Abstract: An apparatus and method for adaptive memory refresh scheduling utilize a score-and-filter-based arbitration logic to select bank pairs for refresh operations. A score generation logic determines priority scores for memory banks based on current operational states, including bank page states and transaction histories. A multi-stage filtering process identifies eligible banks by evaluating timing constraints and bank-refresh window requirements. A dual-stage arbitration logic processes the priority scores of the eligible banks to identify an optimal bank index and bank group pair. The arbitration logic employs parallel arbiters and weighted summation logic to select a winning bank index for generating a dual-bank refresh command. By dynamically selecting refresh candidates based on real-time bank activity and state transitions, the system manages refresh scheduling to maintain high throughput and bank utilization across varying workload conditions.

Abstract: A method includes receiving a TTS end event indicating that audible output of first TTS audio from a user device is finished. Based on receiving the TTS end event, the method also includes instructing an ASR system to use a first level of ASR processing for performing speech recognition. As a user speaks a natural language query that solicits a second response from a LLM-powered assistant, the method includes performing, by the ASR system, using the first level of ASR processing, speech recognition on the audio data to generate a speech recognition result and processing, by the LLM-powered assistant, the speech recognition result to generate the second response. Based on receiving a TTS start event indicating that second TTS audio is about to be audibly output from the user device, the method includes instructing the ASR system to use a second level of ASR processing.

Abstract: A method includes receiving a response generated by an assistant LLM that is directed toward the user during an assistant turn in conversation between a user and the assistant LLM. Based on the response generated by the assistant LLM, the method also includes predicting, by the assistant LLM, one or more possible terms the user may speak in a follow-up query to the response generated by the assistant LLM. The method also includes biasing an automated speech recognition (ASR) model toward recognizing the one or more possible terms predicted by the assistant LLM, receiving audio data characterizing the follow-up query to the response generated by the assistant LLM during the user turn subsequent to the assistant turn in the voice-based conversation, and processing, using the ASR model biased toward recognizing the one or more possible terms, the audio data to generate a transcription of the follow-up query.

Abstract: A media application receives an initial image and a textual request to change the initial image, the initial image including a subject with a face. The media application generates, from the initial image, a preserving mask that corresponds to the face of the subject. The media application provides the textual request, the initial image, and the preserving mask as input to a diffusion model. The diffusion model outputs a denoised initial image based on the initial image; performs text conditioning of the textual request and forward diffusion to generate a noisy translated image that satisfies the textual request; and outputs based on the noisy translated image, extracted features, and self-attention maps, a denoised translated image. The media application blends the denoised initial image, the preserving mask, and the denoised translated image to form an output image, wherein the preserving mask prevents modification to the face from the initial image.

Abstract: Provided are systems and methods that improve the computational efficiency of Transformers or other attention-based neural networks or machine learning models by re-using a number of attention scores between layers and/or heads of the model. To reduce the computational cost of self-attention-based models while achieving comparable or even superior results, example aspects of the present disclosure propose a novel architecture that reuses attention scores computed in one layer in one or multiple subsequent layers.

Abstract: A display device may receive eye tracking data. A display device may detect an eye gaze intent state based on the eye tracking data. A display device may receive an input device signal of an input device. A display device may determine to switch from a first input modality to a second input modality based on at least one of the eye gaze intent state or the input device signal. A display device may move a position indicator on a display of the display device using at least one of the first input modality or the second input modality.

Abstract: Provided is an approach that aligns multi-modal tokens using cross-attention without losing the advantages of global self-attention. In contrast to previous works that concatenate the unimodal tokens along the sequence dimension, example approaches described herein align per-modality tokens by chaining them along the channels. Specifically, the tokens from one modality can be used to query the other modality and the output can be concatenated with the query tokens on the channels. An analogous process can also be repeated (or performed in parallel) where the roles of the two modalities are switched. The resulting sets of compound tokens can be concatenated and fed into a self-attention encoder such as a transformer encoder that performs self-attention.

Abstract: Entropy coding a sequence of syntax elements is described where an observation for a syntax element of the sequence is determined, and the observation is arithmetic coded using the probability model. Thereafter, the probability model is updated using a time-variant update rate to produce an updated probability model. Updating the probability model includes regularizing one or more probability values of the probability model so no probability of the updated probability model is below a defined minimum resolution. As a result, the use of a minimum probability value during the arithmetic coding, which can distort probability model, may be omitted.

