Search NASA⌕ Search

SEARCH · Search NASA

Results for “Computational optimization”

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.

At least 595 records · Page 33

Livermore Big Artificial Neural Network Toolkit

LBANN is a toolkit that is designed to train artificial neural networks efficiently on high performance computing architectures. It is optimized to take advantages of key High Performance Computing features to accelerate neural network training. Specifically it is optimized for low-latency, high bandwidth interconnects, node-local NVRAM, node-local GPU accelerators, and high bandwidth parallel file systems. It is built on top of the open source Elemental distributed-memory dense and spars-direct linear algebra and optimization library that is released under the BSD license. The algorithms contained within LBANN are drawn from the academic literature and implemented to work within a distributed-memory framework.

Essen, Brian Van↗

Black-box optimization of CT acquisition and reconstruction parameters: a reinforcement learning approach

Protocol optimization is critical in Computed Tomography (CT) for achieving desired diagnostic image quality while minimizing radiation dose. Due to the inter-effect of influencing CT parameters, traditional optimization methods rely on the testing of exhaustive combinations of these parameters. This poses a notable limitation due to the impracticality of exhaustive parameter testing. This study introduces a novel methodology leveraging Virtual Imaging Trials (VITs) and reinforcement learning to more efficiently optimize CT protocols. Computational phantoms with liver lesions were imaged using a validated CT simulator and reconstructed with a novel CT reconstruction Toolkit. The optimization parameter space included tube voltage, tube current, reconstruction kernel, slice thickness, and pixel size. The optimization process was done using a Proximal Policy Optimization (PPO) agent which was trained to maximize the Detectability Index (d’) of the liver lesion for each reconstructed image. Results showed that our reinforcement learning approach found the absolute maximum d’ across the test cases while requiring 79.7% fewer steps compared to an exhaustive search, demonstrating both accuracy and computational efficiency, offering a efficient and robust framework for CT protocol optimization. The flexibility of the proposed technique allows for use of varying image quality metrics as the objective metric to maximize for. Our findings highlight the advantages of combining VIT and reinforcement learning for CT protocol management.

Fenwick, David [Duke University Medical Center]↗

3-regular three-XORSAT planted solutions benchmark of classical and quantum heuristic optimizers

With current semiconductor technology reaching its physical limits, special-purpose hardware has emerged as an option to tackle specific computing-intensive challenges. Optimization in the form of solving quadratic unconstrained binary optimization problems, or equivalently Ising spin glasses, has been the focus of several new dedicated hardware platforms. These platforms come in many different flavors, from highly-efficient hardware implementations on digital-logic of established algorithms to proposals of analog hardware implementing new algorithms. In this work, we use a mapping of a specific class of linear equations whose solutions can be found efficiently, to a hard constraint satisfaction problem (three-regular three-XORSAT, or an Ising spin glass) with a 'golf-course' shaped energy landscape, to benchmark several of these different approaches. We perform a scaling and prefactor analysis of the performance of Fujitsu's digital annealer unit (DAU), the D-Wave advantage quantum annealer, a virtual MemComputing machine, Toshiba's simulated bifurcation machine (SBM), the SATonGPU algorithm from Bernaschi et al, and our implementation of parallel tempering. We identify the SATonGPU and DAU as currently having the smallest scaling exponent for this benchmark, with SATonGPU having a small scaling advantage and in addition having by far the smallest prefactor thanks to its use of massive parallelism. Furthermore, our work provides an objective assessment and a snapshot of the promise and limitations of dedicated optimization hardware relative to a particular class of optimization problems.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Decentralized Distributed Proximal Policy Optimization (DD-PPO) for High Performance Computing Scheduling on Multi-User Systems

