuses are described for adaptive phase-changing device power-saving operations. In aspects, a base station determines to transition an adaptive phase-changing device (APD) into an enabled APD-PS mode and determines an APD-PS configuration for the APD that specifies a framework for operating in the enabled APD-PS mode. The base station then directs the APD to operate in the enabled APD-PS mode by communicating the APD-PS configuration to the APD and transmits or receives wireless signals using a surface of the APD and based on the APD-PS configuration.

Abstract: A method for predicting color variance using a proxy includes generating a first 3D object proxy based on a stored 3D object, generating a second 3D object proxy based on the stored 3D object, transforming the first 3D object proxy based on a 3D object identified in a frame of a video, transforming the second 3D object proxy based on the 3D object identified in a key frame of the video, mapping color attributes from the 3D object identified in the frame of the video to the transformed first 3D object proxy, mapping color attributes from the 3D object identified in the key frame to the transformed second 3D object proxy, and generating color data for the 3D object based on the color attributes for the transformed first 3D object proxy and the color attributes for the transformed second 3D object proxy.

Abstract: Implementations set forth herein relate to an automated assistant that can operate according various driving-optimized modes depending on a degree of confidence that a user is predicted to be traveling in a vehicle. For instance, the automated assistant can automatically operate according to a driving-optimized mode when a prediction that the user is traveling corresponds to a certain degree of confidence. Alternatively, when a prediction that a user is traveling corresponds to lesser degree of confidence, the automated assistant may not operate according to the driving-optimized mode until the user expressly selects to transition the automated assistant into the driving-optimized mode. When the user selects the driving-optimized mode, a driving mode GUI can be rendered with a navigation interface that may include directions to a predicted destination of the user and/or another interface with content suggestions for the user.

Abstract: Adapting an automated assistant based on detecting: movement of a mouth of a user; and/or that a gaze of the user is directed at an assistant device that provides an automated assistant interface (graphical and/or audible) of the automated assistant. The detecting of the mouth movement and/or the directed gaze can be based on processing of vision data from one or more vision components associated with the assistant device, such as a camera incorporated in the assistant device. The mouth movement that is detected can be movement that is indicative of a user (to whom the mouth belongs) speaking.

Abstract: A method for determining a placement for machine learning model operations across multiple hardware devices is described.

Abstract: Among other things, an apparatus comprises quantum units; and couplers among the quantum units. Each coupler is configured to couple a pair of quantum units according to a quantum Hamiltonian characterizing the quantum units and the couplers. The quantum Hamiltonian includes quantum annealer Hamiltonian and a quantum governor Hamiltonian. The quantum annealer Hamiltonian includes information bearing degrees of freedom. The quantum governor Hamiltonian includes non-information bearing degrees of freedom that are engineered to steer the dissipative dynamics of information bearing degrees of freedom.

Abstract: According to an aspect, a method of restoring a computing session includes receiving, over a network, session data from a server computer, where the session data includes information about at least one session item that is active during a computing session of a first computing device. The at least one session item includes at least one of a web application or a native application. The method includes restoring the at least one session item of the computing session on a second computing device based on the session data, where the at least one session item is arranged on a user interface of the second computing device according to a display arrangement that corresponds to a display arrangement of the at least one session item on a user interface of the first computing device.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer-storage media, for privacy preserving digital component provider. In some implementations, a method includes accessing a buffer including one or more sets of bits; generating a random sequence of values; generating, from the random sequence of values, a sequence of indices representing an order in which to access particular sets of bits of the buffer; in response to determining an index of the sequence of indices corresponds to a location in the buffer, accessing a set of the particular sets of bits of the buffer at the index in the order of the sequence of indices; and performing one or more memory operations on the set of the one or more sets of bits after accessing the set.

