SEARCH · Search NASA

Results for “predictor”

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.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 487 records · Page 27

All sources NASA NTRS DOE OSTI

DOE OSTI2024

Structural phase diagram for Sm-substituted BiFeO 3 multiferroics

The structural evolution of Sm substituted BiFe⁢O 3 is studied by total x-ray scattering and structure modeling. It is shown that the crystal structure changes from polar to antipolar and then to nonpolar when the Sm to Bi ratio in the material approaches 20% and 40%, respectively, with no intermixing between the structure types. The evolution is driven by lattice strain induced by the difference in the size of Sm and Bi atoms, leading to changes in the pattern of octahedral tilts and Bi off-centering, which, in turn, induce changes in the multiferroic properties. Furthermore, the substitution ratio at which the different structure types emerge appears to be tied up with the average radius of the atomic species occupying the Bi sites in the perovskite lattice and volume occupied by a formula unit, rendering both quantities useful predictor variables for guiding computational searches for substituted BiFe⁢O 3 multiferroics with improved functional properties.

Ferroelectricity↗

DOE OSTI2025

Phase-field model of alloy solidification far from chemical equilibrium at the solid-liquid interface

We further develop a recently introduced phase-field model of far-from-equilibrium alloy solidification under additive manufacturing conditions [K. Ji et al., Phys. Rev. Lett. 130, 026203 (2023)]. This model utilizes enhanced solute diffusivity within the spatially diffuse interface region to quantitatively capture solute trapping with a larger interface width, thereby making simulations on experimentally relevant length and timescales computationally feasible. The main developments presented here include testing the robustness of different variational formulations, extending the model to concentrated alloys by incorporating solid and liquid free energies from thermodynamic databases, as illustrated for hypoeutectic Al-Ag alloys with CALPHAD, extending convergence tests as a function of interface width to 3D, and carrying out simulations in both 2D and 3D to examine existing theories of microstructure development. Our results indicate that the simplest variational formulation that interpolates the bulk free-energy density between its solid and liquid forms is the most robust. Remarkably, for hypoeutectic Al-Ag alloys, this formulation yields a high-velocity nonequilibrium phase diagram that is independent of interface width, thereby demonstrating that the framework of enhanced solute diffusivity can be nontrivially extended to concentrated alloys. Other variational formulations have restricted ranges of materials or processing parameters that can be reliably modeled. We use 2D simulations to construct high-velocity microstructure selection maps for dilute Al-Cu alloys. The results validate the important role of latent heat rejection at the interface and extend the limited predictions of linear stability analysis [A. Karma and A. Sarkissian, Phys. Rev. E 47, 513 (1993)] and sharp-interface 1D simulations to fully nonlinear regimes. Furthermore, 3D simulations, carried out using a computationally tractable axisymmetric cellular/dendritic interface shape, demonstrate a good convergence similar to that observed in 2D as a function of interface width. Full 3D simulations, in turn, reveal that the standard theory of absolute stability is a good predictor of the upper critical velocity beyond which steady-state growth becomes unstable, despite the different morphological manifestations of this instability in 2D and 3D.

36 MATERIALS SCIENCE↗

DOE OSTI2026

The Use of Machine Learning Models for Predicting the Dielectric Strength of Gases

Technological advancements in high voltage systems have pushed sulfur hexafluoride (SF6) to its operational limits. Furthermore, this gas has other drawbacks including a high liquefaction temperature and a high global warming potential. Therefore, there has been an urgent need to find alternative gases with high dielectric strength (DS). In this work, density functional theory (DFT) is used to calculate molecular descriptors that are fed into an artificial neural network (ANN) and a random forest (RF). These machine learning (ML) models are then used to predict the DS for hundreds of molecules. A finite element model (FEM) is also used to calculate the electric field profile of multiple simple electrode geometries as the applied voltage to the system is increased. Results indicate that the random forest model has better generalization to unseen data than the neural network. The highest DS value predicted by the RF was 2.16 relative to the experimental DS of SF6. The results also demonstrate how choosing a gas with a higher DS and a geometry with minimal edges and corners can significantly increase the operating voltage of an electrical system. Due to its superior generalization, the RF represents the most promising path toward an accurate DS predictor once sufficient experimental data are available.