Resource allocation in High Performance Computing (HPC) environments presents a complex and multifaceted challenge for job scheduling algorithms. Beyond the efficient allocation of system resources, schedulers must account for and optimize multiple performance metrics, including job wait time and system throughput. Traditional heuristic-based scheduling algorithms increasingly struggle and lack the efficiency needed to meet the demands and address the complexity and scale of modern HPC systems. Consequently, recent research efforts have focused on leveraging advancements in Artificial Intelligence (AI) and Deep Learning (DL), particularly Reinforcement Learning (RL), to develop more adaptable and intelligent scheduling strategies. Previous RL-based scheduling approaches have explored a range of algorithms, from Deep Q-Networks (DQN) to Proximal Policy Optimization (PPO), and more recently, hybrid methods that integrate Graph Neural Networks (GNNs) with RL techniques. However, a common limitation across these methods is their reliance on relatively small datasets, with few methods being evaluated using large-scale, multi-million-job trace datasets representative of real-world HPC workloads. Moreover, existing RL schedulers face scalability issues due to centralized policy updates, which hinder training efficiency and performance when applied to large datasets. This study introduces a novel RL-based scheduler utilizing Decentralized Distributed Proximal Policy Optimization (DD-PPO) algorithm, which supports large-scale distributed training across multiple workers without requiring parameter synchronization at every step. By eliminating reliance on centralized updates to a shared policy, the DD-PPO scheduler enhances scalability, training efficiency, and sample utilization. Experimental validation using a large real-world dataset containing over 11.5 million job traces collected from petascale HPC systems over six years assesses the influence of dataset scale on training effectiveness and compares DD-PPO performance to traditional and advanced scheduling approaches. The experimental results demonstrate improved scheduling performance in comparison to both heuristic-based schedulers and existing RL-based scheduling algorithms.

AI↗

A Study of Complex Deep Learning Networks on High Performance, Neuromorphic, and Quantum Computers

Current Deep Learning models use highly optimized convolutional neural networks (CNN) trained on large graphical processing units (GPU)-based computers with a fairly simple layered network topology, i.e., highly connected layers, without intra-layer connections. Complex topologies have been proposed, but are intractable to train on current systems. Building the topologies of the deep learning network requires hand tuning, and implementing the network in hardware is expensive in both cost and power.In this paper, we evaluate deep learning models using three different computing architectures to address these problems: quantum computing to train complex topologies, high performance computing (HPC) to automatically determine network topology, and neuromorphic computing for a low-power hardware implementation. Due to input size limitations of current quantum computers we use the MNIST dataset for our evaluation.The results show the possibility of using the three architectures in tandem to explore complex deep learning networks that are untrainable using a von Neumann architecture. We show that a quantum computer can find high quality values of intra-layer connections and weights, while yielding a tractable time result as the complexity of the network increases; a high performance computer can find optimal layer-based topologies; and a neuromorphic computer can represent the complex topology and weights derived from the other architectures in low power memristive hardware.This represents a new capability that is not feasible with current von Neumann architecture. It potentially enables the ability to solve very complicated problems unsolvable with current computing technologies.

97 MATHEMATICS AND COMPUTING↗

A Study of Complex Deep Learning Networks on High Performance, Neuromorphic, and Quantum Computers

Current Deep Learning models use highly optimized convolutional neural networks (CNN) trained on large graphical processing units (GPU)-based computers with a fairly simple layered network topology, i.e., highly connected layers, without intra-layer connections. Complex topologies have been proposed, but are intractable to train on current systems. Building the topologies of the deep learning network requires hand tuning, and implementing the network in hardware is expensive in both cost and power. In this paper, we evaluate deep learning models using three different computing architectures to address these problems: quantum computing to train complex topologies, high performance computing (HPC) to automatically determine network topology, and neuromorphic computing for a low-power hardware implementation. Due to input size limitations of current quantum computers we use the MNIST dataset for our evaluation. The results show the possibility of using the three architectures in tandem to explore complex deep learning networks that are untrainable using a von Neumann architecture. We show that a quantum computer can find high quality values of intra-layer connections and weights, while yielding a tractable time result as the complexity of the network increases; a high performance computer can find optimal layer-based topologies; and a neuromorphic computer can represent the complex topology and weights derived from the other architectures in low power memristive hardware. This represents a new capability that is not feasible with current von Neumann architecture. It potentially enables the ability to solve very complicated problems unsolvable with current computing technologies.

Potok, Thomas E↗

Demons deformable registration for CBCT-guided procedures in the head and neck: Convergence and accuracy

