Emulating quantum computers with physics-aware neural networks
Poster for DOE/SC/ASCR PI's meeting
Proctor, Timothy James [Sandia National Laboratori↗
SEARCH · Search NASA
Results for “COMPUTING NETWORK”
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.
Model reduction for fractured porous media: a machine learning approach for identifying main flow pathways
Discrete fracture networks (DFN) are often used to model flow and transport in fractured porous media. The accurate resolution of flow and transport behavior on a large DFN involving thousands of fractures is computationally expensive. This makes uncertainty quantification studies of quantities of interest such as travel time through the network computationally intractable, since hundreds to thousands of runs of the DFN model are required to get bounds on the uncertainty of the predictions. Prior works on the subject demonstrated that the complexity of a DFN could be reduced by considering a sub-network of it (often termed a “backbone” sub-network), one whose flow and transport properties were then shown to be similar to that of the full network. The technique is tantamount to partitioning the complete set of fractures of a network into two disjoint sets, one of which is the backbone sub-network while the other its complement. It is in this context that we present a system-reduction technique for DFNs using supervised machine learning via a Random Forest Classifier that selects a backbone sub-network from the full set of fractures. The in-sample errors (in terms of precision and recall scores) of the trained classifier are found to be very accurate indicators of the out-of-sample errors, thus exhibiting that the classifier generalizes well to test data. Moreover, this system-reduction technique yields sub-networks as small as 12% of the full DFN that still recover transport characteristics of the full network such as the peak dosage and tailing behavior for late times. Most importantly, the sub-networks remain connected, and their size can be controlled by a single dimensionless parameter. Furthermore, measures of KL-divergence and KS-statistic for the breakthrough curves of the sub-networks with respect to the full network show physically realistic trends in that the measures decrease monotonically as the size of the sub-networks increase. In conclusion, the computational efficiency gained by this technique depends on the size of the sub-network, but large reductions in computational time can be expected for small sub-networks, yielding as much as 90% computational savings for sub-networks that are as small as 10-12% of the full network.
97 MATHEMATICS AND COMPUTING↗
Towards computational polar-topotronics: Multiscale neural-network quantum molecular dynamics simulations of polar vortex states in SrTiO3/PbTiO3 nanowires
Recent discoveries of polar topological structures ( e.g ., skyrmions and merons) in ferroelectric/paraelectric heterostructures have opened a new field of polar topotronics. However, how complex interplay of photoexcitation, electric field and mechanical strain controls these topological structures remains elusive. To address this challenge, we have developed a computational approach at the nexus of machine learning and first-principles simulations. Our multiscale neural-network quantum molecular dynamics molecular mechanics approach achieves orders-of-magnitude faster computation, while maintaining quantum-mechanical accuracy for atoms within the region of interest. This approach has enabled us to investigate the dynamics of vortex states formed in PbTiO 3 nanowires embedded in SrTiO 3 . We find topological switching of these vortex states to topologically trivial, uniformly polarized states using electric field and trivial domain-wall states using shear strain. These results, along with our earlier results on optical control of polar topology, suggest an exciting new avenue toward opto-electro-mechanical control of ultrafast, ultralow-power polar topotronic devices.
Linker, Thomas↗
A QUBO formulation for top-τ eigencentrality nodes
The efficient calculation of the centrality or “hierarchy” of nodes in a network has gained great relevance in recent years due to the generation of large amounts of data. The eigenvector centrality (aka eigencentrality) is quickly becoming a good metric for centrality due to both its simplicity and fidelity. In this work we lay the foundations for solving the eigencentrality problem of ranking the importance of the nodes of a network with scores from the eigenvector of the network, using quantum computational paradigms such as quantum annealing and gate-based quantum computing. The problem is reformulated as a quadratic unconstrained binary optimization (QUBO) that can be solved on both quantum architectures. The results focus on correctly identifying a given number of the most important nodes in numerous networks given by the sparse vector solution of our QUBO formulation of the problem of identifying the top- τ highest eigencentrality nodes in a network on both the D-Wave and IBM quantum computers.
