Search NASA⌕ Search

SEARCH · Search NASA

Results for “Bayesian model averaging”

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 163 records · Page 9

GIGA-Lens: Fast Bayesian Inference for Strong Gravitational Lens Modeling

We present GIGA-Lens: a gradient-informed, GPU-accelerated Bayesian framework for modeling strong gravitational lensing systems, implemented in TensorFlow and JAX. The three components, optimization using multistart gradient descent, posterior covariance estimation with variational inference, and sampling via Hamiltonian Monte Carlo, all take advantage of gradient information through automatic differentiation and massive parallelization on graphics processing units (GPUs). We test our pipeline on a large set of simulated systems and demonstrate in detail its high level of performance. The average time to model a single system on four Nvidia A100 GPUs is 105 s. The robustness, speed, and scalability offered by this framework make it possible to model the large number of strong lenses found in current surveys and present a very promising prospect for the modeling of ${ \mathcal O }({10}^{5})$ lensing systems expected to be discovered in the era of the Vera C. Rubin Observatory, Euclid, and the Nancy Grace Roman Space Telescope.

79 ASTRONOMY AND ASTROPHYSICS↗

Semi-supervised Bayesian Low-shot Learning

Deep neural networks (NNs) typically outperform traditional machine learning (ML) approaches for complicated, non-linear tasks. It is expected that deep learning (DL) should offer superior performance for the important non-proliferation task of predicting explosive device configuration based upon observed optical signature, a task which human experts struggle with. However, supervised machine learning is difficult to apply in this mission space because most recorded signatures are not associated with the corresponding device description, or “truth labels.” This is challenging for NNs, which traditionally require many samples for strong performance. Semi-supervised learning (SSL), low-shot learning (LSL), and uncertainty quantification (UQ) for NNs are emerging approaches that could bridge the mission gaps of few labels and rare samples of importance. NN explainability techniques are important in gaining insight into the inferential feature importance of such a complex model. In this work, SSL, LSL, and UQ are merged into a single framework, a significant technical hurdle not previously demonstrated. Exponential Average Adversarial Training (EAAT) and Pairwise Neural Networks (PNNs) are chosen as the SSL and LSL methods of choice. Permutation feature importance (PFI) for functional data is used to provide explainability via the Variable importance Explainable Elastic Shape Analysis (VEESA) pipeline. A variety of uncertainty quantification approaches are explored: Bayesian Neural Networks (BNNs), ensemble methods, concrete dropout, and evidential deep learning. Two final approaches, one utilizing ensemble methods and one utilizing evidential learning, are constructed and compared using a well-quantified synthetic 2D dataset along with the DIRSIG Megascene.

97 MATHEMATICS AND COMPUTING↗

Lower soil moisture and deep soil temperatures in thermokarst features increase old soil carbon loss after 10 years of experimental permafrost warming

Almost half of the global terrestrial soil carbon (C) is stored in the northern circumpolar permafrost region, where air temperatures are increasing two times faster than the global average. As climate warms, permafrost thaws and soil organic matter becomes vulnerable to greater microbial decomposition. Long-term soil warming of ice-rich permafrost can result in thermokarst formation that creates variability in environmental conditions. Consequently, plant and microbial proportional contributions to ecosystem respiration may change in response to long-term soil warming. Natural abundance δ 13 C and Δ 14 C of aboveground and belowground plant material, and of young and old soil respiration were used to inform a mixing model to partition the contribution of each source to ecosystem respiration fluxes. We employed a hierarchical Bayesian approach that incorporated gross primary productivity and environmental drivers to constrain source contributions. We found that long-term experimental permafrost warming introduced a soil hydrology component that interacted with temperature to affect old soil C respiration. Old soil C loss was suppressed in plots with warmer deep soil temperatures because they tended to be wetter. When soil volumetric water content significantly decreased in 2018 relative to 2016 and 2017, the dominant respiration sources shifted from plant aboveground and young soil respiration to old soil respiration. The proportion of ecosystem respiration from old soil C accounted for up to 39% of ecosystem respiration and represented a 30-fold increase compared to the wet-year average. Our findings show that thermokarst formation may act to moderate microbial decomposition of old soil C when soil is highly saturated. However, when soil moisture decreases, a higher proportion of old soil C is vulnerable to decomposition and can become a large flux to the atmosphere. Furthermore, as permafrost systems continue to change with climate, we must understand the thresholds that may propel these systems from a C sink to a source.

