Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Learning with errors”

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 1,099 records · Page 61
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

OrbNet Denali: A machine learning potential for biological and organic chemistry with semi-empirical cost and DFT accuracy

We present OrbNet Denali, a machine learning model for an electronic structure that is designed as a drop-in replacement for groundstate density functional theory (DFT) energy calculations. The model is a message-passing graph neural network that uses symmetryadapted atomic orbital features from a low-cost quantum calculation to predict the energy of a molecule. OrbNet Denali is trained on a vast dataset of 2.3 × 10 6 DFT calculations on molecules and geometries. This dataset covers the most common elements in biochemistry and organic chemistry (H, Li, B, C, N, O, F, Na, Mg, Si, P, S, Cl, K, Ca, Br, and I) and charged molecules. OrbNet Denali is demonstrated on several well-established benchmark datasets, and we find that it provides accuracy that is on par with modern DFT methods while offering a speedup of up to three orders of magnitude. For the GMTKN55 benchmark set, OrbNet Denali achieves WTMAD-1 and WTMAD-2 scores of 7.19 and 9.84, on par with modern DFT functionals. For several GMTKN55 subsets, which contain chemical problems that are not present in the training set, OrbNet Denali produces a mean absolute error comparable to those of DFT methods. For the Hutchison conformer benchmark set, OrbNet Denali has a median correlation coefficient of R 2 = 0.90 compared to the reference DLPNO-CCSD(T) calculation and R 2 = 0.97 compared to the method used to generate the training data (ωB97X-D3/def2- TZVP), exceeding the performance of any other method with a similar cost. Similarly, the model reaches chemical accuracy for noncovalent interactions in the S66x10 dataset. For torsional profiles, OrbNet Denali reproduces the torsion profiles of ωB97X-D3/def2-TZVP with an average mean absolute error of 0.12 kcal/mol for the potential energy surfaces of the diverse fragments in the TorsionNet500 dataset.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

A machine learning approach to quantify degradation of nuclear fuels and the effects of fission products

Nuclear fuel performance is critically dependent on understanding the evolution of fuel properties under operational conditions, a complex challenge driven by chemical changes and substantial radiation damage during fission. Traditionally, property evolution has been determined via empirical data collected following irradiation. However, these empirical correlations are limited in their applicability beyond the specific conditions in which they were obtained. This study explores a novel approach to address this challenge by applying materials informatics to develop a machine learning random forest (ML-RF) model that captures the effects of fission products on fuel compounds. The model predicts formation enthalpy (ΔH f ) by leveraging extensive quantum materials property data and correlating it with material descriptors such as composition, atomic and site features, and crystal lattice properties. This ML-RF model enables rapid interpolation across the compositional and structural spaces covered by the training data, thus supporting high-throughput screening and energetic ranking of candidate phases. The model demonstrates the ability to predict ΔH f with a mean absolute error (MAE) of approximately 0.1 to 0.2 eV/atom across a wide range of compounds, including key nuclear fuel systems (U-O, U-N, U-C, U-Si, and U-Mo). For example, it was used to assess shifts in stoichiometry for UO 2 (O/M) and UN (N/M) fuels, revealing their distinct tendencies in chemical potential variation and enabling preliminary convex hull analyses. Furthermore, the model provides insights into how individual fission products affect fuel properties. Results indicate that larger fission products (e.g., Nd, Pu, Ce) have a more pronounced impact on UO 2 , while lighter ones (e.g., Zr) strongly influence UN. Here, the model developed in this work can be used to support the Accelerated Fuel Qualification approach by facilitating preliminary evaluations prior to extensive materials modeling and experimentation. To this end, the trained model has been made available to the fuel community to support ongoing fuel development efforts.

