Abstract: Apparatuses, systems, and techniques are presented to select segments of content. In at least one embodiment, one or more neural networks are used to select one or more video segments from one or more video files based at least in part upon an indication of one or more emotions, of one or more viewers of the one or more video segments, during presentation of the one or more video segments.

Abstract: In various examples, hybrid models for determining intents in conversational AI systems and applications are disclosed. Systems and methods are disclosed that use a machine learning model(s) and a data file(s) that associates intents with one another (e.g., using a tree-like structure) in order to determine a final intent associated with text. For example, the text may initially be processed using the machine learning model(s) (e.g., a first machine learning model) in order to determine a first intent associated with the text. The data file(s) may then be used to determine information (e.g., anchors) for one or more second intents (e.g., one or more sub-intents) that are related to the first intent. The text and the information may then be processed using the machine learning model(s) (e.g., a second machine learning model) to determine a second intent, from the one or more second intents, that is associated with the text.

Abstract: Systems and methods to support remote client access to server-based software development within a server that manages a container cluster, are disclosed. Exemplary implementations may launch one or more pods that may include sets of containers, including a first pod that executes a container management software application; launch a first set of containers including a first container; receive a connection instruction to establish a secure channel between a client computing platform and a remotely-accessible server-based software development environment (SDE) in the first container; establish the secure channel; receive user input for particular execution in the remotely-accessible server-based software development environment; perform the particular execution; and/or other actions.

Abstract: In various examples, metadata-based image harmonization for image stitching systems and applications are disclosed. Systems and methods are disclosed that preprocess images with respect to rendering parameters, with the effect of blending those parameters at a border between images to facilitate a smooth rendering when those images are stitched together. An image signal processing (ISP) parameter harmonization function may input metadata parameters associated with a set of images to match and blend one or more of the rendering parameters across an overlapping border between images prior to applying those images to a stitching algorithm. A scaling of the metadata parameter may be performed using a parameter gain function. Pixels in both images located along the border are adjusted to the same boundary metadata parameter value, and smoothed based on the parameter gain function. A discontinuity in rendering parameters is avoided, substantially avoiding corresponding artifacts in the resulting stitched image.

Abstract: An HD map system represents landmarks on a high definition map for autonomous vehicle navigation, including describing spatial location of lanes of a road and semantic information about each lane, and along with traffic signs and landmarks. The system generates lane lines designating lanes of roads based on, for example, mapping of camera image pixels with high probability of being on lane lines into a three-dimensional space, and locating/connecting center lines of the lane lines. The system builds a large connected network of lane elements and their connections as a lane element graph. The system also represents traffic signs based on camera images and detection and ranging sensor depth maps. These landmarks are used in building a high definition map that allows autonomous vehicles to safely navigate through their environments.

Abstract: In various examples, a technique for generating speed change decisions for a mobile robot includes identifying, using one or more maps of a physical environment, one or more obstacles associated with one or more portions of a path of the mobile robot in the physical environment. The technique also includes generating, based at least on the one or more obstacles, one or more speed constraints, each speed constraint specifying a speed limit for a respective portion of the path. The technique further includes generating one or more speed change decisions specifying actions to be performed by the mobile robot to cause a speed profile of the mobile robot to satisfy the one or more speed constraints.

Abstract: Apparatuses, systems, and techniques are presented to reduce an amount of data to be transmitted for media content. In at least one embodiment, one or more neural networks are used to generate video and audio information corresponding to one or more people based, at least in part, on at least one image and voice information corresponding to the one or more people.

Abstract: Apparatuses, systems, and techniques to store information within one or more non-uniform memory access (NUMA) storages. In at least one embodiment, one or more circuits are to perform an application programming interface (API) to cause information to be stored within one or more NUMA storages or one or more graphics processor unit (GPU) physical storages based, at least in part, on one or more indicators to be indicated by one or more users of the API.

Abstract: Systems and methods include a fan and a fan controller configured to set a default speed of the fan. An initialization controller is configured to determine that the fan is powered on and to issue a control signal to the fan controller, responsive to the determination that the fan is powered on. The control signal overrides the default speed of the fan.

Abstract: Various embodiments can perform efficient, large scale fluid simulation using a method, such as the fluid-implicit particle (FLIP) method, over a sparse hierarchy of grids. The grids may be represented in a number of different formats, and may include GVDB Voxels. Such approaches can handle tens of millions of particles within a virtually unbounded simulation domain. Embodiments can utilize a parallel, sparse grid hierarchy construction and provide for fast incremental updates on graphics processing unit (GPU) hardware, for example, for moving particles. In addition, a FLIP-based technique can be used to perform sparse, work-efficient parallel data gathering from particle to voxel. Various embodiments can also utilize a matrix-free GPU-based conjugate gradient solver optimized for sparse grids. Such approaches can provide orders of magnitude faster simulations on GPU hardware with respect to conventional simulations.