permafrost↗

Disentangling the Black Hole Mass Spectrum with Photometric Microlensing Surveys

Abstract From the formation mechanisms of stars and compact objects to nuclear physics, modern astronomy frequently leverages surveys to understand populations of objects to answer fundamental questions. The population of dark and isolated compact objects in the Galaxy contains critical information related to many of these topics, but is only practically accessible via gravitational microlensing. However, photometric microlensing observables are degenerate for different types of lenses, and one can seldom classify an event as involving either a compact object or stellar lens on its own. To address this difficulty, we apply a Bayesian framework that treats lens type probabilistically and jointly with a lens population model. This method allows lens population characteristics to be inferred despite intrinsic uncertainty in the lens class of any single event. We investigate this method’s effectiveness on a simulated ground-based photometric survey in the context of characterizing a hypothetical population of primordial black holes (PBHs) with an average mass of 30 M ⊙ . On simulated data, our method outperforms current black hole (BH) lens identification pipelines and characterizes different subpopulations of lenses while jointly constraining the PBH contribution to dark matter to ≈25%. Key to robust inference, our method can marginalize over population model uncertainty. We find the lower mass cutoff for stellar origin BHs, a key observable in understanding the BH mass gap, particularly difficult to infer in our simulations. This work lays the foundation for cutting-edge PBH abundance constraints to be extracted from current photometric microlensing surveys.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

WE-E-BRE-05: Ensemble of Graphical Models for Predicting Radiation Pneumontis Risk

Purpose: We propose a prior knowledge-based approach to construct an interaction graph of biological and dosimetric radiation pneumontis (RP) covariates for the purpose of developing a RP risk classifier. Methods: We recruited 59 NSCLC patients who received curative radiotherapy with minimum 6 month follow-up. 16 RP events was observed (CTCAE grade ≥2). Blood serum was collected from every patient before (pre-RT) and during RT (mid-RT). From each sample the concentration of the following five candidate biomarkers were taken as covariates: alpha-2-macroglobulin (α2M), angiotensin converting enzyme (ACE), transforming growth factor β (TGF-β), interleukin-6 (IL-6), and osteopontin (OPN). Dose-volumetric parameters were also included as covariates. The number of biological and dosimetric covariates was reduced by a variable selection scheme implemented by L1-regularized logistic regression (LASSO). Posterior probability distribution of interaction graphs between the selected variables was estimated from the data under the literature-based prior knowledge to weight more heavily the graphs that contain the expected associations. A graph ensemble was formed by averaging the most probable graphs weighted by their posterior, creating a Bayesian Network (BN)-based RP risk classifier. Results: The LASSO selected the following 7 RP covariates: (1) pre-RT concentration level of α2M, (2) α2M level mid- RT/pre-RT, (3) pre-RT IL6 level, (4) IL6 level mid-RT/pre-RT, (5) ACE mid-RT/pre-RT, (6) PTV volume, and (7) mean lung dose (MLD). The ensemble BN model achieved the maximum sensitivity/specificity of 81%/84% and outperformed univariate dosimetric predictors as shown by larger AUC values (0.78∼0.81) compared with MLD (0.61), V20 (0.65) and V30 (0.70). The ensembles obtained by incorporating the prior knowledge improved classification performance for the ensemble size 5∼50. Conclusion: We demonstrated a probabilistic ensemble method to detect robust associations between RP covariates and its potential to improve RP prediction accuracy. Our Bayesian approach to incorporate prior knowledge can enhance efficiency in searching of such associations from data. The authors acknowledge partial support by: 1) CREATE Medical Physics Research Training Network grant of the Natural Sciences and Engineering Research Council (Grant number: 432290) and 2) The Terry Fox Foundation Strategic Training Initiative for Excellence in Radiation Research for the 21st Century (EIRR21)

60 APPLIED LIFE SCIENCES↗

Bayesian Calibration of the Community Land Model using Surrogates

We present results from the Bayesian calibration of hydrological parameters of the Community Land Model (CLM), which is often used in climate simulations and Earth system models. A statistical inverse problem is formulated for three hydrological parameters, conditioned on observations of latent heat surface fluxes over 48 months. Our calibration method uses polynomial and Gaussian process surrogates of the CLM, and solves the parameter estimation problem using a Markov chain Monte Carlo sampler. Posterior probability densities for the parameters are developed for two sites with different soil and vegetation covers. Our method also allows us to examine the structural error in CLM under two error models. We find that accurate surrogate models can be created for CLM in most cases. The posterior distributions lead to better prediction than the default parameter values in CLM. Climatologically averaging the observations does not modify the parameters’ distributions significantly. The structural error model reveals a correlation time-scale which can potentially be used to identify physical processes that could be contributing to it. While the calibrated CLM has a higher predictive skill, the calibration is under-dispersive.