Accelerated fuel qualification↗
DOE OSTI2020

Machine-learning prediction for quasiparton distribution function matrix elements

There have been rapid developments in the direct calculation in lattice QCD (LQCD) of the Bjorken-x dependence of hadron structure through large-momentum effective theory (LaMET). LaMET overcomes the previous limitation of LQCD to moments (that is, integrals over Bjorken x) of hadron structure, allowing LQCD to directly provide the kinematic regions where the experimental values are least known. LaMET requires large-momentum hadron states to minimize its systematics and allow us to reach small- x reliably. This means that very fine lattice spacing to minimize lattice artifacts at order $(P_za)^n$ will become crucial for next-generation LaMET-like structure calculations. Furthermore, such calculations require operators with long Wilson-link displacements, especially in finer lattice units, increasing the communication costs relative to that of the propagator inversion. In this work, we explore whether machine-learning algorithms can make predictions of correlators to reduce the computational cost of these LQCD calculations. We consider two algorithms, gradient-boosting decision tree and linear models, applied to LaMET data, the matrix elements needed to determine the kaon and $η_s$ unpolarized parton distribution functions (PDFs), meson distribution amplitude (DA), and the nucleon gluon PDF. We find that both algorithms can reliably predict the target observables with different prediction accuracy and systematic errors. The predictions from smaller displacement z to larger ones work better than those for momentum p due to the higher correlation among the data.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2025

Multi-modality deep learning for pulse prediction in homogeneous nonlinear systems via parametric conversion

In this Letter, we introduce FusionNet, a multi-modality deep learning framework designed to predict and analyze output pulses in high-power rare-earth-doped laser systems driving parametric conversion in homogeneous guided nonlinear media. FusionNet integrates temporal, spectral, and physical experimental conditions to model ultrafast nonlinear phenomena, including parametric nonlinear frequency conversion, self-phase modulation, and cross-phase modulation in homogeneous guided systems such as gas-filled hollow-core fibers. These systems bridge physical models with experimental data, advancing our understanding of light-guiding principles and nonlinear interactions while expediting the design and optimization of on-demand high-power, high-brightness systems. Our results demonstrate a 73% reduction in prediction error and an 83% improvement in computational efficiency compared to conventional neural networks. This work establishes a new paradigm for accelerating parametric simulations and optimizing experimental designs in high-power laser systems, with further implications for high-precision spectroscopy, quantum information science, and distributed entangled interconnects.

47 OTHER INSTRUMENTATION↗
DOE OSTI2024

Prediction of the Cu oxidation state from EELS and XAS spectra using supervised machine learning

Abstract Electron energy loss spectroscopy (EELS) and X-ray absorption spectroscopy (XAS) provide detailed information about bonding, distributions and locations of atoms, and their coordination numbers and oxidation states. However, analysis of XAS/EELS data often relies on matching an unknown experimental sample to a series of simulated or experimental standard samples. This limits analysis throughput and the ability to extract quantitative information from a sample. In this work, we have trained a random forest model capable of predicting the oxidation state of copper based on its L-edge spectrum. Our model attains an R 2 score of 0.85 and a root mean square error of 0.24 on simulated data. It has also successfully predicted experimental L-edge EELS spectra taken in this work and XAS spectra extracted from the literature. We further demonstrate the utility of this model by predicting simulated and experimental spectra of mixed valence samples generated by this work. This model can be integrated into a real-time EELS/XAS analysis pipeline on mixtures of copper-containing materials of unknown composition and oxidation state. By expanding the training data, this methodology can be extended to data-driven spectral analysis of a broad range of materials.

36 MATERIALS SCIENCE↗
DOE OSTI2020

Fast Scanning Probe Microscopy via Machine Learning: Non-Rectangular Scans with Compressed Sensing and Gaussian Process Optimization

Fast scanning probe microscopy enabled via machine learning allows for a broad range of nanoscale, temporally resolved physics to be uncovered. However, such examples for functional imaging are few in number. Here, using piezoresponse force microscopy (PFM) as a model application, a factor of 5.8 reduction in data collection using a combination of sparse spiral scanning with compressive sensing and Gaussian process regression reconstruction is demonstrated. It is found that even extremely sparse spiral scans offer strong reconstructions with less than 6% error for Gaussian process regression reconstructions. Further, the error associated with each reconstructive technique per reconstruction iteration is analyzed, finding the error is similar past ≈15 iterations, while at initial iterations Gaussian process regression outperforms compressive sensing. Finally, this study highlights the capabilities of reconstruction techniques when applied to sparse data, particularly sparse spiral PFM scans, with broad applications in scanning probe and electron microscopies.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Next generation Arctic vegetation maps: Aboveground plant biomass and woody dominance mapped at 30 m resolution across the tundra biome

