Stein variational inference and sparsification for uncertainty quantification of physical neural networks
Explore the source record for details and available documents.
Jones, Reese Edward [Sandia National Laboratories ↗
SEARCH · Engineering Papers
Results for “Network inference”
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.
Prediction of electric and magnetic fields from spectral data using machine learning algorithms for Doppler-free saturation spectroscopy diagnostics
The prediction of electric and magnetic field amplitudes from atomic spectral data is critical for plasma control in fusion devices such as tokamaks. Conventional approaches that rely on physics-based models are computationally expensive and unsuitable for real-time applications. In this work, we develop and benchmark three machine learning algorithms—simulation-based inference (SBI), fully connected neural networks (FCNN), and histogram-based gradient boosting regression (GBR-Hist)—to infer field intensities directly from Doppler-free saturation spectroscopy (DFSS) spectra. Synthetic datasets of spectra were generated using the EZSSS code and evaluated both with and without added Poisson noise to mimic experimental conditions. We find that SBI achieves the highest accuracy and robustness, FCNN provides a strong balance of accuracy and computational efficiency for real-time applications, and GBR-Hist offers the fastest inference but is more sensitive to noise. Furthermore, these results demonstrate the potential of machine learning to accelerate DFSS analysis and enhance its utility for plasma diagnostics and control.
Doppler-free saturation spectroscopy↗
Uncertainty Quantification for Neutron Shield Using Convolutional Neural Networks
Uncertainty quantification from radiation transport calculations was conducted using a Bayesian inference approach. A surrogate model, using a convolutional neural network, was employed to emulate the neutron fluence, which was simulated with a Monte Carlo radiation transport model. This allowed for a computationally cheap approach to evaluate input parameters and to sample their corresponding posterior probability distributions. Experimental data from the literature were employed to perform uncertainty quantification studies for concrete shields. As a result, the method is a nonintrusive approach that enables studies with multiple input parameters and can be applied to any radiation transport model.
Bayesian inference↗
Deep inference of simulated strong lenses in ground-based surveys
The large number of strong lenses discoverable in future astronomical surveys will likely enhance the value of strong gravitational lensing as a cosmic probe of dark energy and dark matter. However, leveraging the increased statistical power of such large samples will require further development of automated lens modeling techniques. We show that deep learning and simulation-based inference (SBI) methods produce informative and reliable estimates of parameter posteriors for strong lensing systems in ground-based surveys. We present the examination and comparison of two approaches to lens parameter estimation for strong galaxy-galaxy lenses — Neural Posterior Estimation (NPE) and Bayesian Neural Networks (BNNs). We perform inference on 1-, 5-, and 12-parameter lens models for ground-based imaging data that mimics the Dark Energy Survey (DES). We find that NPE outperforms BNNs, producing posterior distributions that are more accurate, precise, and well-calibrated for most parameters. For the 12-parameter NPE model, the calibration is consistently within <10% of optimal calibration for all parameters, while the BNN is rarely within 20% of optimal calibration for any of the parameters. Similarly, residuals for most of the parameters are smaller (by up to an order of magnitude) with the NPE model than the BNN model. This work takes important steps in the systematic comparison of methods for different levels of model complexity.
46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
Industrial PLC Network Modeling and Parameter Identification Using Sensitivity Analysis and Mean Field Variational Inference
A multiple input multiple output (MIMO) power line communication (PLC) model for industrial facilities was developed that uses the physics of a bottom-up model but can be calibrated like top-down models. The PLC model considers 4-conductor cables (three-phase conductors and a ground conductor) and has several load types, including motor loads. The model is calibrated to data using mean field variational inference with a sensitivity analysis to reduce the parameter space. The results show that the inference method can accurately identify many of the model parameters, and the model is accurate even when the network is modified.