Bayesian calibration↗

Documentation of a Linked-Population Viability Model to Evaluate Recovery Options for Snake River Fall Chinook Salmon

We developed an age- and life-history based population viability analysis model for linked populations of fall Chinook salmon (LPVA). LPVA was designed to quantify the costs and benefits of re-establishing access to an upstream spawning area using managed reconnection options such as hatchery stocking and translocation. The model incorporates stochasticity in the age at which salmon return to spawn, sex ratio, juvenile life history, and survival rates. Response variables include the proportion of replicate populations that persisted for 80 y (roughly 20 generations), the linked population’s growth rate, λ, time to extinction and summaries of population sizes including average female spawner abundance, final female spawner abundance, and total female abundance. This report attempts to provide a complete model description, including simulation of management options and options for simulating forward with a joint distribution of parameters based on their likelihoods with respect to historical data. Intermediate results illustrating the Bayesian approach to identifying poorly-known parameters are reported here. LPVA simulates reconnection of linked populations by representing management processes such as trapping of adults migrating upstream at two dams and juveniles migrating downstream, and hatchery operations for one or two hatcheries. The purpose of the model is to support efforts to recover the threatened Snake River fall Chinook salmon ESU by evaluating population viability using existing spawning areas with and without a hatchery and to assess the potential costs and benefits of programs to re-establish an upstream spawning area.

Jager, Henriette (Yetta) I.↗

BISON Fuel Fragmentation Relocation and Dispersal (FFRD) Assessment Database for Eventual Use in Bayesian Calibration

Existing light-water reactor (LWR) fuel vendors have been interested in seeking increased discharge burnups of nuclear fuel rods for improved economics for quite a few years. With increased burnups come additional challenges that must be addressed. It has been experimentally observed that average burnups higher than the current regulatory limit of 62 MWd/kgU may undergo fuel fragmentation, relocation, and dispersal (FFRD) during a loss-of-coolant accident (LOCA). Industry must demonstrate approaches to mitigate FFRD. In an effort to support industry, the Nuclear Energy Advanced Modeling and Simulation (NEAMS) program within the U.S. Department of Energy (DOE) has for several years invested in developing multiscale models and creating a validation/assessment database for these models to study the mechanisms driving FFRD. This report provides an update on changes made to the assessment database and new models added to BISON to support the study of fuel rod behavior during FFRD. An effort has been initiated this year to begin adding dedicated inputs to the publicly available Virtual Test Bed (VTB) repository for industry use. A section of this report details the efforts made in this area. NEAMS has recently developed new capabilities in the Multiphysics Object-Oriented Simulation Environment (MOOSE) framework’s stochastic tools module for calibration using Bayesian inference. The goal in the future is to use these capabilities to calibrate and identify weaknesses in the existing BISON models for FFRD. The report concludes with a discussion on the models most likely to benefit the most from such calibration.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