The Arctic is warming faster than anywhere else on Earth, placing tundra ecosystems at the forefront of global climate change. Plant biomass is a fundamental ecosystem attribute that is sensitive to changes in climate, closely tied to ecological function, and crucial for constraining ecosystem carbon dynamics. However, the amount, functional composition, and distribution of plant biomass are only coarsely quantified across the Arctic. Therefore, we developed the first moderate resolution (30 m) maps of live aboveground plant biomass (g m −2 ) and woody plant dominance (%) for the Arctic tundra biome, including the mountainous Oro Arctic. We modeled biomass for the year 2020 using a new synthesis dataset of field biomass harvest measurements, Landsat satellite seasonal synthetic composites, ancillary geospatial data, and machine learning models. Additionally, we quantified pixel-wise uncertainty in biomass predictions using Monte Carlo simulations and validated the models using a robust, spatially blocked and nested cross-validation procedure. Observed plant and woody plant biomass values ranged from 0 to ∼6000 g m −2 (mean ≈ 350 g m −2 ), while predicted values ranged from 0 to ∼4000 g m −2 (mean ≈ 275 g m −2 ), resulting in model validation root-mean-squared-error (RMSE) ≈ 400 g m −2 and R 2 ≈ 0.6. Our maps not only capture large-scale patterns of plant biomass and woody plant dominance across the Arctic that are linked to climatic variation (e.g., thawing degree days), but also illustrate how fine-scale patterns are shaped by local surface hydrology, topography, and past disturbance. By providing data on plant biomass across Arctic tundra ecosystems at the highest resolution to date, our maps can significantly advance research and inform decision-making on topics ranging from Arctic vegetation monitoring and wildlife conservation to carbon accounting and land surface modeling.

Climate change↗
DOE OSTI2025

A machine learning method of modern urban building energy modeling: A case study of Chicago

Urban-scale building energy modeling is vital for urban planning. However, it can be challenging to assimilate reliable non-geometry building data for urban-scale modeling without extensive investment. Here, this study introduces a novel approach to developing modern urban-scale building energy stock data using geographic information systems and machine learning algorithms without necessarily requiring pre-supplied non-geometric metadata. The proposed framework integrates building footprint and height data to estimate gross floor areas, and matches each building to a pool of candidate records from ComStock or ResStock—filtered to the same county and ranked by geometric similarity—demonstrate a proof-of-concept case study in Chicago for predicting energy use intensity (EUI) using scalable datasets. The model achieved a mean bias error (MBE) of 0.08 kWh/m² and root mean square error (RMSE) of 14.84 kWh/m² under full metadata input for EUI prediction. With only location inputs, the model captured 69.2 % of EUI within predicted ranges. These results demonstrate the model’s potential to support early-stage urban planning, identify candidates for energy-efficient retrofits. By removing the dependency on detailed pre-surveys or extensive building metadata, the approach overcomes a key barrier in traditional urban-scale building energy modeling, illustrating a pathway toward broader and more cost-effective application, though further multi-city validation and improved treatment of pre-1925 buildings are needed.

Energy Use Intensity↗
DOE OSTI2021

Understanding Growth Dynamics and Yield Prediction of Sorghum Using High Temporal Resolution UAV Imagery Time Series and Machine Learning