Mileski, Matthew [AFIT]↗

DOE OSTI2025

Computational Performance Bounds Prediction in Quantum Computing With Unstable Noise

Quantum computing has significantly advanced in recent years, boasting devices with hundreds of quantum bits (qubits), hinting at its potential quantum advantage over classical computing. Yet, noise in quantum devices poses significant barriers to realizing this supremacy. Understanding noise’s impact is crucial for reproducibility and application reuse; moreover, the next-generation quantum-centric supercomputing essentially requires efficient and accurate noise characterization to support system management (e.g., job scheduling), where ensuring correct functional performance (i.e., fidelity) of jobs on available quantum devices can even be higher-priority than traditional objectives. However, noise fluctuates over time, even on the same quantum device, which makes predicting the computational bounds for on-the-fly noise is vital. Noisy quantum simulation can offer insights but faces efficiency and scalability issues. Here, in this work, we propose a data-driven workflow, namely QuBound, to predict computational performance bounds. It decomposes historical performance traces to isolate noise sources and devises a novel encoder to embed circuit and noise information processed by a Long Short-Term Memory (LSTM) network. For evaluation, we compare QuBound with a state-of-the-art learning-based predictor, which only generates a single performance value instead of a bound. Experimental results show that the result of the existing approach falls outside of performance bounds, while all predictions from our QuBound with the assistance of performance decomposition better fit the bounds. Moreover, QuBound can efficiently produce practical bounds for various circuits with over 106 speedup over simulation; in addition, the range from QuBound is over 10× narrower than the state-of-the-art analytical approach.

Li, Jinyang [George Mason Univ., Fairfax, VA (Unit↗

DOE OSTI2024

Climate, food and humans predict communities of mammals in the United States

Abstract Aim The assembly of species into communities and ecoregions is the result of interacting factors that affect plant and animal distribution and abundance at biogeographic scales. Here, we empirically derive ecoregions for mammals to test whether human disturbance has become more important than climate and habitat resources in structuring communities. Location Conterminous United States. Time Period 2010–2021. Major Taxa Studied Twenty‐five species of mammals. Methods We analysed data from 25 mammal species recorded by camera traps at 6645 locations across the conterminous United States in a joint modelling framework to estimate relative abundance of each species. We then used a clustering analysis to describe 8 broad and 16 narrow mammal communities. Results Climate was the most important predictor of mammal abundance overall, while human population density and agriculture were less important, with mixed effects across species. Seed production by forests also predicted mammal abundance, especially hard‐mast tree species. The mammal community maps are similar to those of plants, with an east–west split driven by different dominant species of deer and squirrels. Communities vary along gradients of temperature in the east and precipitation in the west. Most fine‐scale mammal community boundaries aligned with established plant ecoregions and were distinguished by the presence of regional specialists or shifts in relative abundance of widespread species. Maps of potential ecosystem services provided by these communities suggest high herbivory in the Rocky Mountains and eastern forests, high invertebrate predation in the subtropical south and greater predation pressure on large vertebrates in the west. Main Conclusions Our results highlight the importance of climate to modern mammals and suggest that climate change will have strong impacts on these communities. Our new empirical approach to recognizing ecoregions has potential to be applied to expanded communities of mammals or other taxa.

Kays, Roland↗

DOE OSTI2025

Forest Age Rivals Climate to Explain Reproductive Allocation Patterns in Forest Ecosystems Globally

ABSTRACT Forest allocation of net primary productivity (NPP) to reproduction (carbon required for flowers, fruits, and seeds) is poorly quantified globally, despite its critical role in forest regeneration and a well‐supported trade‐off with allocation to growth. Here, we present the first global synthesis of a biometric proxy for forest reproductive allocation (RA) across environmental and stand age gradients from a compiled dataset of 824 observations across 393 sites. We find that ecosystem‐scale RA increases ~60% from boreal to tropical forests. Climate shows important non‐linear relationships with RA, but is not the sole predictor. Forest age effects are comparable to climate in magnitude (MAT: ß = 0.24,p = 0.021; old growth forest: ß = 0.22,p < 0.001), while metrics of soil fertility show small but significant relationships with RA (soil pH: ß = 0.07,p = 0.001; soil N: ß = −0.07,p = 0.001). These results provide strong evidence that ecosystem‐scale RA is mediated by climate, forest age, and soil conditions, and is not a globally fixed fraction of positive NPP as assumed by most vegetation and ecosystem models. Our dataset and findings can be used by modellers to improve predictions of forest regeneration and carbon cycling.