Abstract: Systems and methods can perform self-supervised machine learning for improved medical image analysis. As one example, self-supervised learning on ImageNet, followed by additional self-supervised learning on unlabeled medical images from the target domain of interest, followed by fine-tuning on labeled medical images from the target domain significantly improves the accuracy of medical image classifiers such as, for example diagnostic models. Another example aspect of the present disclosure is directed to a novel Multi-Instance Contrastive Learning (MICLe) method that uses multiple different medical images that share one or more attributes (e.g., multiple images that depict the same underlying pathology and/or the same patient) to construct more informative positive pairs for self-supervised learning.

Abstract: A corpus of textual data is generated with a machine-learned text generation model. The corpus of textual data includes a plurality of sentences. Each sentence is descriptive of a type of audio. For each of a plurality of audio recordings, the audio recording is processed with a machine-learned audio classification model to obtain training data including the audio recording and one or more sentences of the plurality of sentences closest to the audio recording within a joint audio-text embedding space of the machine-learned audio classification model. The sentence(s) are processed with a machine-learned generation model to obtain an intermediate representation of the one or more sentences. The intermediate representation is processed with a machine-learned cascaded diffusion model to obtain audio data. The machine-learned cascaded diffusion model is trained based on a difference between the audio data and the audio recording.

Abstract: A user device (UE) equipped with a first universal subscriber identity module (USIM) for communicating with a first cellular network and a second USIM for communicating with a second cellular network, wherein the first cellular network includes a first radio access network (RAN) and the second cellular network includes a second RAN, determines whether the first cellular network supports access to a gateway of the second cellular network (402), and when the first cellular network supports the access: (i) connects to the second cellular network via the first RAN and the gateway (408), and (ii) prevents the second cellular network from transmitting information to the UE via the second RAN (410).

Abstract: Aspects of the disclosure are directed to network optimization of various workload servers running in a distributed cloud platform through closed loop machine learning inferencing performed locally on the workload servers. The workload servers can each be equipped with one or more machine learning accelerators to respectively perform local predictions for the workload servers. In response to the local predictions, attributes of the workload servers can be adjusted automatically for optimizing the network.

Abstract: In a cloud computing system, groups of virtual machines include within one or more pods interacting with one another. A set of hosts is dedicated to operation of the virtual machines in each pod, and such dedication persists during routine maintenance operations when one or more virtual machines is uninstalled. The pod and the hosts can be scheduled for maintenance during a single interval. The cloud computing system may notify the owner of the virtual machines in advance of the interval, and may allow the owner of the virtual machines to either reschedule the maintenance of the pod and hosts in a single interval, or to schedule maintenance for each virtual machine in a separate interval.

Abstract: Provided are systems and methods for sequence-to-sequence modeling with neural quality metrics. More particularly, example aspects of the present disclosure relate to minimum bayes risk (MBR) decoding with neural metrics for machine translation. According to example aspects of the present disclosure, a set of candidate outputs can be sampled from a machine translation model given a source sequence. Given the set of candidate outputs, systems and methods according to example aspects of the present disclosure can select a hypothesis with high expected utility with respect to the distribution over a set of pseudo-references from the machine translation model.

Abstract: Provided are systems and methods for privacy-preserving learning and analytics of a shared embedding space for data split across multiple separate data silos. A central computing system can generate a plurality of synthetic data examples having respective feature data within an aggregate feature-space that represents an aggregation of different component feature-spaces associated with the multiple separate data silos. The synthetic data examples can be used by different computing systems associated with the data silos to generate embeddings within a shared embedding space. Once the embeddings have been generated in the shared embedding space, multiple different types of analytics can be performed on the shared embedding space. As one example, the multiple data silos can correspond to multiple separate entity domains and an analysis of embeddings generated in the shared embedding space can be used to facilitate identification or classification of malicious actors across the multiple separate entity domains.