Unmanned aerial vehicles (UAV) carrying multispectral cameras are increasingly being used for high-throughput phenotyping (HTP) of above-ground traits of crops to study genetic diversity, resource use efficiency and responses to abiotic or biotic stresses. There is significant unexplored potential for repeated data collection through a field season to reveal information on the rates of growth and provide predictions of the final yield. Generating such information early in the season would create opportunities for more efficient in-depth phenotyping and germplasm selection. This study tested the use of high-resolution time-series imagery (5 or 10 sampling dates) to understand the relationships between growth dynamics, temporal resolution and end-of-season above-ground biomass (AGB) in 869 diverse accessions of highly productive (mean AGB = 23.4 Mg/Ha), photoperiod sensitive sorghum. Canopy surface height (CSM), ground cover (GC), and five common spectral indices were considered as features of the crop phenotype. Spline curve fitting was used to integrate data from single flights into continuous time courses. Random Forest was used to predict end-of-season AGB from aerial imagery, and to identify the most informative variables driving predictions. Improved prediction of end-of-season AGB (RMSE reduction of 0.24 Mg/Ha) was achieved earlier in the growing season (10 to 20 days) by leveraging early- and mid-season measurement of the rate of change of geometric and spectral features. Early in the season, dynamic traits describing the rates of change of CSM and GC predicted end-of-season AGB best. Late in the season, CSM on a given date was the most influential predictor of end-of-season AGB. The power to predict end-of-season AGB was greatest at 50 days after planting, accounting for 63% of variance across this very diverse germplasm collection with modest error (RMSE 1.8 Mg/ha). End-of-season AGB could be predicted equally well when spline fitting was performed on data collected from five flights versus 10 flights over the growing season. This demonstrates a more valuable and efficient approach to using UAVs for HTP, while also proposing strategies to add further value.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2024

Quantifying Streambed Grain Size, Uncertainty, and Hydrobiogeochemical Parameters Using Machine Learning Model YOLO

Abstract Streambed grain sizes control river hydro‐biogeochemical (HBGC) processes and functions. However, measuring their quantities, distributions, and uncertainties is challenging due to the diversity and heterogeneity of natural streams. This work presents a photo‐driven, artificial intelligence (AI)‐enabled, and theory‐based workflow for extracting the quantities, distributions, and uncertainties of streambed grain sizes from photos. Specifically, we first trained You Only Look Once, an object detection AI, using 11,977 grain labels from 36 photos collected from nine different stream environments. We demonstrated its accuracy with a coefficient of determination of 0.98, a Nash–Sutcliffe efficiency of 0.98, and a mean absolute relative error of 6.65% in predicting the median grain size of 20 ground‐truth photos representing nine typical stream environments. The AI is then used to extract the grain size distributions and determine their characteristic grain sizes, including the 10th, 50th, 60th, and 84th percentiles, for 1,999 photos taken at 66 sites within a watershed in the Northwest US. The results indicate that the 10th, median, 60th, and 84th percentiles of the grain sizes follow log‐normal distributions, with most likely values of 2.49, 6.62, 7.68, and 10.78 cm, respectively. The average uncertainties associated with these values are 9.70%, 7.33%, 9.27%, and 11.11%, respectively. These data allow for the computation of the quantities, distributions, and uncertainties of streambed HBGC parameters, including Manning's coefficient, Darcy‐Weisbach friction factor, top layer interstitial velocity magnitude, and nitrate uptake velocity. Additionally, major sources of uncertainty in grain sizes and their impact on HBGC parameters are examined.

58 GEOSCIENCES↗
DOE OSTI2021

Machine learning for neutron reflectometry data analysis of two-layer thin films

Neutron reflectometry (NR) is a powerful tool for probing thin films at length scales down to nanometers. We investigated the use of a neural network to predict a two-layer thin film structure to model a given measured reflectivity curve. Application of this neural network to predict a thin film structure revealed that it was accurate and could provide an excellent starting point for traditional fitting methods. Employing prediction-guided fitting has considerable potential for more rapidly producing a result compared to the labor-intensive but commonly-used approach of trial and error searches prior to refinement. A deeper look at the stability of the predictive power of the neural network against statistical fluctuations of measured reflectivity profiles showed that the predictions are stable. We conclude that the approach presented here can provide valuable assistance to users of NR and should be further extended for use in studies of more complex n-layer thin film systems. This result also opens up the possibility of developing adaptive measurement systems in the future.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Intercomparison of Deep Learning Model Architectures for Atmospheric River Prediction