97 MATHEMATICS AND COMPUTING↗
Neuromorphic overparameterisation and few-shot learning in multilayer physical neural networks
Abstract Physical neuromorphic computing, exploiting the complex dynamics of physical systems, has seen rapid advancements in sophistication and performance. Physical reservoir computing, a subset of neuromorphic computing, faces limitations due to its reliance on single systems. This constrains output dimensionality and dynamic range, limiting performance to a narrow range of tasks. Here, we engineer a suite of nanomagnetic array physical reservoirs and interconnect them in parallel and series to create a multilayer neural network architecture. The output of one reservoir is recorded, scaled and virtually fed as input to the next reservoir. This networked approach increases output dimensionality, internal dynamics and computational performance. We demonstrate that a physical neuromorphic system can achieve an overparameterised state, facilitating meta-learning on small training sets and yielding strong performance across a wide range of tasks. Our approach’s efficacy is further demonstrated through few-shot learning, where the system rapidly adapts to new tasks.
Science & Technology - Other Topics↗
Quantum Gauge Networks: A New Kind of Tensor Network
Although tensor networks are powerful tools for simulating low-dimensional quantum physics, tensor network algorithms are very computationally costly in higher spatial dimensions. We introduce quantum gauge networks: a different kind of tensor network ansatz for which the computation cost of simulations does not explicitly increase for larger spatial dimensions. We take inspiration from the gauge picture of quantum dynamics, which consists of a local wavefunction for each patch of space, with neighboring patches related by unitary connections. A quantum gauge network (QGN) has a similar structure, except the Hilbert space dimensions of the local wavefunctions and connections are truncated. We describe how a QGN can be obtained from a generic wavefunction or matrix product state (MPS). All 2k-point correlation functions of any wavefunction for M many operators can be encoded exactly by a QGN with bond dimension O(M k ). In comparison, for just k = 1, an exponentially larger bond dimension of 2 M/6 is generically required for an MPS of qubits. We provide a simple QGN algorithm for approximate simulations of quantum dynamics in any spatial dimension. The approximate dynamics can achieve exact energy conservation for time-independent Hamiltonians, and spatial symmetries can also be maintained exactly. We benchmark the algorithm by simulating the quantum quench of fermionic Hamiltonians in up to three spatial dimensions.
71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
Reducing Cascading Failure Risk by Increasing Infrastructure Network Interdependence
Increased coupling between critical infrastructure networks, such as power and communication systems, has important implications for the reliability and security of these systems. To understand the effects of power-communication coupling, several researchers have studied models of interdependent networks and reported that increased coupling can increase vulnerability. However, these conclusions come largely from models that have substantially different mechanisms of cascading failure, relative to those found in actual power and communication networks, and that do not capture the benefits of connecting systems with complementary capabilities. In order to understand the importance of these details, this paper compares network vulnerability in simple topological models and in models that more accurately capture the dynamics of cascading in power systems. First, we compare a simple model of topological contagion to a model of cascading in power systems and find that the power grid model shows a higher level of vulnerability, relative to the contagion model. Second, we compare a percolation model of topological cascading in coupled networks to three different models of power networks coupled to communication systems. Again, the more accurate models suggest very different conclusions than the percolation model. In all but the most extreme case, the physics-based power grid models indicate that increased power-communication coupling decreases vulnerability. This is opposite from what one would conclude from the percolation model, in which zero coupling is optimal. Only in an extreme case, in which communication failures immediately cause grid failures, did we find that increased coupling can be harmful. Together, these results suggest design strategies for reducing the risk of cascades in interdependent infrastructure systems.