Purpose: The accuracy and convergence behavior of a variant of the Demons deformable registration algorithm were investigated for use in cone-beam CT (CBCT)-guided procedures of the head and neck. Online use of deformable registration for guidance of therapeutic procedures such as image-guided surgery or radiation therapy places trade-offs on accuracy and computational expense. This work describes a convergence criterion for Demons registration developed to balance these demands; the accuracy of a multiscale Demons implementation using this convergence criterion is quantified in CBCT images of the head and neck. Methods: Using an open-source ''symmetric'' Demons registration algorithm, a convergence criterion based on the change in the deformation field between iterations was developed to advance among multiple levels of a multiscale image pyramid in a manner that optimized accuracy and computation time. The convergence criterion was optimized in cadaver studies involving CBCT images acquired using a surgical C-arm prototype modified for 3D intraoperative imaging. CBCT-to-CBCT registration was performed and accuracy was quantified in terms of the normalized cross-correlation (NCC) and target registration error (TRE). The accuracy and robustness of the algorithm were then tested in clinical CBCT images of ten patients undergoing radiation therapy of the head and neck. Results: The cadaver model allowed optimization of the convergence factor and initial measurements of registration accuracy: Demons registration exhibited TRE=(0.8{+-}0.3) mm and NCC=0.99 in the cadaveric head compared to TRE=(2.6{+-}1.0) mm and NCC=0.93 with rigid registration. Similarly for the patient data, Demons registration gave mean TRE=(1.6{+-}0.9) mm compared to rigid registration TRE=(3.6{+-}1.9) mm, suggesting registration accuracy at or near the voxel size of the patient images (1x1x2 mm{sup 3}). The multiscale implementation based on optimal convergence criteria completed registration in 52 s for the cadaveric head and in an average time of 270 s for the larger FOV patient images. Conclusions: Appropriate selection of convergence and multiscale parameters in Demons registration was shown to reduce computational expense without sacrificing registration performance. For intraoperative CBCT imaging with deformable registration, the ability to perform accurate registration within the stringent time requirements of the operating environment could offer a useful clinical tool allowing integration of preoperative information while accurately reflecting changes in the patient anatomy. Similarly for CBCT-guided radiation therapy, fast accurate deformable registration could further augment high-precision treatment strategies.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Optimization Using Metamodeling in the Context of Integrated Computational Materials Engineering (ICME)

Predictive Design Technologies, LLC (PDT) proposed to employ Integrated Computational Materials Engineering (ICME) tools to help the manufacturing industry in the United States regain the competitive advantage in the global economy. ICME uses computational materials science tools within a holistic system in order to accelerate materials development, improve design optimization, and unify design and manufacturing. With the advent of accurate modeling and simulation along with significant increases in high performance computing (HPC) power, virtual design and manufacturing using ICME tools provide the means to reduce product development time and cost by alleviating costly trial-and-error physical design iterations while improving overall quality and manufacturing efficiency. To reduce the computational cost necessary for the large-scale HPC simulations and to make the methodology accessible for small and medium-sized manufacturers (SMMs), metamodels are employed. Metamodels are approximate models (functional relationships between input and output variables) that can reduce the simulation times by one to two orders of magnitude. In Phase I, PDT, partnered with Mississippi State University (MSU), demonstrated the feasibility of the proposed methodology by employing MSU?s internal state variable (ISV) plasticity-damage model with the help of metamodels to optimize the microstructure-process-property-cost for tube manufacturing processes used by Plymouth Tube Company (PTC), which involves complicated temperature and mechanical loading histories. PDT quantified the microstructure-property relationships for PTC?s SAE J525 electric resistance-welded cold drawn low carbon hydraulic 1010 steel tube manufacturing processes at seven different material states and calibrated the ISV plasticity material parameters to fit experimental tensile stress-strain curves. PDT successfully performed large scale finite element (FE) simulations in an HPC environment using the ISV plasticity model in Abaqus FE analyses of the tube forming, sizing, drawing, welding, and normalizing processes. The simulation results coupled with the manufacturing cost data were used to develop prototype metamodeling (quick response) codes which could be used to predict and optimize the microstructure-process-property-cost relationships. The developed ICME metamodeling toolkits are flexible enough to be applied to other manufacturing processes (e.g. forging, forming, casting, extrusion, rolling, stamping, and welding/joining) and metamodeling codes can run on laptop computers. Based on the work completed in Phase I, in Phase II, PDT proposes to continue to refine the ISV model by correlating and incorporating the uncertainties in the microstructure, mechanical testing, and modeling. Following the model refinement, FE analyses will be simulated and will provide even more realistic predictions as they include an appropriate window of uncertainty. Using the HPC output (FE analyses) as input, the quick-response metamodel codes will more accurately predict and optimize the microstructure-process-property-cost relationships. Furthermore, PDT propose to employ the ICME metamodeling toolkits to help develop a new tube product using entirely new high strength steel. The modeling of the high strength steel manufacturing process will replace the costly and time consuming trial-and-error methods that were used in the tubing industry previously. This simulation-based process prototyping will greatly benefit our industrial partners by opening up new market spaces due to new products with greater capabilities.