With a rapid surge in the application of machine learning (ML) for a diverse range of tasks in climate science, the present study addresses a challenge for climate scientists when selecting the optimal ML or deep learning (DL) architecture for a given application. In particular, a DL intercomparison study was performed with a focus on forecasting the position of atmospheric rivers (ARs) on short-range time scales (up to 5-day lead times). AR predictions from multiple DL architectures, including various types of convolutional autoencoders and a vision transformer (ViT), were compared against ECMWF ERA5 reanalysis and hindcasts from a global climate model. DL models with similar trainable parameters were trained on ERA5 reanalysis data and AR positions derived from a thresholding algorithm to ensure a fair comparison among the DL models. Each model’s performance and accuracy in forecasting AR location and key input fields within a 5-day window were assessed using metrics of root-mean-square error, anomaly correlation, and mean intersection over union. The ViT architecture outperformed other autoencoder models in most of the metrics. Incorporating additional meteorological fields only yielded slight improvements in forecasting certain fields at longer lead times. The results also suggest that a smaller number of input time steps or smaller number of autoregressive steps can achieve better prediction skills, while also improving the overall computational efficiency. This research offers valuable insights into the strengths and weaknesses of different DL techniques for AR forecasting, hopefully guiding the development of improved models for forecasting this phenomenon.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

Deep Learning Reconstruction of Daily Soil CO 2 Efflux Reveals Biogeochemical Insights and Reduces Annual Estimate Uncertainty Despite Limited Daily Predictability

Soil CO 2 efflux is commonly measured monthly or seasonally, leaving daily dynamics poorly resolved and contributing to global estimation uncertainty. We trained a single Long Short-Term Memory (LSTM) model to predict daily soil CO 2 efflux across 82 globally distributed sites in COSORE, with 0.2%–46.9% daily data coverage from 2003 to 2020. Despite using far fewer sites than are typically used to train a single deep learning model, with observations biased toward temperate mesic sites, the LSTM model performed well at approximately one-third of sites, reconstructed nearly 2 decades of daily efflux, and outperformed commonly used approaches for estimating daily efflux when applied to the same data set. Performance was weakest at pronounced peaks and troughs and at non-temperate sites with <1.5 years of observations and irregular data patterns. Nevertheless, annual efflux from reconstructed daily data had <40% error even at underperforming sites, substantially improving estimates derived from monthly and seasonal sampling (maximum errors of 95% and 136%, respectively). Temperature sensitivity (Q 10 ) estimated from reconstructed daily predictions closely matched estimates from daily observations, whereas Q 10 values derived from monthly or seasonal observations deviated substantially, suggesting that coarse temporal sampling may contribute to uncertainty in reported Q 10 values. Consistent daily reconstructions further enabled trend analyses for well-performing, predominantly temperate sites and showed increasing soil CO 2 efflux at most sites from 2003 to 2020, with more variable summer trends. Despite limitations, these results demonstrate the potential of LSTM models to reconstruct daily soil CO 2 efflux and reduce estimation uncertainties from sparse observations.

Smykalov, Valerie [Pennsylvania State University, ↗
DOE OSTI2025

Perspectives on Systematic Cloud Microphysics Scheme Development With Machine Learning

Cloud microphysics—the collection of processes that govern the small‐scale formation, evolution, and interactions of liquid droplets and ice crystals in clouds and precipitation—remains a major source of uncertainty in weather and climate models. Although too small in scale to be explicitly resolved in any large‐eddy simulation, weather, or climate model, the representation of cloud microphysical processes has significant impact at the climate scale. Current microphysical schemes are limited by both parametric uncertainty, linked to uncertainty in physical parameter values, and structural uncertainty, arising from incomplete physical understanding of the processes at play or approximations made for computational efficiency. Recent advances in the application of machine learning (ML) to the physical sciences show significant potential for minimizing these limitations by leveraging high‐fidelity simulations and observations. Here we outline the challenges that must be addressed to apply ML toward cloud microphysics scheme development. This perspectives paper synthesizes recent progress in using data‐driven methods, including ML, to improve cloud microphysics parameterizations and highlights opportunities to address key uncertainties. We discuss the roles of aleatoric (irreducible, or statistical) and epistemic (reducible, or systematic) errors in contributing to microphysics parameterization uncertainty. ML can leverage observations to improve microphysical schemes via bottom‐up and top‐down constraints. Methods such as differentiable programming and ML‐enhanced sampling strategies and the creation of large scale benchmark data sets promise to bridge the gap between observations and models and to improve the consistency of cloud microphysical representation across temporal and spatial scales.