Environmental Sciences & Ecology↗

DOE OSTI2024

Nitrogen Deposition Weakens Soil Carbon Control of Nitrogen Dynamics Across the Contiguous United States

ABSTRACT Anthropogenic nitrogen (N) deposition is unequally distributed across space and time, with inputs to terrestrial ecosystems impacted by industry regulations and variations in human activity. Soil carbon (C) content normally controls the fraction of mineralized N that is nitrified ( ƒ nitrified ), affecting N bioavailability for plants and microbes. However, it is unknown whether N deposition has modified the relationships among soil C, net N mineralization, and net nitrification. To test whether N deposition alters the relationship between soil C and net N transformations, we collected soils from coniferous and deciduous forests, grasslands, and residential yards in 14 regions across the contiguous United States that vary in N deposition rates. We quantified rates of net nitrification and N mineralization, soil chemistry (soil C, N, and pH), and microbial biomass and function (as beta‐glucosidase (BG) and N ‐acetylglucosaminidase (NAG) activity) across these regions. Following expectations, soil C was a driver of ƒ nitrified across regions, whereby increasing soil C resulted in a decline in net nitrification and ƒ nitrified . The ƒ nitrified value increased with lower microbial enzymatic investment in N acquisition (increasing BG:NAG ratio) and lower active microbial biomass, providing some evidence that heterotrophic microbial N demand controls the ammonium pool for nitrifiers. However, higher total N deposition increased ƒ nitrified , including for high soil C sites predicted to have low ƒ nitrified , which decreased the role of soil C as a predictor of ƒ nitrified . Notably, the drop in contemporary atmospheric N deposition rates during the 2020 COVID‐19 pandemic did not weaken the effect of N deposition on relationships between soil C and ƒ nitrified . Our results suggest that N deposition can disrupt the relationship between soil C and net N transformations, with this change potentially explained by weaker microbial competition for N. Therefore, past N inputs and soil C should be used together to predict N dynamics across terrestrial ecosystems.

Nieland, Matthew A. [Stockbridge School of Agricul↗

DOE OSTI2024

Evaluation of GlassNet for physics-informed machine learning of glass stability and glass-forming ability

Glassy materials form the basis of many modern applications, including nuclear waste immobilization, touch-screen displays, and optical fibers, and also hold great potential for future medical and environmental applications. However, their structural complexity and large composition space make design and optimization challenging for certain applications. Of particular importance for glass processing and design is an estimate of a given composition's glass-forming ability (GFA). However, there remain many open questions regarding the underlying physical mechanisms of glass formation, especially in oxide glasses. It is apparent that a proxy for GFA would be highly useful in glass processing and design, but identifying such a surrogate property has proven itself to be difficult. While glass stability (GS) parameters have historically been used as a GFA surrogate, recent research has demonstrated that most of these parameters are not accurate predictors of the GFA of oxide glasses. Here, in this work, we explore the application of an open-source pre-trained neural network model, GlassNet, that can predict the characteristic temperatures necessary to compute GS with reasonable performance and assess the feasibility of using these physics-informed machine learning (PIML)-predicted GS parameters to estimate GFA. In doing so, we track the uncertainties at each step of the computation—from the original ML prediction errors to the compounding of errors during GS estimation, and finally to the final estimation of GFA. While GlassNet exhibits reasonable accuracy on all individual properties, we observe a large compounding of error in the combination of these individual predictions for the PIML prediction of GS, finding that random forest models offer similar accuracy to GlassNet. We also break down the performance of GlassNet on different glass families and find that the error in GS prediction is correlated with the error in crystallization peak temperature prediction. Lastly, we utilize this finding to assess the relationship between top-performing GS parameters and GFA for two ternary glass systems: sodium borosilicate and sodium iron phosphate glasses. We conclude that to obtain true ML predictive capability of GFA, significantly more data needs to be collected.