CLASH: JOINT ANALYSIS OF STRONG-LENSING, WEAK-LENSING SHEAR, AND MAGNIFICATION DATA FOR 20 GALAXY CLUSTERS

We present a comprehensive analysis of strong-lensing, weak-lensing shear and magnification data for a sample of 16 X-ray-regular and 4 high-magnification galaxy clusters at 0.19≲z≲0.69 selected from Cluster Lensing And Supernova survey with Hubble (CLASH). Our analysis combines constraints from 16-band Hubble Space Telescope observations and wide-field multi-color imaging taken primarily with Suprime-Cam on the Subaru Telescope, spanning a wide range of cluster radii (10″–16′). We reconstruct surface mass density profiles of individual clusters from a joint analysis of the full lensing constraints, and determine masses and concentrations for all of the clusters. We find the internal consistency of the ensemble mass calibration to be ≤5% ± 6% in the one-halo regime (200–2000 kpc h{sup −1}) compared to the CLASH weak-lensing-only measurements of Umetsu et al. For the X-ray-selected subsample of 16 clusters, we examine the concentration–mass (c–M) relation and its intrinsic scatter using a Bayesian regression approach. Our model yields a mean concentration of c|{sub z=0.34}=3.95±0.35 at M{sub 200c} ≃ 14 × 10{sup 14} M{sub ⊙} and an intrinsic scatter of σ(lnc{sub 200c})=0.13±0.06, which is in excellent agreement with Λ cold dark matter predictions when the CLASH selection function based on X-ray morphological regularity and the projection effects are taken into account. We also derive an ensemble-averaged surface mass density profile for the X-ray-selected subsample by stacking their individual profiles. The stacked lensing signal is detected at 33σ significance over the entire radial range ≤4000 kpc h{sup −1}, accounting for the effects of intrinsic profile variations and uncorrelated large-scale structure along the line of sight. The stacked mass profile is well described by a family of density profiles predicted for cuspy dark-matter-dominated halos in gravitational equilibrium, namely, the Navarro–Frenk–White (NFW), Einasto, and DARKexp models, whereas the single power-law, cored isothermal and Burkert density profiles are disfavored by the data. We show that cuspy halo models that include the large-scale two-halo term provide improved agreement with the data. For the NFW halo model, we measure a mean concentration of c{sub 200c}=3.79{sub −0.28}{sup +0.30} at M{sub 200c}=14.1{sub −1.0}{sup +1.0}×10{sup 14} M{sub ⊙}, demonstrating consistency between the complementary analysis methods.

79 ASTRONOMY AND ASTROPHYSICS↗

Coping with Trial-to-Trial Variability of Event Related Signals: A Bayesian Inference Approach

In electro-neurophysiology, single-trial brain responses to a sensory stimulus or a motor act are commonly assumed to result from the linear superposition of a stereotypic event-related signal (e.g. the event-related potential or ERP) that is invariant across trials and some ongoing brain activity often referred to as noise. To extract the signal, one performs an ensemble average of the brain responses over many identical trials to attenuate the noise. To date, h s simple signal-plus-noise (SPN) model has been the dominant approach in cognitive neuroscience. Mounting empirical evidence has shown that the assumptions underlying this model may be overly simplistic. More realistic models have been proposed that account for the trial-to-trial variability of the event-related signal as well as the possibility of multiple differentially varying components within a given ERP waveform. The variable-signal-plus-noise (VSPN) model, which has been demonstrated to provide the foundation for separation and characterization of multiple differentially varying components, has the potential to provide a rich source of information for questions related to neural functions that complement the SPN model. Thus, being able to estimate the amplitude and latency of each ERP component on a trial-by-trial basis provides a critical link between the perceived benefits of the VSPN model and its many concrete applications. In this paper we describe a Bayesian approach to deal with this issue and the resulting strategy is referred to as the differentially Variable Component Analysis (dVCA). We compare the performance of dVCA on simulated data with Independent Component Analysis (ICA) and analyze neurobiological recordings from monkeys performing cognitive tasks.

Ding, Mingzhou↗

PROVABGS: The Probabilistic Stellar Mass Function of the BGS One-percent Survey

We present the probabilistic stellar mass function (pSMF) of galaxies in the DESI Bright Galaxy Survey (BGS), observed during the One-percent Survey. The One-percent Survey was one of DESI's survey validation programs conducted from 2021 April to May, before the start of the main survey. It used the same target selection and similar observing strategy as the main survey and successfully observed the spectra and redshifts of 143,017 galaxies in the r < 19.5 magnitude-limited BGS Bright sample and 95,499 galaxies in the fainter surface-brightness- and color-selected BGS Faint sample over z < 0.6. We derive pSMFs from posteriors of stellar mass, M*, inferred from DESI photometry and spectroscopy using the Hahn et al. PRObabilistic Value-Added BGS (PROVABGS) Bayesian spectral energy distribution modeling framework. We use a hierarchical population inference framework that statistically and rigorously propagates the M* uncertainties. Furthermore, we include correction weights that account for the selection effects and incompleteness of the BGS observations. We present the redshift evolution of the pSMF in BGS, as well as the pSMFs of star-forming and quiescent galaxies classified using average specific star formation rates from PROVABGS. Overall, the pSMFs show good agreement with previous stellar mass function measurements in the literature. Our pSMFs showcase the potential and statistical power of BGS, which in its main survey will observe >100 × more galaxies. Moreover, we present the statistical framework for subsequent population statistics measurements using BGS, which will characterize the global galaxy population and scaling relations at low redshifts with unprecedented precision.