Abstract: A system includes multiple sensors and, for each sensor, a respective sensor controller of multiple sensor controllers. Each sensor controller is configured to implement a respective decimation filter that is configured to generate a single output value from multiple input samples generated by a corresponding sensor of the multiple sensors. The system further includes a master sensor controller of the multiple sensor controllers, which is configured to generate a sync signal upon receiving a threshold number of input samples. Each sensor controller other than the master sensor controller is configured to monitor sync signals generated by the master sensor controller and to provide an output value generated from input samples upon determining that the master sensor controller generated a sync signal.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, that identify and issue search queries expected to be issued in the future. A set of search queries that have been issued by multiple user devices can be obtained. For each query instance, contextual data can be obtained. A first query and its contextual data can be input to a model that outputs the query's likelihood of being issued in the future. The model can be trained using contextual data for training queries and a corresponding labels for the training queries. The learning model outputs the first query's likelihood of being issued in future, and this query is stored as a repeatable query if the likelihood satisfying a repeatability threshold. Subsequently, a stored repeatable query is issued upon a selection of a user selectable interface component and the search engine provides search results for the query.

Abstract: Provided are improved end-to-end self-supervised pre-training frameworks that leverage a combination of contrastive and masked modeling loss terms. In particular, the present disclosure provides framework that combines contrastive learning and masked modeling, where the former trains the model to discretize input data (e.g., continuous signals such as continuous speech signals) into a finite set of discriminative tokens, and the latter trains the model to learn contextualized representations via solving a masked prediction task consuming the discretized tokens. In contrast to certain existing masked modeling-based pre-training frameworks which rely on an iterative re-clustering and re-training process or other existing frameworks which concatenate two separately trained modules, the proposed framework can enable a model to be optimized in an end-to-end fashion by solving the two self-supervised tasks (the contrastive task and masked modeling) simultaneously.

Abstract: A method and apparatus for distribution of applications amongst a number of servers, ensuring that changes to application data on a master for that application are asynchronously replicated to a number of slaves for that application. Servers may be located in geographically diverse locations; the invention permits data replication over high-latency and lossy network connections and failure-tolerance under hardware and network failure conditions. Access to applications is mediated by a distributed protocol handler which allows any request for any application to be addressed to any server, and which, when working in tandem with the replication system, pauses connections momentarily to allow seamless, consistent live-migration of applications and their state between servers.

Abstract: Reference voltage generators provide stable reference voltages to facilitate correct circuit operation. Bandgap voltage generators are becoming less feasible as process nodes shrink. In example implementations, to accommodate smaller devices, an apparatus for generating a reference voltage includes a current generator having at least one current branch that receives a supply voltage, with the current branch including at least one cascode transistor that receives a bias voltage value. The apparatus also includes a first voltage generator that receives a first bias current from the current generator and provides a first voltage value being complementary to absolute temperature. The apparatus further includes a second voltage generator that receives a second bias current from the current generator and provides a second voltage value being proportional to absolute temperature.

Abstract: A multimedia communication system and computer-implemented method for transmitting auxiliary display content to an end-user communication device to be rendered on a display device with a special effect to emphasize an image included in the auxiliary display content, comprising analyzing image data included in an auxiliary display content to detect an object image or a background image, determining a special effect based on the analysis of the image data, applying the special effect to the auxiliary display content to modify display properties for the auxiliary display content such that the object image is emphasized or pops out, and sending the auxiliary display content with modified display properties to the end-user communication device. The special effect can comprise a non-customization special effect, a simple foreground special effect or a selective foreground special effect.

Abstract: An antenna system including multiple range antennas and multiple angle-of-arrival antennas for electronic devices is provided. In some embodiments, the antenna system can include a plurality of substantially orthogonal range antennas and a plurality of angle-of-arrival antennas, where a processing system on an initiator device can determine which of the plurality of substantially orthogonal range antennas to select for subsequent angle-of-arrival measurements. This can, for example, reduce polarization mismatch between the antenna systems of initiator devices and responder devices.

