Search

Effective, reliable, and scalable development of machine learning (ML) solutions for structured electronic health record (EHR) data requires the ability to reliably generate high-quality baseline models for diverse supervised learning tasks in an efficient and performant manner. Historically, producing such baseline models has been a largely manual effort--individual researchers would need to decide on the particular featurization and tabularization processes to apply to their individual raw, longitudinal data; and then train a supervised model over those data to produce a baseline result to compare novel methods against, all for just one task and one dataset. In this work, powered by complementary advances in core data standardization through the MEDS framework, we dramatically simplify and accelerate this process of tabularizing irregularly sampled time-series data, providing researchers the ability to automatically and scalably featurize and tabularize their longitudinal EHR data across tens of thousands of individual features, hundreds of millions of clinical events, and diverse windowing horizons and aggregation strategies, all before ultimately leveraging these tabular data to automatically produce high-caliber XGBoost baselines in a highly computationally efficient manner. This system scales to dramatically larger datasets than tabularization tools currently available to the community and enables researchers with any MEDS format dataset to immediately begin producing reliable and performant baseline prediction results on various tasks, with minimal human effort required. This system will greatly enhance the reliability, reproducibility, and ease of development of powerful ML solutions for health problems across diverse datasets and clinical settings.

We report a systematic comparison of niobium nitride thin films deposited on oxidized silicon substrates by reactive DC magnetron sputtering and reactive high-power impulse magnetron sputtering (HiPIMS). After determining the nitrogen gas concentration that produces the highest superconducting critical temperature for each process, we characterize the dependence of the critical temperature on film thickness. The optimal nitrogen concentration is higher for HiPIMS than for DC sputtering, and HiPIMS produces higher critical temperatures for all thicknesses studied. We attribute this to the HiPIMS process enabling the films to get closer to optimal stoichiometry before beginning to form a hexagonal crystal phase that reduces the critical temperature, along with the extra kinetic energy in the HiPIMS process improving crystallinity. We also study the ability to increase the critical temperature of the HiPIMS films through the use of an aluminum nitride buffer layer and substrate heating., Comment: 7 pages, 7 figures

Heart failure is a debilitating condition that affects millions of people worldwide and has a significant impact on their quality of life and mortality rates. An objective assessment of cardiac pressures remains an important method for the diagnosis and treatment prognostication for patients with heart failure. Although cardiac catheterization is the gold standard for estimating central hemodynamic pressures, it is an invasive procedure that carries inherent risks, making it a potentially dangerous procedure for some patients. Approaches that leverage non-invasive signals - such as electrocardiogram (ECG) - have the promise to make the routine estimation of cardiac pressures feasible in both inpatient and outpatient settings. Prior models trained to estimate intracardiac pressures (e.g., mean pulmonary capillary wedge pressure (mPCWP)) in a supervised fashion have shown good discriminatory ability but have been limited to the labeled dataset from the heart failure cohort. To address this issue and build a robust representation, we apply deep metric learning (DML) and propose a novel self-supervised DML with distance-based mining that improves the performance of a model with limited labels. We use a dataset that contains over 5.4 million ECGs without concomitant central pressure labels to pre-train a self-supervised DML model which showed improved classification of elevated mPCWP compared to self-supervised contrastive baselines. Additionally, the supervised DML model that uses ECGs with access to 8,172 mPCWP labels demonstrated significantly better performance on the mPCWP regression task compared to the supervised baseline. Moreover, our data suggest that DML yields models that are performant across patient subgroups, even when some patient subgroups are under-represented in the dataset. Our code is available at https://github.com/mandiehyewon/ssldml

Self-supervised learning (SSL) for clinical time series data has received significant attention in recent literature, since these data are highly rich and provide important information about a patient's physiological state. However, most existing SSL methods for clinical time series are limited in that they are designed for unimodal time series, such as a sequence of structured features (e.g., lab values and vitals signs) or an individual high-dimensional physiological signal (e.g., an electrocardiogram). These existing methods cannot be readily extended to model time series that exhibit multimodality, with structured features and high-dimensional data being recorded at each timestep in the sequence. In this work, we address this gap and propose a new SSL method -- Sequential Multi-Dimensional SSL -- where a SSL loss is applied both at the level of the entire sequence and at the level of the individual high-dimensional data points in the sequence in order to better capture information at both scales. Our strategy is agnostic to the specific form of loss function used at each level -- it can be contrastive, as in SimCLR, or non-contrastive, as in VICReg. We evaluate our method on two real-world clinical datasets, where the time series contains sequences of (1) high-frequency electrocardiograms and (2) structured data from lab values and vitals signs. Our experimental results indicate that pre-training with our method and then fine-tuning on downstream tasks improves performance over baselines on both datasets, and in several settings, can lead to improvements across different self-supervised loss functions., Comment: ICML 2023

