Abstract: A method includes obtaining a pre-trained machine learning model and a training embedding snapshot from a remote system, and obtaining one or more input data samples captured by a user device. The method includes, for each particular input data sample of the one or more input data samples, processing, using an on-device machine learning model corresponding to the pre-trained machine learning model, the particular input data sample to generate a corresponding on-device embedding and one or more corresponding predicted outputs, and generating, using the training embedding snapshot and the corresponding on-device embedding, corresponding performance data. The method includes aggregating the corresponding performance data for the one or more input data samples to determine one or more performance metrics for the on-device machine learning model, and transmitting the one or more performance metrics to the remote system.

Abstract: This document describes systems and techniques directed at zonal attenuation compensation. In aspects, a system includes a graphics processing unit configured to provide image data to a display panel. A zonal attenuation module is configured to combine a zonal attenuation mask with the image data to generate masked image data, the masked image data having a reduced brightness for portions of data corresponding to one or more regions on the display panel based on the zonal attenuation mask. An inverse zonal attenuation module is configured to apply an inverse zonal attenuation mask to the masked image data to reduce a brightness for additional portions of data corresponding to one or more additional regions on the display panel effective to offset increased brightness in the one or more additional regions on the display panel.

Abstract: A method including receiving, from a developer device, a request to build an execution environment for a software application, the software application comprising a manifest of dependencies. The method also includes generating, using a bootstrap execution environment based on the manifest of dependencies, the execution environment for the software application comprising a set of application dependencies, and storing the execution environment at a data store. The method further includes executing the software application in the execution environment.

Abstract: A system and method for identifying unauthorized uploaded content that has been uploaded before a validated live reference stream has been ingested is disclosed herein. The live reference stream is compared against the indexed uploaded content repeatedly as the live reference stream is received. The matching process is done once per a time period until a match meeting a minimum match duration threshold is identified. The match is then determined to be unauthorized, and a claim is issued against the unauthorized uploaded content. The time period can be based on a utility based analysis that factors the computational costs of repeated matching versus the diminishing value of the live reference stream as time progresses.

Abstract: Methods and apparatus for providing query suggestions to a user based on one or more past queries submitted by the user. Candidate query suggestions responsive to a current query may be identified. A candidate query similarity measure may be determined for a given candidate query suggestion based on matching entities related to the given candidate query suggestion and the one or more past queries. In some implementations, the similarity measure of the given candidate query suggestion may be based on a comparison of current entities of the given candidate query suggestion that match entities of one or more past queries, to a group of the current entities that includes entities that do not match the entities of one or more past queries. In some implementations a ranking of the candidate query suggestions may be determined based on the similarity measure.

Abstract: Deep recurrent neural networks applied to speech recognition. The deep recurrent neural networks (RNNs) are preferably implemented by stacked long short-term memory bidirectional RNNs. The RNNs are trained using end-to-end training with suitable regularisation.

Abstract: The present disclosure provides various transformations to be used in analysis of a large number of transactions to detect anomalies that would indicate potential fraudulent or criminal activity. Such transformations may be applied, for example, using a machine learning system. According to some examples, each of various transformations may be used to detect a particular type of behavioral anomaly. When multiple disparate transformations are considered together by the machine learning system, anomalous activity related to potential fraudulent or criminal activity can be detected more frequently and with greater accuracy.

Abstract: A media application generates training data that includes a first set of visual media items and a second set of visual media items, where the first set of visual media items correspond to the second set of visual items and include distracting objects that are manually segmented. The media application trains a segmentation machine-learning model based on the training data to receive a visual media item with one or more distracting objects and to output a segmentation mask for one or more segmented objects that correspond to the one or more distracting objects.

Abstract: An electronic apparatus includes an electric motor, driver circuitry to drive the electric motor, a temperature sensor to detect a temperature of the electric motor, and one or more processors. The one or more processors control the driver circuitry to drive the electric motor according to a predetermined voltage signal to cause the electric motor to stall and generate heat, receive information about the temperature of the electric motor detected by the temperature sensor while the electric motor is stalled, and maintain a temperature of the electric motor below a threshold temperature value while the electric motor is stalled based on the information about the temperature of the electric motor detected by the temperature sensor while the electric motor is stalled.