Abstract: A method includes receiving image data representing an environment of a robotic device from a camera on the robotic device. The method further includes applying a trained dense network to the image data to generate a set of feature values, where the trained dense network has been trained to accomplish a first robot vision task. The method additionally includes applying a trained task-specific head to the set of feature values to generate a task-specific output to accomplish a second robot vision task, where the trained task-specific head has been trained to accomplish the second robot vision task based on previously generated feature values from the trained dense network, where the second robot vision task is different from the first robot vision task. The method also includes controlling the robotic device to operate in the environment based on the task-specific output generated to accomplish the second robot vision task.

Abstract: A method can include receiving audio input of speech, receiving visual input while receiving the audio input, generating a semantic description based on the visual input, and presenting a transcription of the speech based on the audio input and the semantic description.

Abstract: A multi-pass encoding operation using a shared reference frame is implemented to encode one or more gaming frames into a game stream. The multi-pass encoding operation includes producing a shared reference frame based upon a second-pass reference frame used by a second pass encoding of the multi-pass encoding operation. The encoding operation also includes performing a first encoding pass on a current frame using the shared reference frame. As a result of the first encoding pass, an estimated complexity for the current frame is determined. A second pass encoding is then performed on the current frame according to the second-pass reference frame and the estimated complexity, resulting in an encoded frame. This encoded frame is then transmitted as part of a stream to a client system.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for batched reinforcement learning. For example, the batched reinforcement learning techniques can be used to determine a control policy for a robot in simulation and the control policy can then be used to control the robot in the real world. In one aspect, a method includes obtaining a plurality of current observations, each current observation characterizing a current state of a respective environment replica; processing the current observations in parallel using the action selection neural network in accordance with current values of the network parameters to generate an action batch; obtaining a transition tuple batch including a respective transition tuple for each of the environment replicas, the respective transition tuple for each environment replica including: (i) a subsequent observation and (ii) a reward; and training the action selection neural network on the batch of transition tuples.

Abstract: A system and method for balancing data storage among a plurality of groups of computing devices, each group comprising one or more respective computing devices, each group having an available storage capacity. The method may involve, for each group of computing devices, determining an amount of used storage at the group of computing devices exceeding a predefined first threshold value that is less than the available storage capacity and calculating a storage cost based on the determined amount of used storage exceeding the predefined first threshold value, determining a total storage cost of the plurality of groups of computing devices based on a sum of the calculated storage costs, determining a transfer of one or more projects between the groups of computing devices that reduces the total storage and directing the plurality of groups of computing devices to execute the determined transfer.

Abstract: A system and method provides for the transfer of the output of content from a user device to an external device for output by the external device. External devices may be detected in a physical space, and identified based on previous connection with the user device, based on a shared network or shared system of connected devices including the user device, based on image information captured by the user device and previously stored anchoring information that identifies the external devices, and the like. An external device may be selected for potential output of the content based on previously stored configuration information associated with the external device including, for example, output capabilities associated with the external device. The identified external device may output the transferred content in response to a user verification input, verifying that the content is to be output by the external device.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network to perform 3D object detection. One of the methods includes training a student neural network to perform 3D object detection using pseudo-labels generated by a teacher neural network.

Abstract: A method for generating video suggestions to users of an online video streaming platform is provided. A video suggestion engine may scan and analyze the comments section of an online video to generate a sentiment score associated with the video. The sentiment score may be indicative of the level of association between a given sentiment and a given video. After, the video suggestion engine may suggest the video to a user, based on the computed sentiment score. In some cases, more than one sentiment score may be computed for a given video. In some cases, multiple sentiment scores may be computed for multiple videos. In some cases sentiment scores across videos can be compared to identify a video for suggesting to a user, and the method may generate a video suggestion based on such a comparison.

Abstract: A display panel includes: a display active area having multiple pixels; a bezel surrounding the display active area; a connector tab attached to the bezel at a first edge of the bezel; and a power line(s) formed from an opaque material, the power line(s) extending from the connector tab to a second edge of the bezel opposite the first edge, the power line(s) being configured to supply electrical power to the display active area, the power line(s) having a first width in a plane of the display along a side edge of the bezel between the first and second edges and, at a portion of the bezel at the second edge, a second width narrower than the first width or a gap in the power lines, the portion of the bezel at the second edge being substantially transparent.