36 MATERIALS SCIENCE↗

Parallelized POD-based suboptimal economic model predictive control of a state-constrained Boussinesq approximation

Motivated by an energy efficient building application, we want to optimize a quadratic cost functional subject to the Boussinesq approximation of the Navier-Stokes equations and to bilateral state and control constraints. Since the computation of such an optimal solution is numerically costly, we design an efficient strategy to compute a sub-optimal (but applicationally acceptable) solution with significantly reduced computational effort. We employ an economic Model Predictive Control (MPC) strategy to obtain a feedback control. The MPC sub-problems are based on a linear-quadratic optimal control problem subjected to mixed control and state constraints and a convection-diffusion equation, reduced with proper orthogonal decomposition. Finally, to solve each sub-problem, we apply a primal-dual active set strategy. The method can be fully parallelized, which enables the solution of large problems with real-world parameters.

97 MATHEMATICS AND COMPUTING↗

Convex quadratic relaxations for mixed-integer nonlinear programs in power systems

Here, this paper presents a set of new convex quadratic relaxations for nonlinear and mixed-integer nonlinear programs arising in power systems. The considered models are motivated by hybrid discrete/continuous applications where existing approximations do not provide optimality guarantees. The new relaxations offer computational efficiency along with minimal optimality gaps, providing an interesting alternative to state-of-the-art semidefinite programming relaxations. Finally, three case studies in optimal power flow, optimal transmission switching and capacitor placement demonstrate the benefits of the new relaxations.

42 ENGINEERING↗

Toward developing a computational capability for PEM fuel cell design and optimization.

In this paper, we report the progress made in our project recently funded by the US Department of Energy (DOE) toward developing a computational capability, which includes a two-phase, three-dimensional PEM (polymer electrolyte membrane) fuel cell model and its coupling with DAKOTA (a design and optimization toolkit developed and being enhanced by Sandia National Laboratories). We first present a brief literature survey in which the prominent/notable PEM fuel cell models developed by various researchers or groups are reviewed. Next, we describe the two-phase, three-dimensional PEM fuel cell model being developed, tested, and later validated by experimental data. Results from case studies are presented to illustrate the utility of our comprehensive, integrated cell model. The coupling between the PEM fuel cell model and DAKOTA is briefly discussed. Our efforts in this DOE-funded project are focused on developing a validated computational capability that can be employed for PEM fuel cell design and optimization.

30 DIRECT ENERGY CONVERSION↗

Genetic algorithm applied to the optimization of quantum cascade lasers with second harmonic generation

A computational model for the optimization of the second order optical nonlinearities in GaInAs/AlInAs quantum cascade laser structures is presented. The set of structure parameters that lead to improved device performance was obtained through the implementation of the Genetic Algorithm. In the following step, the linear and second harmonic generation power were calculated by self-consistently solving the system of rate equations for carriers and photons. This rate equation system included both stimulated and simultaneous double photon absorption processes that occur between the levels relevant for second harmonic generation, and material-dependent effective mass, as well as band nonparabolicity, were taken into account. The developed method is general, in the sense that it can be applied to any higher order effect, which requires the photon density equation to be included. Specifically, we have addressed the optimization of the active region of a double quantum well In{sub 0.53}Ga{sub 0.47}As/Al{sub 0.48}In{sub 0.52}As structure and presented its output characteristics.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Solving the power flow equations: a monotone operator approach

The AC power flow equations underlie all operational aspects of power systems. They are solved routinely in operational practice using the Newton-Raphson method and its variants. These methods work well given a good initial “guess” for the solution, which is always available in normal system operations. However, with the increase in levels of intermittent generation, the assumption of a good initial guess always being available is no longer valid. In this paper, we solve this problem using the theory of monotone operators. We show that it is possible to compute (using an offline optimization) a “monotonicity domain” in the space of voltage phasors. Given this domain, there is a simple efficient algorithm that will either find a solution in the domain, or provably certify that no solutions exist in it. We validate the approach on several IEEE test cases and demonstrate that the offline optimization can be performed tractably and the computed “monotonicity domain” includes all practically relevant power flow solutions.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Computationally guided discovery of thermoelectric materials

