Abstract: Techniques for live stream processing and sensor-based measurement presentation therein are described. In an example, a computer system determines, in-real time relative to a player being in a space, real-time data of the player. The real-time data is generated during a play in the space. The computer system generates, during the play, an input to an artificial intelligence model based on the real-time data. The computer system determines an output of the artificial intelligence model. The output is based on the input and indicates a prediction of an action to be performed by the player during the play. The computer system causes, based on the prediction, a presentation of a user interface element during the play, the user interface element indicates the prediction.

Abstract: Techniques are generally described for scene-aware generation of music visualizations. An example method includes receiving a music sample and a projection surface image, and generating a constraint image based on the projection surface image. The example method also includes generating a sentiment analysis of the music sample and generating an intermediate prompt using the music sample metadata and the sentiment analysis. The example method also includes processing the intermediate prompt using a large language model and generating a subsequent set of keyframe prompts for a latent diffusion model. The example method also includes generating a set of keyframes using the latent diffusion model and generation interpolation images from the keyframes. Finally, the example method includes aligning the keyframes and interpolation images to beats detected in the music sample and generating a music visualization video.

Abstract: This disclosure describes systems for enabling a just-walk-out (JWO) facility in which users may enter the facility, physically pick one or more items, and walk out of the facility to rent, purchase, or otherwise acquire the items without needing to perform a manual checkout for the items. In some instances, a tag, such as an RFID tag, may attach to each item in the facility, which may include an exit system that includes RFID readers for detecting when a tag exits the facility. The exit system may also include one or more cameras that may be used to help identify which of multiple checkout lanes the tag and item exited through for ensuring that the item is associated with the proper user account and/or the appropriate payment instrument is charged.

Abstract: A quantum computing service includes a quantum simulator using classical hardware configured to perform a sparse simulation of a Clifford dominated quantum circuit that includes at least some non-Clifford gates. The non-Clifford gates are represented using a Clifford portion and a non-Clifford portion, wherein application of the Clifford portion of the non-Clifford gates includes updating a Clifford tableau for the quantum circuit being simulated and application of the non-Clifford portion of the non-Clifford gates includes applying a rotational term that is at most 2-sparse to a set of key-value pairs representing a sparse state of the quantum circuit being simulated.

Abstract: Techniques implemented by batch-processing service to provide users with visibility around execution metrics for their batch jobs, and determine more optimal scheduling policies and/or scaling algorithms for executing batch jobs. The batch-processing service may collect telemetry data for batch jobs, and analyze the telemetry data to generate execution metrics for the batch jobs. The batch-processing service uses these execution metrics to generate various visualizations that may be presented to users to provide the users with more visibility into the performance of their batch jobs. Additionally, the batch-processing service may analyze the execution metrics, and determine more optimized scheduling policies and/or scaling algorithms to use for running the batch jobs. In this way, the batch-processing service can increase the visibility given to users around the performance of their batch jobs, and can also identify modifications to scheduling and scaling in order to improve the performance of the batch jobs.

Abstract: Some embodiments provide a method for training parameters of a network. The method receives an MT network having multiple layers of computation nodes that use first sets of weight values. Each computation node of a first layer computes an output value by computing an intermediate value based on a set of input values and a set of weight values and computing a node output value by applying an activation function to the intermediate value. The method replaces the first layer with a second layer of computation nodes followed by a third layer of computation nodes. The replacement involves (i) decomposing the sets of weight values of the first layer into second sets of weight values for the second layer and third sets of weight values for the third layer and (ii) inserting into the network a set of additional activation functions for the second layer.

Abstract: Systems and methods are provided for managing objects. In one implementation, a computer-implemented method is provided. The method includes receiving a query comprising a tag and executing the query. An object identifier is retrieved from a data table, based on the tag. The method further returns a result of the query. The result includes the object identifier that was retrieved from the data table. The method further performing an action related to an object having the retrieved object identifier.

Abstract: Techniques for supporting a token based service are described. In some examples, a token based service such as a foundation model service supports global endpoints. Global endpoints for the token based service are provided with a global quota that is exposed per account, per model. Invocations against a global endpoint consume against the global quota.

Abstract: Some embodiments provide a method for executing a layer of a neural network, for a circuit that restricts a number of weight values used per layer. The method applies a first set of weights to a set of inputs to generate a first set of results. The first set of weights are restricted to a first set of allowed values. For each of one or more additional sets of weights, the method applies the respective additional set of weights to the same set of inputs to generate a respective additional set of results. The respective additional set of weights is restricted to a respective additional set of allowed values that is related to the first set of allowed values and the other additional sets of allowed values. The method generates outputs for the particular layer by combining the first set of results with each respective additional set of results.

Abstract: A vehicle software deployment management system generates a modifiable deployment plan and one or more associated vehicle software modules that is sent to an edge device at a vehicle activity site. The edge device stores the modifiable deployment plan and the one or more vehicle software modules for updating a given vehicle at a future time. In some embodiments, the vehicle may have insufficient network connectivity to perform the update remotely from the vehicle activity site. In some embodiments, the modifiable deployment plan may be modified by the edge device at the future time when the vehicle is present at the vehicle activity site, based on vehicle information obtained by the edge device and based on vehicle user/technician input obtained by the edge device.