79 ASTRONOMY AND ASTROPHYSICS↗

Bayesian calibration of irradiated graphite property models under high temperatures

Graphite under high temperatures and irradiation is central to advanced reactors. We develop a Bayesian calibration framework for graphite property models that explicitly represents model-data mismatch via a Gaussian-process discrepancy. The approach propagates uncertainty from parameters, experimental noise, and model form, with a hierarchical variance structure to capture group and cross-group noise. Using two predictive models across five grades (IG-110, NBG-18, PCEA, NBG-17, 2114) and four properties-irradiation-induced dimension change, creep, Young’s modulus change ratio, and coefficient of thermal expansion change ratio-we obtain average predictive-error reductions of 54%, 65%, 17%, and 17% when discrepancy is included. We illustrate engineering impact with a multiphysics model of a very-high-temperature reactor prismatic reflector brick, analyzing stresses under high fluence and temperature. Accounting for model discrepancy markedly improves predictive accuracy and provides a robust basis for reliable graphite component design in advanced reactors.

36 - MATERIALS SCIENCE↗

Spatiotemporal Associations Between Social Vulnerability, Environmental Measurements, and COVID-19 in the Conterminous United States

This study summarizes the results from fitting a Bayesian hierarchical spatiotemporal model to coronavirus disease 2019 (COVID-19) cases and deaths at the county level in the United States for the year 2020. Two models were created, one for cases and one for deaths, utilizing a scaled Besag, York, Mollié model with Type I spatial-temporal interaction. Each model accounts for 16 social vulnerability and 7 environmental variables as fixed effects. The spatial pattern between COVID-19 cases and deaths is significantly different in many ways. The spatiotemporal trend of the pandemic in the United States illustrates a shift out of many of the major metropolitan areas into the United States Southeast and Southwest during the summer months and into the upper Midwest beginning in autumn. Analysis of the major social vulnerability predictors of COVID-19 infection and death found that counties with higher percentages of those not having a high school diploma, having non-White status and being Age 65 and over to be significant. Among the environmental variables, above ground level temperature had the strongest effect on relative risk to both cases and deaths. Hot and cold spots, areas of statistically significant high and low COVID-19 cases and deaths respectively, derived from the convolutional spatial effect show that areas with a high probability of above average relative risk have significantly higher Social Vulnerability Index composite scores. The same analysis utilizing the spatiotemporal interaction term exemplifies a more complex relationship between social vulnerability, environmental measurements, COVID-19 cases, and COVID-19 deaths.

spatial epidemiology↗

Data free inference with processed data products

Here, we consider the context of probabilistic inference of model parameters given error bars or confidence intervals on model output values, when the data is unavailable. We introduce a class of algorithms in a Bayesian framework, relying on maximum entropy arguments and approximate Bayesian computation methods, to generate consistent data with the given summary statistics. Once we obtain consistent data sets, we pool the respective posteriors, to arrive at a single, averaged density on the parameters. This approach allows us to perform accurate forward uncertainty propagation consistent with the reported statistics.

97 MATHEMATICS AND COMPUTING↗

Spatially resolved microlensing time-scale distributions across the Galactic bulge with the VVV survey

ABSTRACT We analyse 1602 microlensing events found in the VISTA Variables in the Via Lactea (VVV) near-infrared (NIR) survey data. We obtain spatially resolved, efficiency-corrected time-scale distributions across the Galactic bulge (|ℓ| < 10°, |b| < 5°), using a Bayesian hierarchical model. Spatially resolved peaks and means of the time-scale distributions, along with their marginal distributions in strips of longitude and latitude, are in agreement at a 1σ level with predictions based on the Besançon model of the Galaxy. We find that the event time-scales in the central bulge fields (|ℓ| < 5°) are on average shorter than the non-central (|ℓ| > 5°) fields, with the average peak of the lognormal time-scale distribution at 23.6 ± 1.9 d for the central fields and 29.0 ± 3.0 d for the non-central fields. Our ability to probe the structure of the bulge with this sample of NIR microlensing events is limited by the VVV survey’s sparse cadence and relatively small number of detected microlensing events compared to dedicated optical surveys. Looking forward to future surveys, we investigate the capability of the Roman telescope to detect spatially resolved asymmetries in the time-scale distributions. We propose two pairs of Roman fields, centred on (ℓ = ±9, 5°, b = −0.125°) and (ℓ = −5°, b = ±1.375°) as good targets to measure the asymmetry in longitude and latitude, respectively.

