SEARCH · Search NASA
Results for “Network Layer”
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.
Autoencoders on FPGAs for real-time, unsupervised new physics detection at 40 MHz at the Large Hadron Collider
In this paper, we show how to adapt and deploy anomaly detection algorithms based on deep autoencoders, for the unsupervised detection of new physics signatures in the extremely challenging environment of a real-time event selection system at the Large Hadron Collider (LHC). We demonstrate that new physics signatures can be enhanced by three orders of magnitude, while staying within the strict latency and resource constraints of a typical LHC event filtering system. This would allow for collecting datasets potentially enriched with high-purity contributions from new physics processes. Through per-layer, highly parallel implementations of network layers, support for autoencoder-specific losses on FPGAs and latent space based inference, we demonstrate that anomaly detection can be performed in as little as $80\,$ns using less than 3% of the logic resources in the Xilinx Virtex VU9P FPGA. Opening the way to real-life applications of this idea during the next data-taking campaign of the LHC.
Integrating deep neural networks with full-waveform inversion: Reparameterization, regularization, and uncertainty quantification
Full-waveform inversion (FWI) is an accurate imaging approach for modeling the velocity structure by minimizing the misfit between recorded and predicted seismic waveforms. However, the strong nonlinearity of FWI resulting from fitting oscillatory waveforms can trap the optimization in local minima. We have adopted a neural-network-based full-waveform inversion (NNFWI) method that integrates deep neural networks with FWI by representing the velocity model with a generative neural network. Neural networks can naturally introduce spatial correlations as regularization to the generated velocity model, which suppresses noise in the gradients and mitigates local minima. Furthermore, the velocity model generated by neural networks is input to the same partial differential equation (PDE) solvers used in conventional FWI. The gradients of the neural networks and PDEs are calculated using automatic differentiation, which back propagates gradients through the acoustic PDEs and neural network layers to update the weights of the generative neural network. Experiments on 1D velocity models, the Marmousi model, and the 2004 BP model determine that NNFWI can mitigate local minima, especially for imaging high-contrast features such as salt bodies, and it significantly improves the inversion in the presence of noise. Adding dropout layers to the neural network model also allows analyzing the uncertainty of the inversion results through Monte Carlo dropout. NNFWI opens a new pathway to combine deep learning and FWI for exploiting the characteristics of deep neural networks and the high accuracy of PDE solvers. Because NNFWI does not require extra training data and optimization loops, it provides an attractive and straightforward alternative to conventional FWI.
Exploring Multilayer Network Models to Build a Scientific Basis for Integrated Deterrence: Final Report
The emerging multipolar international security environment represents a fundamental restructuring of global nuclear balance of power to include two nuclear peer competitors, growing non-peer nuclear threats, and concerns of nuclear latency from both allies and adversaries. Conflicts in the grey zone, cyber operations, mis- and disinformation campaigns, and emerging disruptive technologies like drones, and hypersonic missiles are becoming more prevalent. These present a risk of cross-domain and multi-domain conflicts that may not follow known escalatory patterns. In order to prepare for the new deterrence environment, it is critical to have quantitative and qualitative understandings of these cross-domain conflicts, their potential for escalation, and which systems they may impact. To that end, our team created a Multi-Layer Network (MLN) model of ‘integrated deterrence’ where instruments of national power are modeled as individual network graph layers that include efforts from all domains. We then evaluate the potential for escalation against escalation scenarios. Analysis of the escalation scenarios is then used to identify insights of potential risk and escalation within integrated deterrence.
Deep material network via a quilting strategy: visualization for explainability and recursive training for improved accuracy
Recent developments integrating micromechanics and neural networks offer promising paths for rapid predictions of the response of heterogeneous materials with similar accuracy as direct numerical simulations. The deep material network is one such approaches, featuring a multi-layer network and micromechanics building blocks trained on anisotropic linear elastic properties. Once trained, the network acts as a reduced-order model, which can extrapolate the material’s behavior to more general constitutive laws, including nonlinear behaviors, without the need to be retrained. However, current training methods initialize network parameters randomly, incurring inevitable training and calibration errors. Here, we introduce a way to visualize the network parameters as an analogous unit cell and use this visualization to “quilt” patches of shallower networks to initialize deeper networks for a recursive training strategy. The result is an improvement in the accuracy and calibration performance of the network and an intuitive visual representation of the network for better explainability.
Optimized active layer morphology toward efficient and polymer batch insensitive organic solar cells
Abstract Morphology control in laboratory and industry setting remains as a major challenge for organic solar cells (OSCs) due to the difference in film-drying kinetics between spin coating and the printing process. A two-step sequential deposition method is developed to control the active layer morphology. A conjugated polymer that self-assembles into a well-defined fibril structure is used as the first layer, and then a non-fullerene acceptor is introduced into the fibril mesh as the second layer to form an optimal morphology. A benefit of the combined fibril network morphology and non-fullerene acceptor properties was that a high efficiency of 16.5% (certified as 16.1%) was achieved. The preformed fibril network layer and the sequentially deposited non-fullerene acceptor form a robust morphology that is insensitive to the polymer batches, solving a notorious issue in OSCs. Such progress demonstrates that the utilization of polymer fibril networks in a sequential deposition process is a promising approach towards the fabrication of high-efficiency OSCs.