47 OTHER INSTRUMENTATION↗
Quantification of neural networks uncertainties with applications to SAFARI-1 axial neutron flux profiles
Deep Neural Networks (DNNs) have been widely used as a data-driven modelling tool in nuclear engineering. However, as a Machine Learning model, Artificial Neural Network (ANN) predictions are subjected to uncertainties originating from the noise in training data, incomplete coverage of the domain, and imperfect neural network architectures. In this work, we target at quantifying the prediction/approximation uncertainties of ANNs using Monte Carlo Dropout (MCD), as well as Bayesian Neural Networks (BNNs) which are solved by variational inference. With a demonstration problem in which neural networks are used to predict the assembly axial neutron flux profiles, the results have shown that the three different neural network models (regular DNNs, DNNs solved with MCD and BNNs) can produce results that agree very well among each other and with the measurement data, on cycles that are not used in the training process. Besides the excellent generalization capability, the uncertainty bands produced by MCD and BNN agree very well, and in general, they can fully envelop the noisy measurement data points. (authors)
21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗
CONCURRENT, CONDENSED STEIN VARIATIONAL GRADIENT DESCENT FOR UNCERTAINTY QUANTIFICATION OF NEURAL NETWORKS
In this work, we propose a Stein variational gradient descent (SVGD) method to concurrently sparsify, train, and provide uncertainty quantification (UQ) of a complexly parameterized model, such as a neural network (NN). It employs a graph reconciliation and condensation process to reduce complexity and increase similarity in the Stein ensemble of parameterizations. Therefore, the proposed concurrent, condensed SVGD (ccSVGD) method can provide UQ on parameters, not just outputs. Furthermore, the parameter reduction speeds up the convergence of the Stein gradient descent as it reduces the combinatorial complexity by aligning and differentiating the sensitivity to parameters. These properties are demonstrated with an illustrative example and an application to a mechanical response representation problem in solid mechanics.
42 ENGINEERING↗
Genetic program activity delineates risk, relapse, and therapy responsiveness in multiple myeloma
Despite recent advancements in the treatment of multiple myeloma (MM), nearly all patients ultimately relapse and many become refractory to multiple lines of therapies. Therefore, we not only need the ability to predict which patients are at high risk for disease progression but also a means to understand the mechanisms underlying their risk. Here, we report a transcriptional regulatory network (TRN) for MM inferred from cross-sectional multi-omics data from 881 patients that predicts how 124 chromosomal abnormalities and somatic mutations causally perturb 392 transcription regulators of 8549 genes to manifest in distinct clinical phenotypes and outcomes. We identified 141 genetic programs whose activity profiles stratify patients into 25 distinct transcriptional states and proved to be more predictive of outcomes than did mutations. The coherence of these programs and accuracy of our network-based risk prediction was validated in two independent datasets. We observed subtype-specific vulnerabilities to interventions with existing drugs and revealed plausible mechanisms for relapse, including the establishment of an immunosuppressive microenvironment. Investigation of the t(4;14) clinical subtype using the TRN revealed that 16% of these patients exhibit an extreme-risk combination of genetic programs (median progression-free survival of 5 months) that create a distinct phenotype with targetable genes and pathways.
59 BASIC BIOLOGICAL SCIENCES↗
Can Simple Machine Learning Tools Extend and Improve Temperature-Based Methods to Infer Streambed Flux?
Temperature-based methods have been developed to infer 1D vertical exchange flux between a stream and the subsurface. Current analyses rely on fitting physically based analytical and numerical models to temperature time series measured at multiple depths to infer daily average flux. These methods have seen wide use in hydrologic science despite strong simplifying assumptions including a lack of consideration of model structural error or the impacts of multidimensional flow or the impacts of transient streambed hydraulic properties. We performed a “perfect-model experiment” investigation to examine whether regression trees, with and without gradient boosting, can extract sufficient information from model-generated subsurface temperature time series, with and without added measurement error, to infer the corresponding exchange flux time series at the streambed surface. Using model-generated, synthetic data allowed us to assess the basic limitations to the use of machine learning; further examination of real data is only warranted if the method can be shown to perform well under these ideal conditions. We also examined whether the inherent feature importance analyses of tree-based machine learning methods can be used to optimize monitoring networks for exchange flux inference.
54 ENVIRONMENTAL SCIENCES↗
Direct inference of nuclear equation-of-state parameters from gravitational-wave observations
The observation of neutron star mergers with gravitational waves (GWs) has provided a new method to constrain the dense-matter equation of state (EOS) and to better understand its nuclear physics. However, inferring nuclear microphysics from GW observations necessitates the sampling of EOS model parameters that serve as input for each EOS used during the GW data analysis. The sampling of the EOS parameters requires solving the Tolman–Oppenheimer–Volkoff (TOV) equations a large number of times—a process that slows down each likelihood evaluation in the analysis on the order of a few seconds. Here, we employ emulators for the TOV equations built using multilayer perceptron neural networks to enable direct inference of nuclear EOS parameters from GW strain data. Our emulators allow us to rapidly solve the TOV equations, taking in EOS parameters and outputting the associated tidal deformability of a neutron star in only a few tens of milliseconds. We implement these emulators in PyCBC to directly infer the EOS parameters using the event GW170817, providing posteriors on these parameters informed solely by GWs. We benchmark these runs against analyses performed using the full TOV solver and find that the emulators achieve speed ups of nearly two orders of magnitude, with negligible differences in the recovered posteriors. Additionally, we constrain the slope and curvature of the symmetry energy at the 90% upper credible interval to be $L$ sym ≲ 106 MeV and $K$ sym ≲ 26 MeV.
