Search NASA⌕ Search

SEARCH · Search NASA

Results for “pruning”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

97 records · Page 6
All sourcesNASA NTRSDOE OSTI
NASA NTRSDate not supplied

Aerial Vehicle Routing and Scheduling for UAS Traffic Management: A Hybrid Monte Carlo Tree Search Approach

We present the Multi-Route Weighted Package Delivery Problem (MRWPDP) and a scalable solution methodology as a major step towards enabling an airspace deconfliction service for drone delivery operations. The problem is motivated by Strategic deconfliction under the FAA’s “Unmanned Aircraft Systems Traffic Management” Concept of Operations. MRWPDP falls under a class of vehicle routing and scheduling problems, and as such is NP-Hard. In MRWPDP, a graph network is given which consists of depots, drop-off sites, and multiple routes connecting the two. In addition, routes are weighted by the associated ground risk and total travel distance for package delivery. The goal is to optimally schedule the departure time and assign routes to a known set of vehicles at the depot. We propose a heuristic solution to the problem by borrowing techniques from Mixed Integer Linear Programming (MILP), Constraint Programming, and Monte Carlo Tree Search (MCTS). The resulting hybrid framework is MCTS with Bound-and-Prune (BP) and rapid simulated updates (U), or MCTS-BP-U. This approach is able to quickly provide a feasible solution for MRWPDP, even for large problem instances up to 1000 vehicles. We provide a MILP formulation of MRWPDP and compare its performance against MCTS-BP-U in terms of solution quality. An agent-based model simulation is conducted as a final step to validate the efficacy of our approach.

air traffic scheduling↗
NASA NTRS2023

Expanding the Deep Space Network to Support the Heliophysics System Observatory

Difficulties associated with receiving telemetry from satellites severely limit the volume of scientific data that can be downlinked to the ground. Current missions employ techniques such as compressing and pruning datasets to reduce the data volume they transmit. While existing mission designs are already restricted by limited telemetry budgets, future Heliophysics System Observatory missions will produce ever larger data volumes with higher resolution and cadence observations from constellations of satellites spread throughout the heliosphere1. In addition, heliophysics missions often produce data for the operational Space Weather community that requires a low latency between observation and downlink. In light of current limitations, the infrastructure to receive NASA satellite telemetry must be expanded and modernized to support the science needs of future data-rich heliophysics missions.

deep space network (DSN)↗
NASA NTRSDate not supplied

System-Theoretic Analysis of Unsafe Collaborative Control in Teaming Systems

The interactions that occur in human-teaming are inspiring novel aerospace designs aimed at improving how humans and machines, or multiple machines, work together. Unfortunately, current Systems Engineering processes are ill-equipped to handle these complex relationships and are unable to design and assure the safety for these systems. To close part of this gap, this paper introduces a novel system-theoretic analytical process to identify unsafe collaborative control actions. It is part of a broader set of techniques that extend the state-of-the-art in hazard analysis, System Theoretic Process Analysis (STPA), to systematically address collaboration. The method rigorously expresses the different ways multiple commands may be unsafe together. Using Systems Theory, it employs abstraction to manage the combinatorial complexity in enumerating control contributions from multiple collaborating components. An algorithm integrates these concepts into an end-to-end process and is supported by automation to enumerate, refine, prune, and prioritize unsafe combinations of control actions. The output of the method feeds the specification of system requirements to implement safety-guided design starting early in concept development. The process is demonstrated on a manned-unmanned aircraft teaming case study and finds new causal factors that were not previously found in a past hazard analysis of the same system.

System Safety↗
NASA NTRS2024

Decision Space Modeling: Trade Space Ontology

As the National Aeronautics and Space Administration (NASA) works to develop a crewed Moon to Mars Architecture, it is dealing with a large decision space consisting of the overlay of human exploration architectures for both the Moon and for Mars. Efforts are underway to enable reasoning, analysis, and deliberation on this decision space. A critical first step is to develop a model of the decision space, which will then allow for various methods and techniques to be applied in support of the larger architecture decision-making process. The Trade Space Ontology consists of a set of terminologies and relations (an ontology) and a MagicDraw resource that enables documentation of decisions and alternatives. It also provides a means by which decisions and alternatives can be traced to other Systems Engineering artifacts. For documenting alternatives, the Trade Space Ontology adapts the Morphological Matrix methodology to The Systems Modeling Language (SysML) through a profile; custom diagrams are also implemented to simplify the profile's use. With the profile and custom diagrams, system architects can specify options for architecture attributes, as well as compatibility between them, in a compact visual format. While the approach shares similarities to a trade tree, the emphasis at this stage is less on enumerating specific combinations of options and instead on specifying the options and their compatibility. Enumeration of alternatives is performed by an external analysis that operates on an output file from a model constructed using the Trade Space Ontology. For decisions, the Trade Space Ontology provides a way to model generic precedence relationships as well as documenting inputs and outputs. These may include what alternatives, criteria, and rationale are understood to be relevant for each decision. Importantly, the decision-making side of the Trade Space Ontology is defined at a more general level, such that it can be adapted to the specific terms in use by projects and programs at NASA. However, this adaptability also means that less capability is provided ``out-of-the-box'' from installation. Currently the resource includes plugin functionality to enumerate paths through generic precedence relationships between decisions and to export these paths to a spreadsheet. Custom dependency stereotypes are included in the profile to indicate the cross-cutting relationships between the trade space and the architecture decisions, providing a means to map which parts of the trade space enumerate alternatives for a decision, and to identify how the output of a decision may modify the trade space through pruning or down-selection. While the motivating use case for this resource is in human exploration architectures, the broad applicability of the Morphological Matrix methodology indicates that the Trade Space Ontology should also be useful for other activities and tasks at the agency.