36 MATERIALS SCIENCE↗

DOE OSTI2025

Spatial Replication Is Important for Developing Landscape Genetic Inferences for a Wetland Salamander

Habitat fragmentation is a pressing threat to wildlife populations, and maintenance of gene flow between populations is an essential goal of conservation. Resistance surfaces have emerged as an important tool for modelling connectivity and developing management strategies to mitigate effects of habitat fragmentation. However, recent studies have noted inconsistencies in the factors most strongly associated with connectivity across different landscapes. Thus, replication of genetic-based resistance surface optimisation across landscapes may be necessary for making robust conclusions about the influence of environmental variables. Accordingly, replication represents a substantive challenge and opportunity in the field of landscape genetics. In this study, we conducted replicated landscape genetic analyses across five landscapes in Tennessee and Kentucky for a threatened wetland amphibian, the four-toed salamander (Hemidactylium scutatum). We tested multiple hypotheses of how different landscape features that could directly affect small, desiccation-intolerant amphibians (e.g., canopy cover) influenced gene flow and assessed the appropriate scale at which to model different features. We found some concordance in the landscape features that influenced gene flow (e.g., a common importance of forest cover and topography), but also some differences—potentially owing to the difference in variability of predictors across landscapes. We also found discordance in the scale of effect of different features across landscapes. In conclusion, our work emphasises that flat areas of moist forest not bisected by roads may be important for H. scutatum conservation, and our replicated design allows us to identify relationships that would have been missed if only using one study site.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2025

Using leaf and stomatal traits to predict biomass production and water use efficiency in Populus

Climate change is reshaping ecosystems, driving plants to adapt through leaf-trait plasticity that reflects strategies for growth and water use. Predicting biomass production and intrinsic water use efficiency (iWUE) remains challenging because of genetic, taxonomic, and environmental variability. Here, we used eastern cottonwood and Populus hybrids as a model system to test whether easily measurable leaf traits can serve as reliable predictors of performance, and whether adding stomatal and biochemical traits improves predictive power. Across two field sites in Mississippi, leaf mass per area (LMA), biomass production, iWUE, leaf area, and foliar nitrogen ( N %) differed significantly among taxa and sites, while other traits were conserved. Factorial analysis of mixed data (FAMD) revealed distinct clustering of taxa and sites, indicating coordinated variation among leaf and stomatal traits. Pairwise correlations highlighted fundamental trade-offs, with biomass positively related to LMA and petiole length but negatively associated with iWUE, N %, and carbon isotopic ratios (δ 13 C). Leaf temperature and leaf angle varied among taxa and were significantly correlated with LMA and petiole length, suggesting mechanisms of heat dissipation and leaf movability that link simple traits to gas exchange and productivity. Weighted multiple linear regression models explained 80%–91% of variation in biomass production and iWUE. Models using only LMA, petiole length, and stomatal metrics performed nearly as well as those incorporating N %, and δ 13 C, with complex traits adding approximately 10% explanatory power. These results demonstrate that simple morphological traits capture integrated functional trade-offs, while complex traits refine predictions. This tiered approach provides an efficient framework for selecting high-yielding, water-efficient genotypes of Populus and other hardwood species, offering practical pathways to enhance carbon uptake and iWUE under climate change.

biomass production↗

DOE OSTI2025

One-shot gas detection with transformer paired neural networks in Mako collected longwave infrared hyperspectral imagery

To date, careful data treatment workflows and statistical detectors are used to perform hyperspectral image (HSI) detection of any gas contained in a spectral library, which is often expanded with physics models to incorporate different spectral characteristics. In general, surrounding evidence or known gas-release parameters are used to provide confidence in or confirm detection capability, respectively. This makes quantifying detection performance difficult as it is nearly impossible to develop an absolute ground truth for gas target pixel presence in collected HSI. Consequently, developing and comparing new detection methods, especially machine learning (ML)-based methods, is susceptible to subjectivity in derived detection map quality. Here, in this work, we demonstrate the first use of transformer-based paired neural networks (PNNs) for one-shot gas target detection for multiple gases while providing quantitative classification and detection metrics for their use on labeled data. Terabytes of training data are generated from a database of long-wave infrared HSI obtained from historical Mako sensor campaigns over Los Angeles. By incorporating labels, singular signature representations, and a model development pipeline, we can tune and select PNNs to detect multiple gas targets that are not seen in training on a quantitative basis. We additionally assess our test set detections using interpretability techniques widely employed with ML-based predictors, but less common with detection methods relying on learned latent spaces.