Lamb, Kara D. [Columbia Univ., New York, NY (Unite↗
DOE OSTI2024

Machine Learning-Enabled Image Classification for Automated Electron Microscopy

Abstract Traditionally, materials discovery has been driven more by evidence and intuition than by systematic design. However, the advent of “big data” and an exponential increase in computational power have reshaped the landscape. Today, we use simulations, artificial intelligence (AI), and machine learning (ML) to predict materials characteristics, which dramatically accelerates the discovery of novel materials. For instance, combinatorial megalibraries, where millions of distinct nanoparticles are created on a single chip, have spurred the need for automated characterization tools. This paper presents an ML model specifically developed to perform real-time binary classification of grayscale high-angle annular dark-field images of nanoparticles sourced from these megalibraries. Given the high costs associated with downstream processing errors, a primary requirement for our model was to minimize false positives while maintaining efficacy on unseen images. We elaborate on the computational challenges and our solutions, including managing memory constraints, optimizing training time, and utilizing Neural Architecture Search tools. The final model outperformed our expectations, achieving over 95% precision and a weighted F-score of more than 90% on our test data set. This paper discusses the development, challenges, and successful outcomes of this significant advancement in the application of AI and ML to materials discovery.

Materials Science↗
DOE OSTI2024

Application of Modified Meshgraphnets for Subsurface Prediction during CO2 Sequestration

In the face of the increasingly dire consequences of anthropogenic climate change, capturing and storing carbon dioxide is paramount. However, several impediments exist to the safe and effective subsurface storage of CO2, such as cost of transport, identification of suitable sites for subsurface storage, and assessment of long-term risk from storage in subsurface aquifers. Accurate subsurface modeling is necessary to ensure that CO2 storage is both safe and effective. Still, such modeling has traditionally required either substantial time and computational power (numerical simulation) or a substantial amount of pre-existing data for training (machine learning models). Additionally, these models lack flexibility in dealing with both changes in discretization of the input data and generalizability beyond the data on which they are trained. In order to address these issues, this research applies graph neural networks (GNNs) to predict subsurface saturation and pressure during CO₂ injection in a model of the Illinois Basin-Decatur Project (IBDP). GNNs provide a flexible, intuitive method for representing and manipulating complex unstructured data, which is often found in many practical domain problems such as fluid flow and subsurface characterization. These unstructured grids are easily represented in GNNs by representing spatially-localized features such as permeability, porosity, saturation, and pressure as nodes in a graph and relationships between these properties as edges connecting these nodes. This research applies a specific GNN model called MeshGraphNets (MGN) to model the change in CO2 saturation and pressure over a 50-month time period (36 months of injection, 14 months post-injection). The MGN model leverages a message passing process that allows the network to learn both the spatial and temporal dynamics of this system simultaneously. Additionally, training on a limited dataset (64 realizations, 20 time points each) resulted in a high degree of accuracy in saturation prediction both within the same timeframe as the training (20 months, 0.039 average RMSE) and when projecting out to the end of injection (36 months, 0.053 average RMSE). Temporal predictions such as those generated by MGNs and other similar models are prone to accumulated error over time; in order to address this, a multi-step rollout (MSR) training process was applied to calculate training loss. This method mimics the forward prediction during inference by “rolling out” multiple time points in a single training step using the previous prediction as input to the MGN model. By calculating the loss several time steps forward from the current prediction, the model is forced to find a more stable state over time. Application of MSR to the MGN model resulted in an average 15% reduction in inference error over time during forward prediction. This study showcases the immense potential of GNNs as a game-changing methodology for predicting pressure and saturation evolution in CCS projects, ultimately paving the way for more sustainable and effective carbon storage solutions. Presentation prepared for the 2024 AiChE Annual Meeting, October 27 to November 1 2024, San Diego, CA.