Abstract: In a general aspect, a method for growing an InGaN optoelectronic in a reaction chamber, by MOCVD, includes controlling a surface temperature of a wafer to be at least 750° C. during growth of a light-emitting layer. The light emitting layer includes an InGaN quantum well layer having an In % of greater than 25%. The method further includes providing an indium-containing metalorganic precursor and a gallium-containing metalorganic precursor into the reaction chamber and to the wafer during growth of the light-emitting layer when the surface temperature of the wafer is greater than 750° C. The method also includes providing an N-containing species to the wafer at a rate such that a partial pressure of the N-containing species at the surface of the wafer is greater than 1.5 atmospheres during growth of the light-emitting layer of the optoelectronic device when the surface temperature of the wafer is greater than 750° C.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a surrogate neural network configured to determine a predicted performance measure of a hardware accelerator having a target hardware configuration on a target application. The trained instance of the surrogate neural network can be used, in addition to or in place of hardware simulation, during a search process for determining hardware configurations for application-specific hardware accelerators, i.e., hardware accelerators on which one or more neural networks can be deployed to perform one or more target machine learning tasks.

Abstract: Various implementations include reducing latency when interacting with a generative model system based on generating predicted complete text based on natural language (NL) text input, where the NL text input is a portion of a user query. In many implementations, predicted completion text can be generated by processing NL text input using a language model. In several implementations, the system can perform initial processing of the NL input text and the predicted completion text (e.g., preform initial preprocessing of the NL input text and predicted completion text for processing using the generative model, performing an initial limited decoding of output using the generative model, etc.). The user can confirm the predicted completion text before the system continues processing the NL input and predicted completion text using the generative model to generate output.

Abstract: Processor(s) of a client device can receive global weights of a global ML model from a remote system, obtain a client device data set, determine a Fisher information matrix for the client data set, and transmit the Fisher information matrix for the client data set to the remote system. Further, processor(s) of the remote system can determine a corresponding elastic weight consolidation (EWC) loss term for each of the global weights based on at least the Fisher information matrix, generate a server update for the global ML model based on (i) processing server data remotely at the remote system and using the global ML model and (ii) based on the corresponding EWC loss term for each of the global weights, and update the global weights of the global ML model based on the server update.

Abstract: A system including a multi-task adapter neural network for performing multiple machine learning tasks is described. The adapter neural network is configured to receive a shared input for the machine learning tasks, and process the shared input to generate, for each of the machine learning tasks, a respective predicted output. The adapter neural network includes (i) a shared encoder configured to receive the shared input and to process the shared input to extract shared feature representations for the machine learning tasks, and (ii) multiple task-adapter encoders, each of the task-adapter encoders being associated with a respective machine learning task in the machine learning tasks and configured to: receive the shared input, receive the shared feature representations from the shared encoder, and process the shared input and the shared feature representations to generate the respective predicted output for the respective machine learning task.

Abstract: An example method includes receiving electrical power via a physical connector of a mobile computing device; determining, by a controller of the mobile computing device, a value of a temperature of a virtual temperature sensor, the value representing a difference between a temperature measured by a first temperature sensor of a plurality of temperature sensors of the mobile computing device and a temperature measured by a second temperature sensor of the plurality of temperature sensors, wherein a distance between the first temperature sensor and the physical connector is less than a distance between the second temperature sensor and the physical connector; determining, based on the value of the temperature of the virtual temperature sensor, whether an overheat event has occurred; and responsive to determining that the overheat event has occurred, reducing an amount of electrical current flowing through the physical connector.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an estimate for a number of users that accessed a digital resource within a time window. In one aspect, a method comprises: obtaining access data for a digital resource; generating a tree model based on the access data; selecting, for each node in the tree model, a respective private access value for the node that: (i) is an approximation of an access value for the node, and (ii) is selected from a finite set of possible private access values; and generating an estimate for the number of users that accessed the digital resource at least the predefined number of times within the time window based on private access values associated with one or more nodes in the tree model.