Hyperspectral imaging↗

DOE OSTI2025

DNA methylation–based age prediction and sex-specific epigenetic aging in a lizard with female-biased longevity

Sex differences in life span are widespread across animal taxa, but their causes remain unresolved. Alterations to the epigenome are hypothesized to contribute to vertebrate aging, and DNA methylation–based aging clocks allow for quantitative estimation of biological aging trajectories. Here, we investigate the influence of age, sex, and their interaction on genome-wide DNA methylation patterns in the brown anole (Anolis sagrei), a lizard with pronounced female-biased survival and longevity. We develop a series of age predictor models and find that, contrary to our predictions, rates of epigenetic aging were not slower in female lizards. However, methylation states at loci acquiring age-associated changes appear to be more “youthful” in young females, suggesting that female DNA methylomes are preemptively fortified in early life in opposition to the direction of age-related drift. Collectively, our findings provide insights into epigenetic aging in reptiles and suggest that early-life epigenetic profiles may be more informative than rates of change for predicting sex biases in longevity.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2025

North Pacific meridional mode has larger impacts on El Niño evolution than the March Madden-Julian Oscillation

The El Niño–Southern Oscillation (ENSO) is a key driver of global climate variability. Early-season westerly wind bursts (WWBs) have long been suggested to be important for ENSO evolution and diversity, with the Madden-Julian Oscillation (MJO) among the main sources of WWBs. However, MJO’s contribution to ENSO evolution has been difficult to quantify. Here, using an ensemble hindcast approach specifically designed to isolate internal atmospheric variability, we evaluate the influences of March MJO on subsequent ENSO development. Our results show that the March MJO, under favorable background conditions, by itself has limited impacts on ENSO due to weak equatorial air-sea coupling in spring. In comparison, the North Pacific Oscillation–induced meridional mode exerts a more sustained influence on ENSO evolution. A cyclonic circulation anomaly over Hawaii, associated with the Pacific-North American pattern, also plays a role. These findings suggest that March MJO activity alone may not be a reliable predictor for ENSO evolution, but underscore the importance of North Pacific atmospheric variability.

Science & Technology - Other Topics↗

DOE OSTI2026

Specific ion effects on ion transport in charged polymer membranes

Selective separation of like-charged ions is a central challenge in applications such as critical mineral recovery. Electrochemical membrane–based separations offer promising pathways to address this need, but a limited fundamental understanding of ion transport in charged polymer membranes hampers the development of highly selective materials. This study elucidates the role of specific ion effects (SIEs) in ion transport through charged polymer membranes by integrating experimental measurements of ion mobility and in situ ion/ion and ion/water interactions in model charged polymer membranes with results from molecular dynamics simulations on analogous systems. We demonstrate that solvent-mediated ion interactions drive pronounced SIEs within the studied membranes, with the ion softness, i.e., the malleability of ion hydration shells, from hard/soft acid/base (HSAB) theory emerging as a key predictor of transport properties. HSAB theory also explains the observed mechanism of solvent-mediated ion interactions. Our findings offer a mechanistic framework for designing membranes with tailored ion selectivity, potentially enabling efficient separations of chemically similar ions.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2026

Predictive analytics to direct clinical attention to complex patients with elevated suicide risk: enhancement of the Veterans Health Administration REACH VET model

Suicide is a major public health concern, particularly among Veterans. The U.S. Department of Veterans Affairs Veterans Health Administration (VHA) employs the Recovery Engagement and Coordination for Health–Veterans Enhanced Treatment (REACH VET) model to prioritise high-risk patients for targeted clinical attention. REACH VET 1.0 (RV 1.0) was developed on 2008–2011 data. To reflect changes in clinical practice and populations, VHA updated it to REACH VET 2.0 (RV 2.0). This study describes its development and validation. RV 2.0 used longitudinal data from 7,248,170 VHA patients (4,967 suicide deaths) in 2018–2019, with 650 time-varying demographic, clinical and area-level predictors derived from a 2-year lookback (2016–2019). An ensemble of Elastic-Net logistic regression models was trained on 2018 data and evaluated monthly at the population level in 2019, focusing on the top 0.1% intervention risk tier. Analyses assessed model discrimination, suicide detection, risk concentration, subgroup consistency (sex, age and race/ethnicity) and performance relative to RV 1.0 using the same percentile-based risk strata. RV 2.0 outperformed RV 1.0 across all risk strata, with better discrimination (C-statistic 0.76 vs 0.69) and consistent performance across demographic subgroups. Within the top 0.1% of predicted risk, RV 2.0 identified more deaths, higher suicide rates and greater mortality risk concentration both when averaged across the 12 monthly 2019 test sets (5.6 vs 3.6; 83.6 vs 53.7 per 100,000 person-years; 21.0 vs 14.1) and when annualised for 2019 (67 vs 43; 2.7% vs 1.7%; 1,003 vs 644 per 100,000 person-years; 26.7 vs 17.1). RV 2.0 improves suicide risk stratification among Veterans, demonstrating better performance and consistent prediction across subgroups and highlighting the need for regular model updates and evaluation.

