UC Berkeley Previously Published Works

Deep Generative Models for Fast Photon Shower Simulation in ATLAS

(2024)

Abstract: The need for large-scale production of highly accurate simulated event samples for the extensive physics programme of the ATLAS experiment at the Large Hadron Collider motivates the development of new simulation techniques. Building on the recent success of deep learning algorithms, variational autoencoders and generative adversarial networks are investigated for modelling the response of the central region of the ATLAS electromagnetic calorimeter to photons of various energies. The properties of synthesised showers are compared with showers from a full detector simulation using geant4. Both variational autoencoders and generative adversarial networks are capable of quickly simulating electromagnetic showers with correct total energies and stochasticity, though the modelling of some shower shape distributions requires more refinement. This feasibility study demonstrates the potential of using such algorithms for ATLAS fast calorimeter simulation in the future and shows a possible way to complement current simulation techniques.

Software Performance of the ATLAS Track Reconstruction for LHC Run 3

(2024)

Charged particle reconstruction in the presence of many simultaneous proton–proton (pp) collisions in the LHC is a challenging task for the ATLAS experiment’s reconstruction software due to the combinatorial complexity. This paper describes the major changes made to adapt the software to reconstruct high-activity collisions with an average of 50 or more simultaneous pp interactions per bunch crossing (pile-up) promptly using the available computing resources. The performance of the key components of the track reconstruction chain and its dependence on pile-up are evaluated, and the improvement achieved compared to the previous software version is quantified. For events with an average of 60pp collisions per bunch crossing, the updated track reconstruction is twice as fast as the previous version, without significant reduction in reconstruction efficiency and while reducing the rate of combinatorial fake tracks by more than a factor two.

Efficiency of ETA Prediction

(2024)

Estimating the effect of realistic improvements of metformin adherence on COVID-19 mortality using targeted machine learning.

(2024)

BACKGROUND: Type 2 diabetes elevates the risk of severe outcomes in COVID-19 patients, with multiple studies reporting higher case fatality rates. Metformin is a widely used medication for glycemic management. We hypothesize that improved adherence to metformin may lower COVID-19 post-infection mortality risk in this group. Utilizing data from the Mexican Social Security Institute (IMSS), we investigate the relationship between metformin adherence and mortality following COVID-19 infection in patients with chronic metformin prescriptions. METHODS: This is a retrospective cohort study consisting of 61,180 IMSS beneficiaries who received a positive polymerase chain reaction (PCR) or rapid test for SARS-CoV-2 and had at least two consecutive months of metformin prescriptions prior to the positive test. The hypothetical intervention is improved adherence to metformin, measured by proportion of days covered (PDC), with the comparison being the observed metformin adherence values. The primary outcome is all-cause mortality following COVID-19 infection. We defined the causal parameter using shift intervention, an example of modified treatment policies. We used the targeted learning framework for estimation of the target estimand. FINDINGS: Among COVID-19 positive patients with chronic metformin prescriptions, we found that a 5% and 10% absolute increase in metformin adherence is associated with a respective 0.26% (95% CI: -0.28%, 0.79%) and 1.26% (95% CI: 0.72%, 1.80%) absolute decrease in mortality risk. INTERPRETATION: Subject to the limitations of a real-world data study, our results indicate a causal association between improved metformin adherence and reduced COVID-19 post-infection mortality risk.

Estimating geographic variation of infection fatality ratios during epidemics.

(2024)

OBJECTIVES: We aim to estimate geographic variability in total numbers of infections and infection fatality ratios (IFR; the number of deaths caused by an infection per 1,000 infected people) when the availability and quality of data on disease burden are limited during an epidemic. METHODS: We develop a noncentral hypergeometric framework that accounts for differential probabilities of positive tests and reflects the fact that symptomatic people are more likely to seek testing. We demonstrate the robustness, accuracy, and precision of this framework, and apply it to the United States (U.S.) COVID-19 pandemic to estimate county-level SARS-CoV-2 IFRs. RESULTS: The estimators for the numbers of infections and IFRs showed high accuracy and precision; for instance, when applied to simulated validation data sets, across counties, Pearson correlation coefficients between estimator means and true values were 0.996 and 0.928, respectively, and they showed strong robustness to model misspecification. Applying the county-level estimators to the real, unsimulated COVID-19 data spanning April 1, 2020 to September 30, 2020 from across the U.S., we found that IFRs varied from 0 to 44.69, with a standard deviation of 3.55 and a median of 2.14. CONCLUSIONS: The proposed estimation framework can be used to identify geographic variation in IFRs across settings.

Understanding Multiprogram Take-Up of Safety Net Programs Among California Families.