Abstract: The present disclosure provides systems and methods for communication efficient distributed mean estimation. In particular, aspects of the present disclosure can be implemented by a system in which a number of vectors reside on a number of different clients, and a centralized server device seeks to estimate the mean of such vectors. According to one aspect of the present disclosure, a client computing device can rotate a vector by a random rotation matrix and then subsequently perform probabilistic quantization on the rotated vector. According to another aspect of the present disclosure, subsequent to quantization but prior to transmission, the client computing can encode the quantized vector according to a variable length coding scheme (e.g., by computing variable length codes).

Abstract: The technology employs a patch-based multi-scale Transformer (300) that is usable with various imaging applications. This avoids constraints on image fixed input size and predicts the quality effectively on a native resolution image. A native resolution image (304) is transformed into a multi-scale representation (302), enabling the Transformer's self-attention mechanism to capture information on both fine-grained detailed patches and coarse-grained global patches. Spatial embedding (316) is employed to map patch positions to a fixed grid, in which patch locations at each scale are hashed to the same grid. A separate scale embedding (318) is employed to distinguish patches coming from different scales in the multiscale representation. Self-attention (508) is performed to create a final image representation. In some instances, prior to performing self-attention, the system may prepend a learnable classification token (322) to the set of input tokens.

Abstract: Implementations are provided for increasing realism of robot simulation by injecting noise into various aspects of the robot simulation. In various implementations, a three-dimensional (3D) environment may be simulated and may include a simulated robot controlled by an external robot controller. Joint command(s) issued by the robot controller and/or simulated sensor data passed to the robot controller may be intercepted. Noise may be injected into the joint command(s) to generate noisy commands. Additionally or alternatively, noise may be injected into the simulated sensor data to generate noisy sensor data. Joint(s) of the simulated robot may be operated in the simulated 3D environment based on the one or more noisy commands. Additionally or alternatively, the noisy sensor data may be provided to the robot controller to cause the robot controller to generate joint commands to control the simulated robot in the simulated 3D environment.

Abstract: To manage sidelink and non-sidelink information, a user device obtains, by processing hardware in a user device communicating with a radio access network (RAN), a first set of information for sidelink communication with another user device (504A), and obtains, by the processing hardware, a second set of information for non-sidelink communication (504B). The user device transmits, by the processing hardware to the RAN, a first message including the first set of information (1010A), and transmits, by the processing hardware to the RAN, a second message including the second set of information (1010B), the first and second messages being separate messages.

Abstract: One or more audio files including a recording of one or more verbal statements provided by a first participant of the conference call are obtained. A determination is made that the recorded one or more verbal statements include a question associated with audio-based polling of a set of second participants of the conference call. Response data indicating one or more responses to the question provided by at least one of the set of second participants is obtained. A report indicating one or more outcomes of the audio-based polling is generated based on at least the question and the one or more responses to the question indicated by the obtained response data.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for automated management of campaigns using scripted rules.

Abstract: A method for presenting content comprises receiving a content item; receiving a plurality of thumbnails from the content item, each thumbnail associated with a point in the content item; playing the content item starting at a first point; receiving in a user interface a selection at a second point in the content item; presenting, in the user interface, first content of the content item at the second point; and presenting, in the user interface, a subset of the plurality of thumbnails, the subset arranged based on a predetermined order and including a thumbnail associated with the second point.

Abstract: A system and methods are disclosed for using a trained machine learning model to identify constituent images within composite images. A method may include providing data identifying a first image as input to a machine learning model trained using training data identifying a plurality of composite images that each include one or more constituent images, and determining, using one or more outputs of the trained machine learning model, that the first image is a composite image that includes a first constituent image, wherein at least a portion of the first constituent image is in a spatial area of the first image, and wherein the first constituent image corresponds to a frame of a video embedded into the first image.

Abstract: Methods, systems, and media for protecting and verifying video files are provided.

Abstract: Systems and methods for processing loops in computational graphs representing machine learning models are disclosed. An example method begins with obtaining data representing a computational graph. Data identifying an allocation of the computational graph across devices is obtained. Additionally, one or more nodes in the computational graph that represent a respective control flow statement are identified. For each identified node, a structure of nodes and edges that represents an operation that provides a current state of recursion or iteration in the respective control flow statement is generated. This structure is inserted into the computational graph and the allocation of nodes to devices is modified to assign the structure to a device.