Neural network models have demonstrated impressive performance in predicting pathologies and outcomes from the 12-lead electrocardiogram (ECG). However, these models often need to be trained with large, labelled datasets, which are not available for many predictive tasks of interest. In this work, we perform an empirical study examining whether training time data augmentation methods can be used to improve performance on such data-scarce ECG prediction problems. We investigate how data augmentation strategies impact model performance when detecting cardiac abnormalities from the ECG. Motivated by our finding that the effectiveness of existing augmentation strategies is highly task-dependent, we introduce a new method, TaskAug, which defines a flexible augmentation policy that is optimized on a per-task basis. We outline an efficient learning algorithm to do so that leverages recent work in nested optimization and implicit differentiation. In experiments, considering three datasets and eight predictive tasks, we find that TaskAug is competitive with or improves on prior work, and the learned policies shed light on what transformations are most effective for different tasks. We distill key insights from our experimental evaluation, generating a set of best practices for applying data augmentation to ECG prediction problems., Comment: Conference on Health, Inference, and Learning (CHIL) 2022

The impact of machine learning models on healthcare will depend on the degree of trust that healthcare professionals place in the predictions made by these models. In this paper, we present a method to provide people with clinical expertise with domain-relevant evidence about why a prediction should be trusted. We first design a probabilistic model that relates meaningful latent concepts to prediction targets and observed data. Inference of latent variables in this model corresponds to both making a prediction and providing supporting evidence for that prediction. We present a two-step process to efficiently approximate inference: (i) estimating model parameters using variational learning, and (ii) approximating maximum a posteriori estimation of latent variables in the model using a neural network, trained with an objective derived from the probabilistic model. We demonstrate the method on the task of predicting mortality risk for patients with cardiovascular disease. Specifically, using electrocardiogram and tabular data as input, we show that our approach provides appropriate domain-relevant supporting evidence for accurate predictions., Comment: ACM Conference on Health, Learning, and Inference 2021

In many domains such as medicine, training data is in short supply. In such cases, external knowledge is often helpful in building predictive models. We propose a novel method to incorporate publicly available domain expertise to build accurate models. Specifically, we use word2vec models trained on a domain-specific corpus to estimate the relevance of each feature's text description to the prediction problem. We use these relevance estimates to rescale the features, causing more important features to experience weaker regularization. We apply our method to predict the onset of five chronic diseases in the next five years in two genders and two age groups. Our rescaling approach improves the accuracy of the model, particularly when there are few positive examples. Furthermore, our method selects 60% fewer features, easing interpretation by physicians. Our method is applicable to other domains where feature and outcome descriptions are available., Comment: Presented at 2016 Machine Learning and Healthcare Conference (MLHC 2016), Los Angeles, CA

We demonstrate rapid loading of a small array of optical tweezers with a single $^{87}$Rb atom per site. We find that loading efficiencies of up to 90% per tweezer are achievable in less than 170 ms for traps separated by more than $1.7 \,\mu\text{m}$. Interestingly, we find the load efficiency is affected by nearby traps and present the efficiency as a function of the spacing between two optical tweezers. This enhanced loading, combined with subsequent rearranging of filled sites, will enable the study of quantum many-body systems via quantum gas assembly.

We investigate a method for generating nonlinear phase shifts on superpositions of photon number states. The light is stored in a Bose-Einstein condensate via electromagnetically-induced transparency memory techniques. The atomic collisions are exploited to generate a nonlinear phase shift of the stored state. The stored light is then revived with the nonlinear phase shift imprinted upon it. We show that this method can be used as a nonlinear-sign gate in the regime where the Thomas-Fermi and mean-field approximations are valid. We test these approximations using realistic parameters and find that these approximations pass the standard tests for validity in a single-component condensate. However, for the two-component condensates considered here, we find that these conditions are insufficiently strict. We find a stronger set of conditions and show for the same set of parameters that the approximations are invalid., Comment: 8 pages, 4 figures