Trade Tree↗
DOE OSTI2024

autoGEMM: Pushing the Limits of Irregular Matrix Multiplication on Arm Architectures

This paper presents an open-source library that pushes the limits of performance portability for irregular General Matrix Multiplication (GEMM) on the widely-used Arm architectures. Our library, autoGEMM, is designed to support a wide range of Arm processors: from edge devices to HPC-grade CPUs. autoGEMM generates optimized kernels for various hardware configurations by auto-combining fragments of autogenerated micro-kernels that employ hand-written optimizations to maximize computational efficiency. We optimize the kernel pipeline by tuning the register reuse and the data load/store overlapping. In addition, we use a dynamic tiling scheme to generate balanced tile shapes. Finally, we position autoGEMM on top of the TVM framework where our dynamic tiling scheme prunes the search space for TVM to identify the optimal combination of parameters for code optimization. Evaluations on five different classes of Arm chips demonstrate the advantages of autoGEMM. For small matrices, autoGEMM achieves 98% of peak and up to 2.0x speedup over state-of-the-art libraries such as LIBXSMM and LibShalom. For irregular matrices (i.e. tall skinny and long rectangles), autoGEMM is 1.3-2.0x faster than widely-used libraries such as OpenBLAS and Eigen. autoGEMM is publicly available at: https://github.com/wudu98/autoGEMM.

Wu, Du↗
DOE OSTI2025

MoE-Inference-Bench: Performance Evaluation of Mixture of Expert Large Language and Vision Models

Mixture of Experts (MoE) models have enabled the scaling of Large Language Models (LLMs) and Vision Language Models (VLMs) by achieving massive parameter counts while maintaining computational efficiency. However, MoEs introduce several inference-time challenges, including load imbalance across experts and the additional routing computational overhead. To address these challenges and fully harness the benefits of MoE, a systematic evaluation of hardware acceleration techniques is essential. We present MoE-Inference-Bench, a comprehensive study to evaluate MoE performance across diverse scenarios. We analyze the impact of batch size, sequence length, and critical MoE hyperparameters such as FFN dimensions and number of experts on throughput. We evaluate several optimization techniques on Nvidia H100 GPUs, including pruning, Fused MoE operations, speculative decoding, quantization, and various parallelization strategies. Our evaluation includes MoEs from the Mixtral, DeepSeek, OLMoE and Qwen families. The results reveal performance differences across configurations and provide insights for the efficient deployment of MoEs.

Chitty-Venkata, Krishna Teja↗
DOE OSTI2026

Surrogate Neural Architecture Codesign Package (SNAC-Pack)

Neural architecture search (NAS) is a powerful approach for automating model design, but existing methods often optimize for accuracy alone or rely on proxy metrics such as bit operations (BOPs) that correlate poorly with hardware cost. This gap is particularly large for FPGA deployment, where cost is dominated by a multi-dimensional budget of lookup tables, DSPs, flip-flops, BRAM, and latency. We present the Surrogate Neural Architecture Codesign Package (SNAC-Pack), an open-source AutoML framework for hardware-aware neural architecture codesign and end-to-end FPGA deployment. SNAC-Pack runs a multi-objective global search with Optuna and NSGA-II, loading trials to a shared SQLite store that enables parallel workers across compute nodes. A hardware surrogate model outputs per-trial resource and latency estimates, avoiding the synthesis cost that would otherwise dominate the search loop. A local search stage then applies quantization-aware training (QAT) together with iterative magnitude pruning in a combined compression loop, after which the final model is synthesized to FPGA firmware via the hls4ml Python library. A YAML configuration and an optional agentic frontend let users run the pipeline on new datasets without modifying the framework. We demonstrate SNAC-Pack on jet classification at the Large Hadron Collider and superconducting qubit readout, discovering compact architectures that match or exceed strong baselines on the task metric while reducing FPGA resource utilization and, in the qubit readout case, reducing the design space exploration process from months of manual fine-tuning to hours of automated search.

Weitz, Jason [UC, San Diego]↗