73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
Advanced data analysis in inertial confinement fusion and high energy density physics
Bayesian analysis enables flexible and rigorous definition of statistical model assumptions with well-characterized propagation of uncertainties and resulting inferences for single-shot, repeated, or even cross-platform data. This approach has a strong history of application to a variety of problems in physical sciences ranging from inference of particle mass from multi-source high-energy particle data to analysis of black-hole characteristics from gravitational wave observations. The recent adoption of Bayesian statistics for analysis and design of high-energy density physics (HEDP) and inertial confinement fusion (ICF) experiments has provided invaluable gains in expert understanding and experiment performance. In this Review, we discuss the basic theory and practical application of the Bayesian statistics framework. We highlight a variety of studies from the HEDP and ICF literature, demonstrating the power of this technique. Due to the computational complexity of multi-physics models needed to analyze HEDP and ICF experiments, Bayesian inference is often not computationally tractable. Two sections are devoted to a review of statistical approximations, efficient inference algorithms, and data-driven methods, such as deep-learning and dimensionality reduction, which play a significant role in enabling use of the Bayesian framework. We provide additional discussion of various applications of Bayesian and machine learning methods that appear to be sparse in the HEDP and ICF literature constituting possible next steps for the community. We conclude by highlighting community needs, the resolution of which will improve trust in data-driven methods that have proven critical for accelerating the design and discovery cycle in many application areas.
47 OTHER INSTRUMENTATION↗
Towards Compact Neural Networks via End-to-End Training: A Bayesian Tensor Approach with Automatic Rank Determination
Post-training model compression can reduce the inference costs of deep neural networks, but uncompressed training still consumes enormous hardware resources and energy. To enable low-energy training on edge devices, it is highly desirable to directly train a compact neural network from scratch with a low memory cost. Low-rank tensor decomposition is an effective approach to reduce the memory and computing costs of large neural networks. However, directly training low-rank tensorized neural networks is a very challenging task because it is hard to determine a proper tensor rank a priori, and the tensor rank controls both model complexity and accuracy. Here, this paper presents a novel end-to-end framework for low-rank tensorized training. We first develop a Bayesian model that supports various low-rank tensor formats (e.g., CANDECOMP/PARAFAC, Tucker, tensor-train, and tensor-train matrix) and reduces neural network parameters with automatic rank determination during training. Then we develop a customized Bayesian solver to train large-scale tensorized neural networks. Our training methods shows orders-of-magnitude parameter reduction and little accuracy loss (or even better accuracy) in the experiments. On a very large deep learning recommendation system with over 4.2 ×10 9 model parameters, our method can reduce the parameter number to 1.6 ×10 5 automatically in the training process (i.e., by 2.6 ×10 4 times) while achieving almost the same accuracy. Code is available at https://github.com/colehawkins/bayesian-tensor-rank-determination.
compact neural networks↗
Toward real-time optimization through model reduction and model discrepancy sensitivities
Optimization problems arise in a range of scenarios, from optimal control to model parameter estimation. In many applications, such as the development of digital twins, it is essential to solve these optimization problems within wall-clock-time limitations. However, this is often unattainable for complex systems, such as those modeled by nonlinear partial differential equations. One strategy for mitigating this issue is to construct a reduced-order model (ROM) that enables more rapid optimization. In particular, the use of nonintrusive ROMs—those that do not require access to the full-order model at evaluation time—is popular because they facilitate the computation of optimization solutions within the wall-clock time requirements. However, the optimization solution will be unreliable if the iterates move outside the ROM training data. This article proposes the use of hyper-differential sensitivity analysis with respect to model discrepancy (HDSA-MD) as a computationally efficient tool to augment ROM-constrained optimization and improve its reliability. The proposed approach consists of two phases: (i) an offline phase where several full-order model evaluations are computed to train the ROM, and (ii) an online phase where a ROM-constrained optimization problem is solved, a limited number of full-order model evaluations are computed, and HDSA-MD is used to enhance the optimization solution. Numerical results are demonstrated for two examples, atmospheric contaminant control and wildfire ignition location estimation, in which a ROM is trained offline using inaccurate atmospheric data. In conclusion, the HDSA-MD update yields a significant improvement in the ROM-constrained optimization solution using only one full-order model evaluation online with corrected atmospheric data.