To date, no research has examined the AB degree through a state policy lens or has acknowledged the scope of involvement of 4-year colleges and universities. This omission limits current understanding of the AB degree as a potential contributor to the higher education system and to the nation's college completion agenda in particular. The primary goals of this study were to document the extent to which AB degrees are offered by higher education institutions in the 50 states, to examine different approaches and models for implementing and awarding these degrees, and to explore the state and institutional policy contexts that surround the awarding of AB degrees by community and technical colleges and 4-year colleges and universities. The study was conducted in two phases, beginning with a national inventory of AB degrees in the U.S. public higher education sector (Phase 1), and extending into a multicase study of 6 states (Phase 2). In 2009, the authors began the second phase of the study by conducting case studies in six states (Arizona, Florida, Kentucky, Oklahoma, Texas, and Washington), which were selected because of their engagement in AB degrees, but also because of their diversity in policy and program approaches to various forms of baccalaureate degrees. In looking at the six states that were the focus of Phase 2 of the study, several themes emerged. First, despite a paucity of impact data, state and institutional administrators believe that AB degrees benefit adult learners, particularly those who are currently working or who are using the degree as a means for job advancement. Also important to this investigation was the fact that interview and online survey results confirmed that students targeted for AB degrees are overwhelmingly adult learners who are working, but this group also included unemployed or dislocated workers and active military personnel. Students of color and students with disabilities were identified by program directors responding to the authors' survey, as was also evident in their field observations and interviews. The findings lead to several conclusions about past developments in and potential of the AB. The authors offer several recommendations to continue to advance research and development concerning the AB degree. Appended are: (1) The Applied Baccalaureate Project Advisory Committee Members; (2) State-by-State Inventory; (3) Identified Applied Baccalaureate Programs in the Six Selected States; and (4) The Applied Baccalaureate (AB) Online Survey Instrument. (Contains 2 figures, 11 tables and 1 footnote.)

The evolution of AB degrees has been influenced by an increased emphasis on workforce development, baccalaureate degree attainment, and transfer on the state and institutional levels. Beginning as primarily programmatic decisions made by a few institutions, over time AB degree program decisions have involved more institutions, and eventually states. Led by Barbara Townsend until her untimely death on June 11, 2009, the authors' research team surveyed education agencies in all 50 states and found that as of 2008, AB degrees were offered in at least one public 4-year institution in 39 states. Lumina Foundation's "Big Goal 2025" (n.d.) focuses on increasing the percentage of Americans with postsecondary credentials to 60% (compared to 39% at the start of Big Goal 2025) and advocates for a policy agenda designed to improve the credentialing of adult and transfer students in the nation. To that end, Lumina encouraged and supported a "Convening" on the AB degree to bring together policy experts, practitioners, and researchers to discuss the status of the AB degree, draw on the authors' research, and encourage dialogue on the potential of and challenges to AB degrees in the future. This paper summarizes some of the most salient themes that emerged during the Convening from sharing the authors' research, hearing from panelists and presenters, and participating in discussions among attendees. In reflecting on the dialogue among participants in the Convening, four primary themes emerged that deserve further exploration in this paper: (1) Defining the AB; (2) Clarifying the relationship between the AB and the CCB; (3) Defining a baccalaureate education; and (4) Accrediting the AB degree.

In the nation's changing economy, there is an increasing necessity for baccalaureate level education for jobs that have never before required that level of education. One potential solution to issues related to baccalaureate attainment and workforce development is the applied baccalaureate degree. Applied baccalaureate degrees have arisen from a number of convergent forces to provide a bachelor's degree option for participants in Associate of Applied Science (AAS) degree or applied associate degree programs offered primarily in occupational-technical (or career-technical) education program areas. These degree programs represent a potentially important curriculum path to the baccalaureate for a significant number of postsecondary students. "The Adult Learner and the Applied Baccalaureate" project is designed to provide federal, state, and local educational leaders and policy makers with information about the applied baccalaureate degree in the United States. This policy brief draws upon results of a 50-state study to inventory applied baccalaureate degree programs. Using data obtained from telephone interviews with state officials, and from Web sites, reports, legislation, and other materials provided by the states, the authors describe the status of applied baccalaureate programs offered by public associate degree-granting and traditional baccalaureate degree-granting colleges and universities. The authors report steady growth in applied baccalaureate degrees since the first emerged in three public traditional baccalaureate degree-granting institutions in the 1970s. Six states were added in the decade of the 1980s, and nine more in the 1990s. Since 2000, fifteen states have begun offering the programs. Interviews with state officials indicate at least four other states have introduced legislation, convened task forces, or conducted hearings to gather information about the applied baccalaureate degree at associate degree-granting institutions, and other states are considering the applied baccalaureate at traditional baccalaureate degree-granting institutions. For many states and higher education institutions, this inventory suggests that the applied baccalaureate offers a means of reaching a diversity of underserved learners, including but not limited to adults, and supporting student enrollments in baccalaureate-level, occupational-technical degree programs. State and institutional rationales often emphasize the importance of using applied baccalaureate degrees to address workforce and economic needs. The authors conclude that, although some critics claim that applied baccalaureate degrees represent a threat to the integrity of the baccalaureate degree, diminishing quality and adding cost to an increasingly expensive higher education system, the applied baccalaureate represents a viable option to reaching adult learners and encouraging their participation in higher education to meet individual and larger systemic educational and economic needs. (Contains 1 map and 1 table.) [This brief summarizes results originally presented in "The Adult Learner and the Applied Baccalaureate: National and State-by-State Inventory" by Townsend, Bragg, and Ruud (2008), along with additional results concerning state and local implementation drawn from 50-state inventory.]