Intelligent network slicing and policy-based routing engine
One or more aspects of the present disclosure are directed to network optimization solutions provided as software agents (applications) executed on network nodes in a heterogenous multi-vendor environment to provide cross-layer network optimization and ensure availability of network resources to meet associated Quality of Experience (QoE) and Quality of Service (QoS). In one aspect, a network slicing engine is configured to receive at least one request from at least one network endpoint for access to the heterogeneous multi-vendor network for data transmission; receive information on state of operation of a plurality of communication links between the plurality of nodes; determine a set of data transmission routes for the request; assign a network slice for serving the request; determine, from the set of data transmission routes, an end-to-end route for the network slice; and send network traffic associated with the request using the network slice and over the end-to-end route.
Predicting PEMFC performance from a volumetric image of catalyst layer structure using pore network modeling
Not Available
Differentiable Neural Architecture, Mixed Precision and Accelerator Co-Search
Quantization, effective Neural Network architecture, and efficient accelerator hardware are three important design paradigms to maximize accuracy and efficiency. Mixed Precision Quantization is a process of assigning different precision to different Neural Network layers for optimized inference. Neural Architecture Search (NAS) is a process of automatically designing the neural network for a task and can also be extended to search for the precision of each weight and activation matrix. In this paper, we develop the following three methods: (i) Fast Differentiable Hardware-aware Mixed Precision Quantization Search method to find optimal precision, (ii) Joint Differentiable hardware-aware Architecture and Mixed Precision Quantization Co-search, (iii) Joint Accelerator, Architecture, and Precision triple co-search to find best possibilities in all the three worlds. We demonstrate the effectiveness of our proposed methods targeting Bitfusion accelerator by searching mixed precision models on MobilenetV2. We achieve better accuracy-latency trade-off models than the manually designed and previously proposed search methods.
Grain boundary slip transfer classification and metric selection with artificial neural networks
An artificial neural network is used to evaluate the effectiveness of six metrics and their combinations to assess whether slip transfers across grain boundaries in coarse-grained oligocrystalline Al foils. This approach extends the one- or two-dimensional projections formerly applied to analyze slip transfer. The accuracy of this binary classification reaches around 87% for the best single metric and around 90% when considering two or more metrics simultaneously. Here, the results suggest slip transfer mostly depends on the geometric relationship between grains. Training a double-layer network having 10 nodes per hidden layer with 40 measurements is sufficient to render the maximum accuracy.
Stripping away ion hydration shells in electrical double-layer formation: Water networks matter
Significance For centuries the double layer at the solid/electrolyte interface has been a central concept in electrochemistry. Today, it is still crucial for virtually all renewable energy storage and conversion technologies. Here, the double-layer formation is probed by THz spectroscopy with ultrabright synchrotron light as a source. Our results capture the molecular details of double-layer formation at positively/negatively charged Au electrodes for an NaCl electrolyte. We reveal a contrasting response applying positive versus negative bias, which is dictated by the interfacial water network and rationalized by accompanying molecular dynamics simulations and electronic-structure calculations. While Na + is directly attracted toward the negatively charged electrode, stripping of the Cl − hydration shell is observed only at larger potential values.
Deep Learning without Global Optimization by Random Fourier Neural Networks
Here we introduce a new training algorithm for deep neural networks that utilize random complex exponential activation functions. Our approach employs a Markov chain Monte Carlo sampling procedure to iteratively train network layers, avoiding global and gradient-based optimization while maintaining error control. It consistently attains the theoretical approximation rate for residual networks with complex exponential activation functions, determined by network complexity. Additionally, it enables efficient learning of multiscale and high-frequency features, producing interpretable parameter distributions. Despite using sinusoidal basis functions, we do not observe Gibbs phenomena in approximating discontinuous target functions.
Modeling Analog Tile-Based Accelerators Using SST
Analog computing has been widely proposed to improve the energy efficiency of multiple important workloads including neural network operations, and other linear algebra kernels. To properly evaluate analog computing and explore more complex workloads such as systems consisting of multiple analog data paths, system level simulations are required. Moreover, prior work on system architectures for analog computing often rely on custom simulators creating signficant additional design effort and complicating comparisons between different systems. To remedy these issues, this report describes the design and implementation of a flexible tile-based analog accelerator element for the Structural Simulation Toolkit (SST). The element focuses on heavily on the tile controller—an often neglected aspect of prior work—that is sufficiently versatile to simulate a wide range of different tile operations including neural network layers, signal processing kernels, and generic linear algebra operations without major constraints. The tile model also interoperates with existing SST memory and network models to reduce the overall development load and enable future simulation of heterogeneous systems with both conventional digital logic and analog compute tiles. Finally, both the tile and array models are designed to easily support future extensions as new analog operations and applications that can benefit from analog computing are developed.