PDE-constrained optimization↗
Neural posterior unfolding
Differential cross section measurements are the currency of scientific exchange in particle and nuclear physics. A key challenge for these analyses is the correction for detector distortions, known as deconvolution or unfolding. Binned unfolding of cross section measurements traditionally rely on the regularized inversion of the response matrix that represents the detector response, mapping pre-detector (`particle level') observables to post-detector (`detector level') observables. In this paper we introduce Neural Posterior Unfolding, a modern, Bayesian approach that leverages normalizing flows for unfolding. By using normalizing flows for neural posterior estimation, NPU offers several key advantages including implicit regularization through the neural network architecture, fast amortized inference that eliminates the need for repeated retraining, and direct access to the full uncertainty in the unfolded result. In addition to introducing NPU, we implement a classical Bayesian unfolding method called Fully Bayesian Unfolding (FBU) in modern Python so it can also be studied. These tools are validated on simple Gaussian examples and then tested on simulated jet substructure examples from the Large Hadron Collider (LHC). We find that the Bayesian methods are effective and worth additional development to be analysis ready for cross section measurements at the LHC and beyond.
Analysis and statistical methods↗
Calibrating Bayesian generative machine learning for Bayesiamplification
Recently, combinations of generative and Bayesian deep learning have been introduced in particle physics for both fast detector simulation and inference tasks. These neural networks aim to quantify the uncertainty on the generated distribution originating from limited training statistics. The interpretation of a distribution-wide uncertainty however remains ill-defined. We show a clear scheme for quantifying the calibration of Bayesian generative machine learning models. For a Continuous Normalizing Flow applied to a low-dimensional toy example, we evaluate the calibration of Bayesian uncertainties from either a mean-field Gaussian weight posterior, or Monte Carlo sampling network weights, to gauge their behaviour on unsteady distribution edges. Well calibrated uncertainties can then be used to roughly estimate the number of uncorrelated truth samples that are equivalent to the generated sample and clearly indicate data amplification for smooth features of the distribution.
97 MATHEMATICS AND COMPUTING↗
Energy Distribution of the Galactic Center Excess’s Sources
The Galactic Center Excess (GCE) may yet herald the discovery of annihilating dark matter. Weighing against that conclusion are analyses showing evidence for dim point sources within the spatial structure of the emission. Because of technical limitations these analyses are purely spatial with all spectral information that could disentangle the excess from astrophysical backgrounds discarded. Here, we demonstrate that a neural network simulation-based inference approach can jointly analyze the spatial and spectra data. The addition is profound: energy information drives the putative point sources to be significantly dimmer, indicating either the GCE is truly diffuse in nature or made of an exceptionally large number of sources. Quantitatively, for our best fit background model, the excess is essentially consistent with Poisson emission as predicted by dark matter. If due to point sources, our median prediction is O(10^{5}) sources, or more than 35 000 at 90% confidence-both orders of magnitude larger than the hundreds preferred by earlier point-source analyses of the GCE, although variations allowed by background systematics could reduce the required number of sources by roughly an order of magnitude.
List, Florian↗
MARIAH PCAP data for Validation Demonstration
This dataset holds simulated PCAP (packet capture) data from the SCEPTRE validation demonstration model as a set of pairwise communications between devices via specific protocols. All connections should be assumed to be symmetric, as this data is an aggregation of the true PCAP. A mapping is also provided associating each IP address with its true device type.
cyber-physical system↗
Prediction of laser beam spatial profiles in a high-energy laser facility by use of deep learning
We adapt the significant advances achieved recently in the field of generative artificial intelligence/machine-learning to laser performance modeling in multipass, high-energy laser systems with application to high-shot-rate facilities relevant to inertial fusion energy. Advantages of neural-network architectures include rapid prediction capability, data-driven processing, and the possibility to implement such architectures within future low-latency, low-power consumption photonic networks. Four models were investigated that differed in their generator loss functions and utilized the U-Net encoder/decoder architecture with either a reconstruction loss alone or combined with an adversarial network loss. We achieved inference times of 1.3 ms for a 256 × 256 pixel near-field beam with errors in predicted energy of the order of 1% over most of the energy range. It is shown that prediction errors are significantly reduced by ensemble averaging the models with different weight initializations. These results suggest that including the temporal dimension in such models may provide accurate, real-time spatiotemporal predictions of laser performance in high-shot-rate laser systems.
47 OTHER INSTRUMENTATION↗