79 ASTRONOMY AND ASTROPHYSICS↗

Recombination smooths the time-signal disrupted by latency in within-host HIV phylogenies

Within-host HIV evolution involves several features that may disrupt standard phylogenetic reconstruction. One important feature is re-activation of latently integrated provirus, which has the potential to disrupt the temporal signal, leading to variation in the branch lengths and apparent evolutionary rates in a tree. Yet, real within-host HIV phylogenies tend to show clear, ladder-like trees structured by the time of sampling. Another important feature is recombination, which violates the fundamental assumption that evolutionary history can be represented by a single bifurcating tree. Thus, recombination complicates the within-host HIV dynamic by mixing genomes and creating evolutionary loop structures that cannot be represented in bifurcating trees. In this paper, we develop a coalescent-based simulator of within-host HIV evolution that includes latency, recombination, and effective population size dynamics that allows us to study the relationship between the true, complex genealogy of within-host HIV evolution, encoded as an Ancestral Recombination Graph (ARG), and the observed phylogenetic tree. To compare our ARG results to the familiar phylogeny format, we calculate the expected bifurcating tree after decomposing the ARG into all unique site trees, their combined distance matrix, and the overall corresponding bifurcating tree. While latency and recombination separately disrupt the phylogenetic signal, remarkably, we find that recombination recovers the temporal signal of within-host HIV evolution caused by latency by mixing fragments of old, latent genomes into the contemporary population. In effect, recombination averages over extant heterogeneity, whether it stems from mixed time-signals or population bottlenecks. Further, we establish that the signals of latency and recombination can be observed in phylogenetic trees despite being an incorrect representation of the true evolutionary history. Using an Approximate Bayesian Computation method, we develop a set of statistical probes to tune our simulation model to nine longitudinally-sampled within-host HIV phylogenies. Because ARGs are exceedingly difficult to infer from real HIV data, our simulation system allows investigating effects of latency, recombination, and population size bottlenecks by matching decomposed ARGs to real data as observed in standard phylogenies.

59 BASIC BIOLOGICAL SCIENCES↗

Large-scale Gravitational Lens Modeling with Bayesian Neural Networks for Accurate and Precise Inference of the Hubble Constant

We investigate the use of approximate Bayesian neural networks (BNNs) in modeling hundreds of time delay gravitational lenses for Hubble constant (H 0 ) determination. Here, our BNN was trained on synthetic Hubble Space Telescope quality images of strongly lensed active galactic nuclei with lens galaxy light included. The BNN can accurately characterize the posterior probability density functions (PDFs) of model parameters governing the elliptical power-law mass profile in an external shear field. We then propagate the BNN-inferred posterior PDFs into an ensemble H 0 inference, using simulated time delay measurements from a plausible dedicated monitoring campaign. Assuming well-measured time delays and a reasonable set of priors on the environment of the lens, we achieve a median precision of 9.3% per lens in the inferred H 0 . A simple combination of a set of 200 test lenses results in a precision of 0.5 km s –1 Mpc –1 (0.7%), with no detectable bias in this H 0 recovery test. The computation time for the entire pipeline—including the generation of the training set, BNN training and H0 inference—translates to 9 minutes per lens on average for 200 lenses and converges to 6 minutes per lens as the sample size is increased. Being fully automated and efficient, our pipeline is a promising tool for exploring ensemble-level systematics in lens modeling for H 0 inference.

79 ASTRONOMY AND ASTROPHYSICS↗

Optimization and stabilization of Fermilab Booster using hybrid Bayesian/RL framework

PIPII project will raise Fermilab Booster intensity and ramp rate. Beam losses will limit average power and are hard to simulate. Presently, Booster uses operator-guided empirical tuning. This task is challenging due to high dimensionality, multiple objectives, critical safety constraints, and drifts. We developed a synergistic suite of Bayesian optimization (BO) and reinforcement learning (RL) tools to optimize and stabilize beam losses. First, active learning was used to build a rough model. Data was collected parasitically using two novel safety constraint types – nonlinear input space restrictions (based on optics model), and uncertainty constraints (to stop bad steps/beam aborts). We then applied online multi-objective BO with scalarized objectives and fitting to improve/rebalance losses, increasing safety margins by 25%. Using BO model as a safety veto, we tried several on/off-policy RL agents for long term stabilization; SAC had best performance. We found that adding contextual (state) information further improved performance, eventually integrating key knobs like linac phase and temperature into the parameter space. Long term testing is ongoing to enable operational use.

Kuklev, Nikita [Fermilab]↗