Abstract: Certain aspects of the present disclosure provide techniques and apparatus for a training and using machine learning models in multi-device network environments. An example computer-implemented method for network communications performed by a host device includes extracting a feature set from a data set associated with a client device using a client-device-specific feature extractor, wherein the feature set comprises a subset of features in a common feature space, training a task-specific model based on the extracted feature set and one or more other feature sets associated with other client devices, wherein the feature sets associated with the other client devices comprise one or more subsets of features in the common feature space, and deploying, to each respective client device of a plurality of client devices, a respective version of the task-specific model.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for improved machine learning. Voice data from a first user is received. In response to determining that the voice data includes an utterance of a defined keyword, a user verification score is generated by processing the voice data using a first user verification machine learning (ML) model, and a quality of the voice data is determined. In response to determining that the user verification score and determined quality satisfy one or more defined criteria, a second user verification ML model is updated based on the voice data.

Abstract: Systems and techniques are provided for processing audio data. For instance, a process can include detecting a first audio data between two or more in-person participants of a plurality of in-person participants of a group communication session. The first audio data can be identified as private based on one or more cues associated with the first audio data. A second audio data can be identified between the two or more in-person participants. The second audio data can be identified as non-private based on one or more additional cues associated with the second audio data, wherein the one or more additional cues associated with the second audio data are different from the one or more cues associated with the first audio data. The second audio data can be output based on identifying the second audio data as non-private.

Abstract: Systems and techniques are provided for processing audio data. For instance, a process can include obtaining a primary audio signal from a user computing device and obtaining first audio data from an additional computing device, wherein the additional computing device is within an auditory range of the user computing device. The primary audio signal can be filtered to generate a filtered primary audio signal. The filtering can include removing a representation of the first audio data from the primary audio signal. The filtered primary audio signal can be output to a network-based meeting.

Abstract: Systems and techniques are provided for processing audio data. For instance, a process can include detecting a first audio data between two or more in-person participants of a plurality of in-person participants of a group communication session. The first audio data can be identified as private based on one or more cues associated with the first audio data. A second audio data can be identified between the two or more in-person participants. The second audio data can be identified as non-private based on one or more additional cues associated with the second audio data, wherein the one or more additional cues associated with the second audio data are different from the one or more cues associated with the first audio data. The second audio data can be output based on identifying the second audio data as non-private.

Abstract: Techniques are provided for motion estimation using hybrid video imaging based on frame-based capture and event-based capture. A methodology implementing the techniques according to an embodiment includes receiving a sequence of pixel events, generated asynchronously by an event-based video camera, and receiving a sequence of image frames generated by a frame-based video camera at a frame sampling period. The method also includes integrating a subset of the sequence of pixel events, occurring within the frame sampling period between two of the image frames, to generate a pixel motion vector representing motion of the pixel between the two image frames. The method further includes mapping the pixel motion vector to a tile of one of the image frames to generate an estimated motion vector associated with that tile.

Abstract: Techniques are provided for motion estimation using hybrid video imaging based on frame-based capture and event-based capture. A methodology implementing the techniques according to an embodiment includes receiving a sequence of pixel events, generated asynchronously by an event-based video camera, and receiving a sequence of image frames generated by a frame-based video camera at a frame sampling period. The method also includes integrating a subset of the sequence of pixel events, occurring within the frame sampling period between two of the image frames, to generate a pixel motion vector representing motion of the pixel between the two image frames. The method further includes mapping the pixel motion vector to a tile of one of the image frames to generate an estimated motion vector associated with that tile.

Abstract: High dynamic range depth generation is described for 3D imaging systems. One example includes receiving a first exposure of a scene having a first exposure level, determining a first depth map for the first depth exposure, receiving a second exposure of the scene having a second exposure level, determining a second depth map for the second depth exposure, and combining the first and second depth map to generate a combined depth map of the scene.

Abstract: An image sensor includes an array of pixels on a semiconductor device for sensing light incident on the pixel array and a plurality of anomaly registers comprising a plurality of nonvolatile elements. Each anomaly register identifies an anomalous pixel cluster and includes a location indicator with a row and column address and a size indicator with a horizontal and vertical range. In other embodiments, each anomaly register includes a first address, second address, first direction flag, and second direction flag. The first and second direction flags use a first state to indicate a row address or a second state to indicate a column address. The first and second direction flags combine to define an anomalous pixel cluster, a pair of anomalous pixel rows, or a pair of anomalous pixel columns. Some embodiments may include an anomaly type indicator and some embodiments may include a shape indicator.

Abstract: In one embodiment, an electronic device comprises a plurality of electrical switches and a plurality of energy storage elements arrayed relative to one another such that the energy storage elements may be connected in series, or in parallel, or both, to an input and an output.