Peluso, Alina [Oak Ridge National Laboratory (ORNL↗

DOE OSTI2025

Adaptive Uncertainty Quantification for Stochastic Hyperbolic Conservation Laws

Here, we propose a predictor-corrector adaptive method for the study of hyperbolic partial differential equations (PDEs) under uncertainty. Constructed around the framework of stochastic finite volume (SFV) methods, our approach circumvents sampling schemes or simulation ensembles while also preserving fundamental properties, in particular hyperbolicity of the resulting systems and conservation of the discrete solutions. Furthermore, we augment the existing SFV theory with a priori convergence results for statistical quantities, in particular push-forward densities, which we demonstrate through numerical experiments. By linking refinement indicators to regions of the physical and stochastic spaces, we drive anisotropic refinements of the discretizations, introducing new degrees of freedom where deemed profitable. To illustrate our proposed method, we consider a series of numerical examples for nonlinear hyperbolic PDEs based on Burgers’ and Euler’s equations.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2026

CAHS: Context-Aware Homology Search

Protein homology search is foundational to bioinformatics: it supports annotation transfer, structure/function inference, and evolutionary analysis over rapidly expanding sequence repositories (e.g., UniProtKB). Profile hidden Markov models (pHMMs), as implemented in HMMER, remain the most widely trusted approach because they provide statistically calibrated E-values; however, their gap behavior is fixed once a profile is trained, despite biological evidence that insertion/deletion tolerance varies across flexible loops and intrinsically disordered regions. We present CAHS (Context-Aware Homology Search), a lightweight query-time adapter for pHMM search that incorporates learned and biologically motivated signals without changing HMMER's downstream search pipeline or its calibrated E-value reporting. Given a query sequence, CAHS computes per-residue representations from a protein language model and a disorder predictor, maps these to profile coordinates, and modulates only match-state transition rows (gap-open and gap-extension probabilities) while preserving Plan7 constraints. We comprehensively evaluate CAHS across six structurally diverse protein families and multi-domain architectures against a 570k-sequence target corpus. CAHS expands detection capability, retrieving thousands of additional remote homologs at relaxed thresholds by maintaining alignment quality through flexible regions. For multi-domain proteins, context-aware modulation resolves 94% of fragmented alignments. Crucially, CAHS preserves hit-set invariance at stringent operating points (E<10-10), demonstrating increased statistical confidence without inflating false positives. Furthermore, sharper statistical distinction between homologs and background noise during early filter stages yields up to a 3.87× acceleration in end-to-end wall-clock time on high-performance computing clusters. Overall, CAHS illustrates a practical AI-for-science design pattern: augmenting a trusted probabilistic model with query-specific learned signals to improve interpretable, reproducible inference in data-rich biology.

Bhattaram, Swethasree [Georgia Institute of Techno↗

DOE OSTI2026

wa-hls4ml: A Benchmark and Surrogate Models for hls4ml Resource and Latency Estimation

As machine learning (ML) is increasingly implemented in hardware to address real-time challenges in scientific applications, the development of advanced toolchains has significantly reduced the time required to iterate on various designs. These advancements have solved major obstacles, but also exposed new challenges. For example, processes that were not previously considered bottlenecks, such as hardware synthesis, are becoming limiting factors in the rapid iteration of designs. To mitigate these emerging constraints, multiple efforts have been undertaken to develop an ML-based surrogate model that estimates resource usage of ML accelerator architectures. We introduce wa-hls4ml, a benchmark for ML accelerator resource and latency estimation, and its corresponding initial dataset of over 680,000 fully connected and convolutional neural networks, all synthesized using hls4ml and targeting Xilinx FPGAs. The benchmark evaluates the performance of resource and latency predictors against several common ML model architectures, primarily originating from scientific domains, as exemplar models, and the average performance across a subset of the dataset. Additionally, we introduce GNN- and transformer-based surrogate models that predict latency and resources for ML accelerators. We present the architecture and performance of the models and find that the models generally predict latency and resources for the 75% percentile within several percent of the synthesized resources on the synthetic test dataset.

Hawks, Benjamin [Fermilab] (ORCID:0000000157000288↗