Without appropriate education, American workers are hampered in their ability to have the requisite technical skills needed to participate fully and effectively in the modern workforce. Higher education as a system must develop programs and infrastructures that ensure adult learners receive postsecondary education, including at the baccalaureate degree level. This report is part of a larger research initiative funded by Lumina Foundation for Education to provide federal, state, and local leaders and policymakers with up-to-date, detailed information about the applied baccalaureate phenomenon in the United States, including the extent to which applied baccalaureate programs are targeting adult learners' educational needs. Appended are: (1) Advisory Committee Members; and (2) State Profiles (constitute the majority of the report). (Contains 2 tables, 1 footnote, and 3 maps.)

This brief summarizes results of a recently published study that examined the effects of career and technical education (CTE) transition programs on student matriculation from secondary to postsecondary education. The study was divided into two components. One focused on secondary institutions' CTE transition programs and their effects on secondary student performance and college transition as well as student perceptions of the transition programs. The other component examined the postsecondary effects of the selected CTE programs on student transition and postsecondary outcomes, taking a retrospective look at students who matriculated from the specified high school CTE transition programs and comparing them to non-participants from the same sampled high schools. This analysis examined student persistence in college and whether the transition programs were related to remediation, credit attainment, number of terms attended, and certificate or degree attainment. Both components of the study examined the relationship between student outcomes and institutional engagement in CTE transition programs, a measure of institution-level commitment as evidenced by number of articulation agreements and faculty involvement in delivering CTE transition-related courses, on postsecondary education outcomes.

Unitary control of qudits can improve the collective spin squeezing of an atomic ensemble. Preparing the atoms in a state with large quantum fluctuations in magnetization strengthens the entangling Faraday interaction. The resulting increase in interatomic entanglement can be converted into metrologically useful spin squeezing. Further control can squeeze the internal atomic spin without compromising entanglement, providing an overall multiplicative factor in the collective squeezing. We model the effects of optical pumping and study the tradeoffs between enhanced entanglement and decoherence. For realistic parameters we see improvements of ~10 dB., Comment: 5 pages, 2 figures

We describe a new approach to spin squeezing based on a double-pass Faraday interaction between an optical probe and an optically dense atomic sample. A quantum eraser is used to remove residual spin-probe entanglement, thereby realizing a single-axis twisting unitary map on the collective spin. This interaction can be phase-matched, resulting in exponential enhancement of squeezing. In practice the scaling and peak squeezing depends on decoherence, technical loss, and noise. A simplified model indicates ~10 dB of squeezing should be achievable with current laboratory parameters., Comment: 4 pages, 2 figures.

The design cycle associated with large engineering systems requires an initial decomposition of the complex system into design processes which are coupled through the transference of output data. Some of these design processes may be grouped into iterative subcycles. In analyzing or optimizing such a coupled system, it is essential to be able to determine the best ordering of the processes within these subcycles to reduce design cycle time and cost. Many decomposition approaches assume the capability is available to determine what design processes and couplings exist and what order of execution will be imposed during the design cycle. Unfortunately, this is often a complex problem and beyond the capabilities of a human design manager. A new feature, a genetic algorithm, has been added to DeMAID (Design Manager's Aid for Intelligent Decomposition) to allow the design manager to rapidly examine many different combinations of ordering processes in an iterative subcycle and to optimize the ordering based on cost, time, and iteration requirements. Two sample test cases are presented to show the effects of optimizing the ordering with a genetic algorithm.