The potential for advances in thermoelectric materials, and thus solid-state refrigeration and power generation, is immense. Progress so far has been limited by both the breadth and diversity of the chemical space and the serial nature of experimental work. In this Review, we discuss how recent computational advances are revolutionizing our ability to predict electron and phonon transport and scattering, as well as materials dopability, and we examine efficient approaches to calculating critical transport properties across large chemical spaces. When coupled with experimental feedback, these high-throughput approaches can stimulate the discovery of new classes of thermoelectric materials. Within smaller materials subsets, computations can guide the optimal chemical and structural tailoring to enhance materials performance and provide insight into the underlying transport physics. Beyond perfect materials, computations can be used for the rational design of structural and chemical modifications (such as defects, interfaces, dopants and alloys) to provide additional control on transport properties to optimize performance. Through computational predictions for both materials searches and design, a new paradigm in thermoelectric materials discovery is emerging.

36 MATERIALS SCIENCE↗

Evaluating Optimization Strategies for Engine Simulations Using Machine Learning Emulators

This work evaluates different optimization algorithms for computational fluid dynamics (CFD) simulations of engine combustion. Due to the computational expense of CFD simulations, emulators built with machine learning algorithms were used as surrogates for the optimizers. Two types of emulators were used: a Gaussian process (GP) and a weighted variety of machine learning methods called SuperLearner (SL). The emulators were trained using a dataset of 2048 CFD simulations that were run concurrently on a supercomputer. The design of experiments (DOE) for the CFD runs was obtained by perturbing nine input parameters using a Monte-Carlo method. The CFD simulations were of a heavy duty engine running with a low octane gasoline-like fuel at a partially premixed compression ignition mode. Ten optimization algorithms were tested, including types typically used in research applications. Each optimizer was allowed 800 function evaluations and was randomly tested 100 times. The optimizers were evaluated for the median, minimum, and maximum merits obtained in the 100 attempts. Some optimizers required more sequential evaluations, thereby resulting in longer wall clock times to reach an optimum. The best performing optimization methods were particle swarm optimization (PSO), differential evolution (DE), GENOUD (an evolutionary algorithm), and micro-genetic algorithm (GA). These methods found a high median optimum as well as a reasonable minimum optimum of the 100 trials. Moreover, all of these methods were able to operate with less than 100 successive iterations, which reduced the wall clock time required in practice. Two methods were found to be effective but required a much larger number of successive iterations: the DIRECT and MALSCHAINS algorithms. A random search method that completed in a single iteration performed poorly in finding optimum designs but was included to illustrate the limitation of highly concurrent search methods. The last three methods, Nelder–Mead, bound optimization by quadratic approximation (BOBYQA), and constrained optimization by linear approximation (COBYLA), did not perform as well.

Probst, Daniel M.↗

Optimal control of fast and high-fidelity quantum state transfer in spin-1/2 chains

Spin chains are promising candidates for quantum communication and computation. Using quantum optimal control (OC) theory based on the Krotov method, we present a protocol to perform quantum state transfer with fast and high fidelity by only manipulating the boundary spins in a quantum spin-1/2 chain. The achieved speed is about one order of magnitude faster than that is possible in the Lyapunov control case for comparable fidelities. Additionally, it has a fundamental limit for OC beyond which optimization is not possible. The controls are exerted only on the couplings between the boundary spins and their neighbors, so that the scheme has good scalability. We also demonstrate that the resulting OC scheme is robust against disorder in the chain.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Accurate optimization of amino acid form factors for computing small-angle X-ray scattering intensity of atomistic protein structures

Structure modelling via small-angle X-ray scattering (SAXS) data generally requires intensive computations of scattering intensity from any given biomolecular structure, where the accurate evaluation of SAXS profiles using coarse-grained (CG) methods is vital to improve computational efficiency. To date, most CG SAXS computing methods have been based on a single-bead-per-residue approximation but have neglected structural correlations between amino acids. To improve the accuracy of scattering calculations, accurate CG form factors of amino acids are now derived using a rigorous optimization strategy, termed electron-density matching (EDM), to best fit electron-density distributions of protein structures. This EDM method is compared with and tested against other CG SAXS computing methods, and the resulting CG SAXS profiles from EDM agree better with all-atom theoretical SAXS data. By including the protein hydration shell represented by explicit CG water molecules and the correction of protein excluded volume, the developed CG form factors also reproduce the selected experimental SAXS profiles with very small deviations. Taken together, these EDM-derived CG form factors present an accurate and efficient computational approach for SAXS computing, especially when higher molecular details (represented by the q range of the SAXS data) become necessary for effective structure modelling.

Tong, Dudu↗