(2024)

INTRODUCTION: The U.S. safety net, which provides critical aid to households with low income, is composed of a patchwork of separate programs, and many people with low income benefit from accessing <1 program. However, little is known about multiprogram take-up, that is, participation conditioned on eligibility. This study examined individual and multiprogram take-up patterns and sociodemographic factors associated with multiprogram take-up of U.S. safety net programs. METHODS: The Assessing California Communities Experiences with Safety Net Supports study interviewed Californians and reviewed their 2019 tax forms between August 2020 and May 2021. Take-up of safety net programs was calculated among eligible participants (n=365), including the Earned Income Tax Credit; Supplemental Nutrition Assistance Program; the Special Supplemental Nutrition Program for Women, Infants, and Children; and Medicaid. Multivariable regressions identified sociodemographic factors associated with take-up of multiple programs. RESULTS: Take-up was highest for Medicaid (90.6%) and lowest for Supplemental Nutrition Assistance Program (57.5%). Among people who received benefits from at least 1 other program, take-up ranged from 81.7% to 84.8% for the Earned Income Tax Credit; 54.4%-62.0% for Supplemental Nutrition Assistance Program; 74.3%-80.1% for Special Supplemental Nutrition Program for Women, Infants, and Children; and 89.7%-98.1% for Medicaid. Having a lower income and being younger were associated with concurrent take-up of Supplemental Nutrition Assistance Program and Special Supplemental Nutrition Program for Women, Infants, and Children. Among Supplemental Nutrition Assistance Program and Special Supplemental Nutrition Program for Women, Infants, and Children recipients, having higher income, being older, and being primarily English speaking were associated with Earned Income Tax Credit take-up. CONCLUSIONS: Individual and multiprogram take-up vary between programs and by sociodemographic factors. Findings suggest opportunities to increase take-up of potentially synergistic programs by improving cross-program coordination, data sharing, and targeted recruitment of underenrolled subgroups (Supplemental Nutrition Assistance Program and Special Supplemental Nutrition Program for Women, Infants, and Children).

Cover page of Gender, racial-ethnic, and socioeconomic disparities in the development of social-emotional competence among elementary school students

Gender, racial-ethnic, and socioeconomic disparities in the development of social-emotional competence among elementary school students

(2024)

Social-emotional competence (SEC) has been demonstrated to be a crucial factor for student mental health and is malleable through the high-quality implementation of effective school-based social and emotional learning (SEL) programs. SEL is now widely practiced in the United States as a Tier 1 strategy for the entire student body, yet it remains unclear whether disparities exist in the development of SEC across socioculturally classified subgroups of students. Also, despite the field’s widespread concern about teacher bias in assessing SEC within diverse student bodies, little evidence is available on the measurement invariance of the SEC assessment tools used to explore and facilitate SEC development. Based on a sociocultural view of student SEC development, this study aimed to measure and examine the extent to which gender, racial-ethnic, and socioeconomic disparities exist in SEC developmental trajectories during elementary school years. Specifically, using 3 years of SEC assessment data collected from a districtwide SEL initiative (N = 5452; Grades K–2 at baseline; nine measurement occasions), this study (a) tested the measurement invariance of a widely-used, teacher-rated SEC assessment tool (DESSA-Mini) across student gender, race and ethnicity, and socioeconomic status (SES); and (b) examined the extent to which multiyear SEC growth trajectories differed across these subgroups under a routine SEL practice condition. The invariance testing results supported strict factorial invariance of the DESSA-Mini across all the examined subgroups, thereby providing a foundation for valid cross-group comparisons of student SEC growth. The piecewise latent growth modeling results indicated that boys (vs. girls), Black students (vs. White students), Hispanic students (vs. White students), and low-income students (vs. middle-to-high-income students) started with a lower level of SEC, with these gaps being sustained or slightly widened throughout 3 elementary school years. Based on these findings, this study calls for future research that can inform practice efforts to ensure equitable SEC assessments and produce more equitable SEL outcomes, thereby promoting equity in school mental health.

A Numerical Simulation of PFPE Lubricant Kinetics in HAMR Air Bearing

(2024)

Abstract: This report investigates the kinetics of lubricant molecules in the HAMR air bearing to understand the initiation and growth of PFPE contamination on the head surface. The collisions with the air bearing induce three forces—drag, thermophoresis, and lift. Of these, we find that lift forces are negligible. Then, a sensitivity analysis of the remaining two forces reveals the conditions where they dominate. Further, a hybrid simulation strategy is utilized to track their movements. The results show that the contaminations (smear) highly depend on the interplay between the thermophoresis and drag forces. We then explain the mechanism of the formation of the various observed patterns. Finally, we offer some recommendations to exploit the air bearing to contain smear on the head.

Encoding and context-dependent control of reward consumption within the central nucleus of the amygdala.

(2024)

Dysregulation of the central amygdala is thought to underlie aberrant choice in alcohol use disorder, but the role of central amygdala neural activity during reward choice and consumption is unclear. We recorded central amygdala neurons in male rats as they consumed alcohol or sucrose. We observed activity changes at the time of reward approach, as well as lick-entrained activity during ongoing consumption of both rewards. In choice scenarios where rats could drink sucrose, alcohol, or quinine-adulterated alcohol with or without central amygdala optogenetic stimulation, rats drank more of stimulation-paired options when the two bottles contained identical options. Given a choice among different options, central amygdala stimulation usually enhanced consumption of stimulation-paired rewards. However, optogenetic stimulation during consumption of the less-preferred option, alcohol, was unable to enhance alcohol intake while sucrose was available. These findings indicate that the central amygdala contributes to refining motivated pursuit toward the preferred available option.