Abstract: Described are systems and processes for compressing image data, such as video data, based on detection of region of interests (ROIs) in the image data. Image data may be analyzed to determine moving objects and stationary objects (e.g., background) depicted in the image data. To compress the image data, the stationary objects are refreshed less frequently than moving objects. ROIs are identified in the moving objects, such as a face of a person, hands of a person, and/or objects moved by the person. Some of these ROI regions/pixels are maintained at high quality, while non-ROI regions/pixels are reduced in quality to further compress the image data. Faces are reduced in quality to obfuscate a person's identity. Multiple ROI regions may be located and then tracked simultaneously. The resulting compression may reduce storage space to store the image data and/or may reduce bandwidth needed to transmit the image data to remote devices while maintaining high quality in ROI regions.

Abstract: Frames may be encoded at different resolutions. A scaling factor may be used to scale video frames used for translational or affine motion compensation. Various improvements to motion compensation processes for different resolution video frames are described.

Abstract: Techniques for an intent-inference system to use large language models (LLMs) to infer intents of applications running in user accounts. The system is granted permissions to obtain metadata for application resources, and uses the LLMs to analyze the metadata and infer the intended functionality of the applications. The system uses prompt engineering to interact with the LLMs by embedding the metadata as input into the LLMs, and has LLMs infer the intended functionality of resources in an application. The system generates a formal specification of the application that defines the intended relationships between the resources, and the overall functionality of the application. The formal specification may be used to generate a human-readable summary for users to validate that the inferred formal specification matches their understanding of their application. Further, the intent-inference system may generate a machine-readable representation of the formal specification that is used for various use cases.

Abstract: A device includes a platform configured to receive a plurality of containers, a fan, a manifold, and a biasing element. The manifold is fluidly connected to the fan and directs air into a container of the plurality of containers. Directing air into the container causes the container to at least partially expand to permit the container to be loaded with one or more items. The biasing element biases the platform in a direction toward the air being directed into the container.

Abstract: Techniques for an optical waveguide with homogenous light intensity output are described herein. In an example, an optical waveguide system includes an optical waveguide substrate and a first holographic layer coupled to the optical waveguide substrate and configured to inject light into the optical waveguide substrate as injected light. The optical waveguide system also includes a second holographic layer coupled to the optical waveguide substrate and configured to receive the injected light and output the injected light as output lights in a direction. The second holographic layer is characterized by a varying diffraction efficiency and the output lights have a homogenous intensity at a distance from the second holographic layer.

Abstract: A technique for deriving Clifford circuits using a Satisfiability Modulo Theories (SMT) solver via encoding a Clifford circuit design problem as an SMT decision problem is disclosed. Both symbolic bit matrix representations of elements of the Clifford circuit design problem and constraint equations for modeling the Clifford circuit are used to construct an encoded SMT solver protocol. Clifford circuit solutions are then solved for by the SMT solver before one or more of the solutions are implemented into quantum hardware. Such a technique can be applied to bottom-up fault-tolerant magic state preparation protocol such that an encoded magic state can be teleported from a color code to a surface code via lattice surgery, resulting in a merged surface and color code. Decoding the stabilizer measurements of the merged code requires a decoding algorithm specific to error correction of the merged code.

Abstract: A machine readable policy generated from a policy specification may be obtained. The machine readable policy may specify an exit reason for a virtual processor (vCPU) to exit execution on a host computing system and a set of register access permissions associated with the exit reason. Upon the vCPU exiting for the exit reason, a separation kernel that separates the hypervisor executing on the host computing system from direct hardware access applies the machine readable policy. Applying the machine readable policy includes determining that the hypervisor is requesting access to a register associated with the vCPU, evaluating the access against the set of register access permissions, and allowing the hypervisor to access the register based on the evaluation.

Abstract: Techniques to improve memory utilization for performing a collective compute operation may include allocating an amount of space in a shared memory of a processing node that is part of an integrated circuit device having multiple processing nodes. A number of sharded portions of parameters are obtained and stored in the shared memory. Each of the sharded portions of parameters stored in the shared memory is multicasted to a state buffer of other processing nodes in the integrated circuit device to provide each state buffer with the number of sharded portions of parameters for the collective compute operation.

Abstract: A system for enhancing security in networked environments through the aggregation, transformation, and dissemination of security incident data. The system collects data from various sources, including human threat intelligence and machine-readable logs, and transforms this data into a machine-executable data structure. The system stores the transformed data in a threat intelligence data store, making the threat intelligence data store accessible via application programming interface (API) calls. The system includes components that implement large language models (LLMs) to generate security incident summaries and visualizations tailored to specific user roles.

Abstract: Disclosed are various embodiments for dynamic location update and verification for radio units in radio-based networks. In one embodiment, a current location of a radio unit in a radio-based network is automatically detected. It is determined that the current location differs from a registered location of the radio unit. One or more actions are initiated in response to determining that the current location differs from the registered location.