Abstract: A radio access network (RAN) can perform a method for managing security protection for multicast and/or broadcast services (MBS). The method includes receiving (802), from a core network (CN) via a DL tunnel, a data packet associated with an MBS session; prior to transmitting the data packet to a plurality of UEs, determining (804) which security protection to apply to the data packet based on an identity of the DL tunnel; and transmitting (806) the data packet to the plurality of UEs using the determined security protection. A UE can receive (902), from a RAN, a configuration for establishing a logical channel with the RAN; receive (904), from the RAN via the logical channel, a data packet associated with an MBS session; determine (906) which security protection to apply to the data packet based on the configuration; and apply (908) the determined security protection to the data packet.

Abstract: This document describes systems and techniques for using secure MPC to select digital components in ways that preserve user privacy and protects the security of data of each party that is involved in the selection process. In one aspect, a method includes obtaining, by a first computer of a secure multi-party computation (MPC) system, at least a first share of a set of contextual properties of an environment in which a selected digital component will be displayed at a client device. For each digital component in a set of digital components, at least a first share of an eligibility expression that defines a relationship between a set of eligibility criteria for the digital component is obtained. A determination is made, based on the at least first share of the set of contextual properties and the at least first share of the eligibility expression, a first share of an eligibility parameter.

Abstract: An example folding device includes a hinge assembly that is coplanar with the continuous display of the device in order to decrease the thickness of the device. The hinge assembly includes torque members that increase the amount of force needed to rotate the assemblies. In this way, the torque members may provide the device with a more rigid feel. Also in this way, the torque members may enable the device to hold intermediate positions between fully open and fully closed.

Abstract: A computing system may receive an indication of a gaming application and an indication of a computing device. The computing system may determine, using collaborative filtering of performance scores of a plurality of gaming applications executing at a plurality of computing devices, a predicted performance of the gaming application executing at the computing device. The computing system may send an indication of the predicted performance of the gaming application executing at the computing device.

Abstract: Techniques and apparatuses are described that implement software-defined hardware. In example aspects, a hardware manager can enable and disable different combinations of peripherals coupled to a host device. Software can be executed on different enabled combinations of peripherals to determine which combinations best meet the requirements of a product. Combinations of peripherals can be updated with alternative or new peripherals that have new features or replace obsolete peripherals. Thus, the hardware manager can provide a modular validation platform that allows a manufacturer of electronic devices to skip a prototype phase, qualify new peripherals as replacements for existing peripherals that may become obsolete, and iterate between different combinations of peripherals to determine ideal combinations. The disclosed techniques and systems can therefore reduce resources, time, and costs associated with design and testing phases of electronic devices.

Abstract: Example embodiments of aspects of the present disclosure provide an example computer-implemented method for improved prompting of a machine-learned model. The example method can include obtaining an instructive sequence descriptive of an instructive query, an instructive response, and an instructive trace of intermediate states from the instructive query to the instructive response. The example method can include inputting, to a machine-learned model, the instructive sequence and an operative query, wherein the machine-learned model is configured to process the operative query with attention over the instructive sequence. The example method can include generating, using the machine-learned model and responsive to the operative query, an operative response.

Abstract: A method is presented, including providing an offset magnetic flux bias to a plurality of superconducting qubits and providing respective control magnetic flux biases, for performing a computation, to the plurality of qubits using a plurality of control lines coupled respectively to each qubit. The qubits are configured such that respective resonance frequencies of the qubits are controlled by the offset magnetic flux bias and the respective control magnetic flux biases. The qubits are arranged to perform the computation when the respective resonance frequencies of the qubits are within an operational dynamic range.

Abstract: Decoding a video stream may include decoding a first block of a current frame by decoding a first motion vector from the encoded video stream, decoding an identifier of a first interpolation filter from the encoded video stream, and reconstructing the first block using the first motion vector and the first interpolation filter. Decoding a second block of the current frame may include identifying the first motion vector from the first block as a selected motion vector for predicting the second block in response to decoding an inter-prediction mode identifier for decoding the second block, identifying the first interpolation filter as a selected interpolation filter for predicting the second block in response to identifying the first motion vector from the first block as the selected motion vector for predicting the second block, and reconstructing the second block using the first motion vector and the first interpolation filter.