Abstract: Various examples, systems, and methods are disclosed relating to a slicing pipeline. A first computing system can process, by a processing component of a system-on-chip (SoC), at least one first frame into a plurality of first slices based at least on a first slice parameter. The first computing system further can update at least one performance metric based at least on timing data corresponding with the processing of the plurality of first slices. The first computing system further can determine a second slice parameter based at least on the at least one performance metric. The first computing system further can apply, in real-time, the second slice parameter to the processing component to cause subsequence processing of at least one second frame into a plurality of second slices based at least on the second slice parameter.

Abstract: In various examples, systems and methods are disclosed relating to implementing parallel computation of Boolean logic operations between 2-D geometric polygon sets. A system can identify a plurality of first edges of a first polygon and a plurality of second edges of a second polygon. The system can generate, based at least on a Boolean operation for the first polygon and the second polygon, a set of intersection points corresponding to intersections between at least one first edge of the plurality of first edges and at least one second edge of the plurality of second edges. The system can generate a directed graph data structure for the Boolean operation using the set of intersection points, and generate an output polygon based at least on the directed graph data structure.

Abstract: Systems and methods for monitoring data center metrics using machine learning are disclosed. A system can obtain a first set of sensor readings from data center components during a first time period. The system can generate a dataset using the first set of sensor readings and a negative sampling function. The dataset can include labels generated according to a learned distribution derived from the first set of sensor readings. The system can update a machine learning model using the dataset to predict likelihoods of future anomalies in the plurality of data center components. The system can obtain a second set of sensor readings from the data center components during a second time period. The system can generate, using the machine learning model and the second set of sensor readings, a prediction of a future anomaly that may occur in at least one of the data center components.

Abstract: In various examples, a system can include one or more processors to determine, using a plurality of retriever models, a quality score of each of a plurality of queries in a first dataset, determine, using a plurality of embedding models, a metric indicative of a relationship between each of the plurality of queries and the set of text information of the first dataset, and select one of the plurality of embedding models as a filter model based at least in part on correlation between the metrics outputted by the plurality of embedding models and the quality scores for the plurality of queries, the filter model applied to filter a plurality of queries and a plurality of sets of text information in a second dataset.

Abstract: In various examples, scenarios may be defined using a declarative description—e.g., defining a behavior of interest—that the present system may convert into a procedural description for generating one or more instances and/or variations of a scenario for testing an autonomous or semi-autonomous machine in a virtual environment. The system may execute observers or evaluators for testing the performance and accuracy of the machine and may compute coverage of various elements based on the generated virtual scenarios, and may feed the results back to the system to generate additional instances and/or variations where the coverage or accuracy is below a desired level. As a result, the system may include an end-to-end framework for generating scenarios in virtual environments, testing and validating the scenarios themselves, and/or testing and validating the underlying autonomous or semi-autonomous systems of the machine—all based on a declarative description.

Abstract: Various examples, systems, and methods are disclosed relating to automated generation of data source descriptions for agentic workflows. A system can query a data collection with a zero-vector to retrieve a subset of data from the data collection. A system can generate, using a machine learning model, a summary of the data collection based at least on the subset of data.

Abstract: In various examples, parking area detection for autonomous and/or semi-autonomous systems and applications is described herein. Systems and methods described herein may use a hybrid method to more accurately determine geometries of the parking areas within environments. For instance, sensor data (e.g., image data, etc.) may be processed using one or more edge detection techniques to determine edge features (e.g., two-dimensional pixels, three-dimensional points, etc.) associated with an environment that includes a parking area. A predicted geometry associated with the parking area, as determined using one or more machine learning models, may then be used to filter the edge features in order to identify a portion of the edge features that represent the parking area. One or more optimization techniques may then be used to determine the final geometry associated with the parking area based at least on the filtered edge features.

Abstract: In various examples, segmented map creation for autonomous and semi-autonomous systems and applications is described herein. For instance, systems and methods described herein may segment a map into various portions, where the systems and methods then update the map based at least on the segmentation. For instance, based at least on receiving instances of data from machines navigating with an environment associated with the HD map, the instances of data may be associated with various portions of the map. Additionally, metrics associated with the portions of the map may be updated, such as by indicating the numbers of instances of data that are associated with the portions of the map. As such, when a metric associated with a portion of a map is satisfied, such as by reaching a threshold, the portion of the map may be updated without updating one or more other portions of the map.

Abstract: Apparatuses, systems, and techniques to scale processor clocks. In at least one embodiment, one or more circuits are to scale one or more clocks of one or more cores based, at least in part, on a proximity of the one or more cores to each other.

Abstract: Apparatuses, systems, and techniques to perform multicast data transmissions in parallel. In at least one embodiment, a plurality of pending multicast data transmissions are analyzed to select a subset of multicast transmissions that can be performed in parallel. In at least one embodiment, the selection is made by prioritizing senders using a rotating priority scheme.