PyTorch Circuit-Aware Bit-Cell Modeling
SAND2024-08549O PyTorch Circuit-Aware Bit-Cell Modeling, a Python library, demonstrates circuit-aware training for analog machine learning accelerators. Built upon PyTorch, this software provides hardware-aware versions of common neural network layers to mitigate non-ideal behavior from novel hardware. The software provides accurate forward and backward pass estimates for hardware-mapped neural networks that are implemented in crossbars (arrays) that contain a non-linear transistor selector device. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.
NeuroCoreX: An Open-Source FPGA-Based Spiking Neural Network Emulator with On-Chip Learning
Spiking Neural Networks (SNNs) are computational models inspired by the event-driven communication and connectivity patterns of biological neural circuits. They enable high energy efficiency and natural support for diverse architectures ranging from layered networks to small-world and graphstructured topologies. In this work, we introduce NeuroCoreX, an open-source, FPGA-based spiking neural network emulator that provides real-time, on-chip learning and flexible network organization. NeuroCoreX supports both feedforward sensory inputs streamed directly from sensors or PCs via UART and recurrent on-chip connectivity, enabling simultaneous processing and learning from external stimuli and internal network dynamics-capabilities rarely available in existing FPGA SNN platforms. The system implements a Leaky Integrate-and-Fire (LIF) neuron model with current-based synapses and supports pair-based STDP learning on both feedforward and recurrent synapses. A lightweight Python interface enables interactive configuration, live monitoring, weight read-back, and experiment control. Importantly, NeuroCoreX is tightly integrated with the SuperNeuroMAT simulator, allowing SNN models to be transferred seamlessly from software to hardware for hardware-in-the-loop development. By combining real-time plasticity, flexible connectivity, and an open-source VHDL implementation, NeuroCoreX provides an extensible and accessible platform for neuromorphic research, algorithm-hardware co-design, and energy-efficient edge intelligence.
Medial packing and elastic asymmetry stabilize the double-gyroid in block copolymers
Triply-periodic networks are among the most complex and functionally valuable self-assembled morphologies, yet they form in nearly every class of biological and synthetic soft matter building blocks. In contrast to simpler assembly motifs – spheres, cylinders, layers – networks require molecules to occupy variable local environments, confounding attempts to understand their formation. Here, we examine the double-gyroid network phase by using a geometric formulation of the strong stretching theory of block copolymer melts, a prototypical soft self-assembly system. The theory establishes the direct link between molecular packing, assembly thermodynamics and the medial map, a generic measure of the geometric center of complex shapes. We show that “medial packing” is essential for stability of double-gyroid in strongly-segregated melts, reconciling a long-standing contradiction between infinite- and finite-segregation theories. Additionally, we find a previously unrecognized non-monotonic dependence of network stability on the relative entropic elastic stiffness of matrix-forming to tubular-network forming blocks. The composition window of stable double-gyroid widens for both large and small elastic asymmetry, contradicting intuitive notions that packing frustration is localized to the tubular domains. This study demonstrates the utility of optimized medial tessellations for understanding soft-molecular assembly and packing frustration via an approach that is readily generalizable far beyond gyroids in neat block copolymers.
A Taxonomy and Feature set for Server-Side Identification of Proxies
Malicious actors frequently use proxies and VPNs to evade detection and hide their origin. Current challenges to information security include the use of residential proxies to blend in with normal traffic and Man-in-the-Middle phishing proxies that are used to compromise accounts protected with mult-factor authentication. We advance a taxonomy and feature set for the identification of proxied traffic based on the network layer where proxying occurs. We describe how these features apply to common proxy types and how to use these features in the classification of the proxied traffic. Collection of these additional features is feasible using existing network sensors and web servers, while only adding about 30% volume to commonly deployed network sensor logs.
Neural network based emulation of galaxy power spectrum covariances: A reanalysis of BOSS DR12 data
We train neural networks to quickly generate redshift-space galaxy power spectrum covariances from a given parameter set (cosmology and galaxy bias). This covariance emulator utilizes a combination of traditional fully connected network layers and transformer architecture to accurately predict covariance matrices for the high redshift, north galactic cap sample of the BOSS DR12 galaxy catalog. We run simulated likelihood analyses with emulated and brute-force computed covariances, and we quantify the network’s performance via two different metrics: (1) difference in Χ 2 and (2) likelihood contours for simulated BOSS DR 12 analyses. We find that the emulator returns excellent results over a large parameter range. We then use our emulator to perform a reanalysis of the BOSS HighZ NGC galaxy power spectrum, and find that varying covariance with cosmology along with the model vector produces Ω m = $0.276$$^{+0.013}_{–0.015}$, H 0 = 70.2 ± 1.9 km/s/Mpc, and σ 8 = $0.674$$^{+0.058}_{–0.077}$. These constraints represent an average 0.46σ shift in best-fit values and a 5% increase in constraining power compared to fixing the covariance matrix (Ω m = 0.293 ± 0.017, H 0 = 70.3 ± 2.0 km/s/Mpc, σ 8 = $0.702$$^{+0.063}_{–0.075}$). As a result, this work demonstrates that emulators for more complex cosmological quantities than second-order statistics can be trained over a wide parameter range at sufficiently high accuracy to be implemented in realistic likelihood analyses.