Holcomb, Paul↗
DOE OSTI2022

Autonomous System Subversion Tactics: Prototypes and Recommended Countermeasures

One of the fielding requirements for Advanced and Small Modular Reactors (AR/SMR) is the ability to support remote and autonomous operations. Autonomous Control Systems (ACS) are found on platforms such as Autonomous Space Vehicles, Cruise Missiles, and advanced driver-assistance systems. Each of these ACS implementations depends upon a set of decision support subsystems responsible for supporting Autonomous Mission Managers (names vary based upon field and author preferences). These Autonomous Mission Managers receive inputs from system sensors (e.g., LIDAR collection from an automobile travelling down a street; transients from a nuclear reactor), and perform a set of classifications (e.g., Red Traffic Light; Small Pedestrian at 10m; Load Rejection; Single Coolant Pump Trip), and then use these classifications in combination with recommendation algorithms to achieve platform goals (e.g., Stop the Vehicle at the Traffic Light, Avoid the Small Pedestrian; Trip the Reactor to prevent a Safety Event). The design, implementation, and fielding of an ACS capability will alter the cyber-attack surface such that existing risk management plans will need to be updated to include how to protect and defend against data-science and decision-support-system attack classes. These attack classes would include protection of the design and training environments where algorithm selection and testing and training data would be obvious attack vectors. These attack classes would also require an informed set of detection and response procedures to identify anomalous behaviors and document best practices for anomaly assessment and vulnerability mitigation and remediation. Last year we published a Cyber Threat Assessment Methodology for Autonomous and Remote Operations for AR/SMRs along with a companion publication on Cyber Attack and Defense Use Cases. The focus of the methodology was on describing and enumerating ACS processes, components, and functions such that security engineers could: evaluate subversion options against the target; identify threat actor attributes and capabilities derived from each subversion option; and identify security controls and response countermeasures. The Use Cases document offered detailed methodology examples including an assessment of a Military Base SMR, an Autonomous System Decision Loop, and implementation of AR/SMR Machine Learning algorithms. Our proposal at the end of last year was to focus on implementation of subversion prototypes related to the last Use Case area: AR/SMR Machine Learning (ML) Algorithms. We included six attack scenarios in our Use Cases paper: a Poisoning Attack against ML functions implemented using an FPGA; a Trojaning Attack against ML classifiers exploiting the excitability of Nuclear Engineers; a Backdooring Attack against ML Training environments to ensure persistence of an attack vector; a False Positive Evasion Attack against multi-factor Access Control Systems using clever inputs; an Inference Attack against ML models by an Insider with access to the Operational environment; and an Adversarial Reprogramming Attack against a Material Access Control Video Surveillance System. At the beginning of this year these six attack scenarios were provided to our research teams at Georgia Tech and Idaho State University and each team successfully implemented a subversion attack against a ML implementation to include transient misclassifications. While this is a notable outcome from this type of research, this paper offers the reader insight into not only how to structure and execute these types of attacks, but into the thought process behind how the researcher investigated the problem space, performed initial algorithm implementation, and the trial-and-error behind arriving at the successful subversion prototypes. We include in this paper a set of associated Scenarios on how these subversion prototypes could be implemented and an initial set of guidance for AR/SMR architects, Nuclear Regulators, and Cyber Defenders to implement awareness and defense capabilities into their current operational portfolios.

42 ENGINEERING↗
DOE OSTI2024

Complexity-calibrated benchmarks for machine learning reveal when prediction algorithms succeed and mislead

Abstract Recurrent neural networks are used to forecast time series in finance, climate, language, and from many other domains. Reservoir computers are a particularly easily trainable form of recurrent neural network. Recently, a “next-generation” reservoir computer was introduced in which the memory trace involves only a finite number of previous symbols. We explore the inherent limitations of finite-past memory traces in this intriguing proposal. A lower bound from Fano’s inequality shows that, on highly non-Markovian processes generated by large probabilistic state machines, next-generation reservoir computers with reasonably long memory traces have an error probability that is at least $$\sim 60\%$$ ∼ 60 % higher than the minimal attainable error probability in predicting the next observation. More generally, it appears that popular recurrent neural networks fall far short of optimally predicting such complex processes. These results highlight the need for a new generation of optimized recurrent neural network architectures. Alongside this finding, we present concentration-of-measure results for randomly-generated but complex processes. One conclusion is that large probabilistic state machines—specifically, large $$\epsilon$$ ϵ -machines—are key to generating challenging and structurally-unbiased stimuli for ground-truthing recurrent neural network architectures.

97 MATHEMATICS AND COMPUTING↗