Abstract: Systems and methods provide for a machine learning system to train a machine learning model to output a penalty-free emission when processing an auditory input. For example, as the system generates paths through a probability lattice, one or more paths may include a penalty-free emission that skips at least one frame associated with the probability lattice, but that does not add a cost to a final path cost. The use of the penalty-free emissions may be represented through one or more graphical representations used for training in order to develop loss functions for models. One or more of these frameworks may be incorporated into automatic speech recognition pipelines to improve training while also reducing coding requirements to simplify debugging operations.

Abstract: An architecture can generate lane graphs or path determinations, for devices such as robots or autonomous vehicles, using multiple sources of data while satisfying applicable requirements and regulations for operation. A system can fuse together data from multiple sources useful to determine localization. To ensure safety compliance, this fused data is compared against data from systems where safety is trusted and, as long as at least two comparators agree with the fused localization data, the fused localization data can be used and verified to be safety regulation compliant. This system can also fuse together available information useful for lane perception. This fused data is compared against data from systems where the safety is trusted, and as long as at least two comparators for these safety-compliant systems agree with the fused lane graph data, then the fused lane graph data can be provided for navigation and verified to be regulation compliant.

Abstract: In various examples, accuracy determinations for localization in autonomous and semi-autonomous systems and applications are described herein. Systems and methods are disclosed that determine one or more errors associated with vehicle localization using various types of sensor data generated using a vehicle. For instance, a first component of the vehicle may use a map and first sensor data to determine an estimated pose of the vehicle. A second component of the vehicle may then determine the error(s) associated with the estimated pose based on both actual motion of the vehicle within the environment, as determined using second sensor data, and comparing features represented by the first sensor data to features represented by the map. In some examples, the second component may further determine information associated with the error(s), such as one or more uncertainties associated with the error(s).

Abstract: Apparatuses, systems, and techniques to concatenate multiple videos together to generate a single video. In at least one embodiment, an object is identified in two or more videos using metadata generated by one or more neural networks, and those videos are concatenated so that an object's movements are captured in a single video.

Abstract: Apparatuses, systems, and techniques relate to neural components for differentiable ray tracing of radio propagation. Differentiable ray tracing may be used to refine the scene geometry of the physical environment, to learn or optimize the scene properties of objects in the scene, to learn or optimize the scene properties of antennas, and to learn or optimize antenna patterns, array geometries, and orientations and positions of transmitters and receivers. Once scene properties have been learned or optimized, the differentiable ray tracer may further be used to simulate the performance of different configurations of the transmitters, receivers, and scene geometry. In an embodiment, one or more of the scene geometry, scene properties, and antenna characteristics are computed by a differentiable parametric function, such as a neural network, etc. and parameters of the differentiable parametric function are learned using the differentiable ray tracing.

Abstract: A device includes a memory and one or more processing devices operatively coupled to the memory. The one or more processing devices to determine that a data request comprises first persistent data, remove the first persistent data from the data request to obtain first dynamic data, generate first modification data representing the first persistent data, combine the first dynamic data and first modification data to obtain a first modified data request, and cause the first modified data request to be transmitted to a second device over a communication link.

Abstract: Apparatuses, systems, and techniques to perform neural networks. In at least one embodiment, a processor comprising one or more circuits uses one or more neural networks to adjust brightness of pixels of images prior to denoising the images.

Abstract: Systems and methods herein address scalable contact-rich simulation in physics engines using one or more processing units to simulate movement between at least two objects in a simulation, the movement based at least on a plurality of sets of reduced points obtained from an iterative reduction using one or more threshold criteria, the iterative reduction applied to a plurality of points associated with at least one contact between the depictions.

Abstract: Apparatuses, systems, and techniques to train one or more neural networks. In at least one embodiment, one or more neural networks are trained based, at least in part, on inferencing output from one or more second neural networks.

Abstract: In various examples, a deep neural network (DNN) may be used to detect and classify animate objects and/or parts of an environment. The DNN may be trained using camera-to-LiDAR cross injection to generate reliable ground truth data for LiDAR range images. For example, annotations generated in the image domain may be propagated to the LiDAR domain to increase the accuracy of the ground truth data in the LiDAR domain—e.g., without requiring manual annotation in the LiDAR domain. Once trained, the DNN may output instance segmentation masks, class segmentation masks, and/or bounding shape proposals corresponding to two-dimensional (2D) LiDAR range images, and the outputs may be fused together to project the outputs into three-dimensional (3D) LiDAR point clouds. This 2D and/or 3D information output by the DNN may be provided to an autonomous vehicle drive stack to enable safe planning and control of the autonomous vehicle.