24 POWER TRANSMISSION AND DISTRIBUTION↗
Vertex finding in neutrino-nucleus interaction: a model architecture comparison
We compare different neural network architectures for machine learning algorithms designed to identify the neutrino interaction vertex position in the MINERvA detector. The architectures developed and optimized by hand are compared with the architectures developed in an automated way using the package “Multi-node Evolutionary Neural Networks for Deep Learning” (MENNDL), developed at Oak Ridge National Laboratory. While the domain-expert hand-tuned network was the best performer, the differences were negligible and the auto-generated networks performed as well. There is always a trade-off between human, and computer resources for network optimization and this work suggests that automated optimization, assuming resources are available, provides a compelling way to save significant expert time.
46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
Adaptive neural network management system
A method and computer system for managing a neural network. Data is sent into an input layer in a portion of layers of nodes in the neural network. The data moves on an encode path through the portion such that an output layer in the portion outputs encoded data. The encoded data is sent into the output layer on a decode path through the portion back to the input layer to obtain a reconstruction of the data by the input layer. A determination is made as to whether an undesired amount of error has occurred in the output layer based on the data sent into the input layer and the reconstruction of the data. A number of new nodes is added to the output layer when a determination is present that the undesired amount of the error occurred, enabling reducing the error using the number of the new nodes.
Draelos, Timothy J.↗
Moving small files in a networked environment
Globally distributed computing infrastructures, such as clouds and supercomputers, are currently used to manage data that is generated with an unprecedented speed from a variety of resources. Coping with this trend, the volume of data exchanged across distant sites increases substantially. To accelerate data transfer, high-speed networks are provided to connect remote sites. Most existing data movement solutions are optimized for moving large files. However, it is still challenging to transfer a large number of small files across networks. This disadvantage not only lowers data transfer performance, but also decreases overall system utilization. Here, we identify that moving small files is mainly constrained by degraded file system throughput, not just network performance as might be suspected. We have built a data transfer pipeline model to analyze the impact of small network I/O and storage I/O on data movement. Extending one of the widely used open source data movement solutions, GridFTP, we demonstrate several appropriate engineering approaches that mitigate the bottleneck and increase data transfer efficiency. We show optimizations that improve data transfer performance more than 5 times. In comparison to existing solutions, our approaches can save a significant amount of system resources for moving lots of small files.
97 MATHEMATICS AND COMPUTING↗
Spike-and-Slab Shrinkage Priors for Structurally Sparse Bayesian Neural Networks
Network complexity and computational efficiency have become increasingly significant aspects of deep learning. Sparse deep learning addresses these challenges by recovering a sparse representation of the underlying target function by reducing heavily overparameterized deep neural networks. Specifically, deep neural architectures compressed via structured sparsity (e.g., node sparsity) provide low-latency inference, higher data throughput, and reduced energy consumption. In this article, we explore two well-established shrinkage techniques, Lasso and Horseshoe, for model compression in Bayesian neural networks (BNNs). To this end, we propose structurally sparse BNNs, which systematically prune excessive nodes with the following: 1) spike-and-slab group Lasso (SS-GL) and 2) SS group Horseshoe (SS-GHS) priors, and develop computationally tractable variational inference, including continuous relaxation of Bernoulli variables. We establish the contraction rates of the variational posterior of our proposed models as a function of the network topology, layerwise node cardinalities, and bounds on the network weights. Furthermore, we empirically demonstrate the competitive performance of our models compared with the baseline models in prediction accuracy, model compression, and inference latency.
97 MATHEMATICS AND COMPUTING↗
Low-Voltage, CMOS-Free Synaptic Memory Based on Li X TiO 2 Redox Transistors
Neuromorphic computers based on analogue neural networks aim to substantially lower computing power by reducing the need to shuttle data between memory and logic units. Artificial synapses containing nonvolatile analogue conductance states enable direct computation using memory elements; however, most nonvolatile analogue memories require high write voltages and large current densities and are accompanied by nonlinear and unpredictable weight updates. Here, we develop an inorganic redox transistor based on electrochemical lithium-ion insertion into Li X TiO 2 that displays linear weight updates at both low current densities and low write voltages. The write voltage, as low as 200 mV at room temperature, is achieved by minimizing the open-circuit voltage and using a low-voltage diffusive memristor selector. We further show that the Li X TiO 2 redox transistor can achieve an extremely sharp transistor subthreshold slope of just 40 mV/decade when operating in an electrochemically driven phase transformation regime.