Abstract: One embodiment of a method for controlling a robot includes performing a plurality of simulations of a robot interacting with one or more objects represented by one or more signed distance functions (SDFs), where performing the plurality of simulations comprises reducing a number of contacts between the one or more objects that are being simulated, and updating one or more parameters of a machine learning model based on the plurality of simulations to generate a trained machine learning model.

Abstract: Apparatuses, systems, and techniques to train one or more neural networks using stratified sampled training data parameters. In at least one embodiment, one or more stochastic training data parameters may be stratified sampled from one or more sampling ranges to compute a gradient for updating the one or more neural networks.

Abstract: A method for generating, by an encoder-based model, a three-dimensional (3D) representation of a two-dimensional (2D) image is provided. The encoder-based model is trained to infer the 3D representation using a synthetic training data set generated by a pre-trained model. The pre-trained model is a 3D generative model that produces a 3D representation and a corresponding 2D rendering, which can be used to train a separate encoder-based model for downstream tasks like estimating a triplane representation, neural radiance field, mesh, depth map, 3D key points, or the like, given a single input image, using the pseudo ground truth 3D synthetic training data set. In a particular embodiment, the encoder-based model is trained to predict a triplane representation of the input image, which can then be rendered by a volume renderer according to pose information to generate an output image of the 3D scene from the corresponding viewpoint.

Abstract: One embodiment of the present invention sets forth a technique for performing nested kernel execution within a parallel processing subsystem. The technique involves enabling a parent thread to launch a nested child grid on the parallel processing subsystem, and enabling the parent thread to perform a thread synchronization barrier on the child grid for proper execution semantics between the parent thread and the child grid. This technique advantageously enables the parallel processing subsystem to perform a richer set of programming constructs, such as conditionally executed and nested operations and externally defined library functions without the additional complexity of CPU involvement.

Abstract: An estimation model utilizes simulations of an autonomous vehicle and objects detected near the automated vehicle to develop estimates of tolerable frame processing latency to develop real world frame processing latency estimates for similar driving conditions. An estimation model can a minimum tolerable latency for processing the frames of image data of an object detection camera on an autonomous vehicle using the object state data of the objects detected near the autonomous vehicle. An autonomous vehicle system process can determine if the processing latency of a sensor is greater than the modeled tolerable latency for that sensor, then a safety check is failed and an alert is sent. An autonomous vehicle system process can determine if the processing latency of a sensor is greater than the modeled tolerable latency for that sensor, then the hardware resources are prioritized to the processing for that sensor.

Abstract: Apparatuses, systems, and techniques to select one or more discontinuous wireless communication patterns. In at least one embodiment, a processor includes one or more circuits to cause one or more different discontinuous wireless communication patterns to be selected based, at least in part, on one or more different beams with which to communicate the one or more different discontinuous wireless communication patterns.

Abstract: Apparatuses, systems, and techniques to determine a matrix multiplication algorithm for a matrix multiplication operation. In at least one embodiment, a matrix multiplication operation is analyzed to determine an appropriate matrix multiplication algorithm to perform the matrix multiplication algorithm.

Abstract: A deep neural network(s) (DNN) may be used to detect objects from sensor data of a three dimensional (3D) environment. For example, a multi-view perception DNN may include multiple constituent DNNs or stages chained together that sequentially process different views of the 3D environment. An example DNN may include a first stage that performs class segmentation in a first view (e.g., perspective view) and a second stage that performs class segmentation and/or regresses instance geometry in a second view (e.g., top-down). The DNN outputs may be processed to generate 2D and/or 3D bounding boxes and class labels for detected objects in the 3D environment. As such, the techniques described herein may be used to detect and classify animate objects and/or parts of an environment, and these detections and classifications may be provided to an autonomous vehicle drive stack to enable safe planning and control of the autonomous vehicle.

Abstract: A machine learning model (MLM) may be trained and evaluated. Attribute-based performance metrics may be analyzed to identify attributes for which the MLM is performing below a threshold when each are present in a sample. A generative neural network (GNN) may be used to generate samples including compositions of the attributes, and the samples may be used to augment the data used to train the MLM. This may be repeated until one or more criteria are satisfied. In various examples, a temporal sequence of data items, such as frames of a video, may be generated which may form samples of the data set. Sets of attribute values may be determined based on one or more temporal scenarios to be represented in the data set, and one or more GNNs may be used to generate the sequence to depict information corresponding to the attribute values.

Abstract: In various examples, an obstacle detector is capable of tracking a velocity state of detected objects or obstacles using LiDAR data. For example, using LiDAR data alone, an iterative closest point (ICP) algorithm may be used to determine a current state of detected objects for a current frame and a Kalman filter may be used to maintain a tracked state of the one or more objects detected over time. The obstacle detector may be configured to estimate velocity for one or more detected objects, compare the estimated velocity to one or more previous tracked states for previously detected objects, determine that the detected objects corresponds to a certain previously detected object, and update the tracked state for the previously detected object with the estimated velocity.