42 ENGINEERING↗
Cloud Services Enable Efficient AI-Guided Simulation Workflows across Heterogeneous Resources
Applications which fuse machine learning and simulation are rarely best served by a single computing resource. Highly parallel simulation codes are best deployed on super- computers, while AI tasks used to decide which simulations to perform may be best suited to specialized accelerators. Here we present a Function-as-a-Service (FaaS) system for executing complex, distributed computational campaigns that achieves performance parity with conventional workflow systems without the complexities of secure network connections between compute providers. One innovation enabling high performance is a subsystem that directly moves task data between sites, separate from the cloud-hosted FaaS system used to distribute task instructions. We also introduce a flexible scheduling system that allows us access factor of 2 trade offs between the amount of resources required to solve a problem at each compute site. We anticipate that this system will upgrade multi-site applications from demonstration projects to routine practice in computational science.
Ward, Logan↗
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, BrianVan↗
Computing architecture for autonomous microgrids
A computing architecture that facilitates autonomously controlling operations of a microgrid is described herein. A microgrid network includes numerous computing devices that execute intelligent agents, each of which is assigned to a particular entity (load, source, storage device, or switch) in the microgrid. The intelligent agents can execute in accordance with predefined protocols to collectively perform computations that facilitate uninterrupted control of the .
24 POWER TRANSMISSION AND DISTRIBUTION↗
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↗
A Scalable Quantum Cryptography Network for Protected Automation Communication (Final Report)
This is the final report for a CEDS-funded project aimed at developing a new quantum technology for securing utility communication networks used to control and monitor electrical grid equipment. Securing these control networks represents a unique challenge as the performance of the security solution has a direct impact on the stability and reliability of the electrical grid. Traditional, software-based solutions - developed for information networks - are not suitable for utility control networks because they introduce latency, require burdensome maintenance and upgrades, are often incompatible with legacy equipment, and introduce operational complexity that reduces grid reliability. Consequently, many U.S. utilities do not use existing solutions and, instead, protect their critical control networks through the careful isolation and obscuration of their networked equipment. With more utilities embracing grid automation, the attack surface that utilities must defend from hackers has grown to an unmanageable size. To address this situation, Qubitekk and its partners proposed and developed a hardware-based solution that can secure critical control networks without negatively impacting grid performance. This new solution is based on quantum key distribution (QKD) techniques that guarantee secure key generation and distribution across a utility control network. Through deployment and field testing of a prototype QKD system, we have shown that this solution delivers long-term network security, is technically feasible to implement and maintain on a utility’s distribution substation network and does not negatively impact grid operations. In addition, the project has identified and solved key challenges associated with generating, transmitting, and measuring coherent photonic quantum states on a real-world fiber optic network. These additional findings are playing a critical role in advancing quantum networks for quantum computing applications. An overview of the QKD prototype development effort, field testing activities and results, and additional findings relevant to emerging quantum networks are presented in this report.
71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
Choosing experiments to accelerate collective discovery
Scientists perform a tiny subset of all possible experiments. What characterizes the experiments they choose? What are the consequences of those choices for the pace of scientific discovery? We model scientific knowledge as a network and science as a sequence of experiments designed to gradually uncover it. By analyzing millions of biomedical articles published over 30 y, we find that biomedical scientists pursue conservative research strategies exploring the local neighborhood of central, important molecules. Although such strategies probably serve scientific careers, we show that they slow scientific advance, especially in mature fields, where more risk and less redundant experimentation would accelerate discovery of the network. Lastly, we also consider institutional arrangements that could help science pursue these more efficient strategies.
97 MATHEMATICS AND COMPUTING↗