skip to main content

Title: Homogeneity Structure Learning in Large-scale Panel Data with Heavy-tailed Errors
Large-scale panel data is ubiquitous in many modern data science applications. Conventional panel data analysis methods fail to address the new challenges, like individual impacts of covariates, endogeneity, embedded low-dimensional structure, and heavy-tailed errors, arising from the innovation of data collection platforms on which applications operate. In response to these challenges, this paper studies large-scale panel data with an interactive effects model. This model takes into account the individual impacts of covariates on each spatial node and removes the exogenous condition by allowing latent factors to affect both covariates and errors. Besides, we waive the sub-Gaussian assumption and allow the errors to be heavy-tailed. Further, we propose a data-driven procedure to learn a parsimonious yet flexible homogeneity structure embedded in high-dimensional individual impacts of covariates. The homogeneity structure assumes that there exists a partition of regression coeffcients where the coeffcients are the same within each group but different between the groups. The homogeneity structure is flexible as it contains many widely assumed low dimensional structures (sparsity, global impact, etc.) as its special cases. Non-asymptotic properties are established to justify the proposed learning procedure. Extensive numerical experiments demonstrate the advantage of the proposed learning procedure over conventional methods especially when the data are more » generated from heavy-tailed distributions. « less
; ;
Award ID(s):
2015539 1953196 1820702
Publication Date:
Journal Name:
Journal of machine learning research
Page Range or eLocation-ID:
Sponsoring Org:
National Science Foundation
More Like this
  1. We study high-dimensional signal recovery from non-linear measurements with design vectors having elliptically symmetric distribution. Special attention is devoted to the situation when the unknown signal belongs to a set of low statistical complexity, while both the measurements and the design vectors are heavy-tailed. We propose and analyze a new estimator that adapts to the structure of the problem, while being robust both to the possible model misspecification characterized by arbitrary non-linearity of the measurements as well as to data corruption modeled by the heavy-tailed distributions. Moreover, this estimator has low computational complexity. Our results are expressed in the formmore »of exponential concentration inequalities for the error of the proposed estimator. On the technical side, our proofs rely on the generic chaining methods, and illustrate the power of this approach for statistical applications. Theory is supported by numerical experiments demonstrating that our estimator outperforms existing alternatives when data is heavy-tailed.« less
  2. Canonical polyadic decomposition (CPD) has been a workhorse for multimodal data analytics. This work puts forth a stochastic algorithmic framework for CPD under β-divergence, which is well-motivated in statistical learning—where the Euclidean distance is typically not preferred. Despite the existence of a series of prior works addressing this topic, pressing computational and theoretical challenges, e.g., scalability and convergence issues, still remain. In this paper, a unified stochastic mirror descent framework is developed for large-scale β-divergence CPD. Our key contribution is the integrated design of a tensor fiber sampling strategy and a flexible stochastic Bregman divergence-based mirror descent iterative procedure, whichmore »significantly reduces the computation and memory cost per iteration for various β. Leveraging the fiber sampling scheme and the multilinear algebraic structure of low-rank tensors, the proposed lightweight algorithm also ensures global convergence to a stationary point under mild conditions. Numerical results on synthetic and real data show that our framework attains significant computational saving compared with state-of-the-art methods.« less
  3. The actual failure times of individual components are usually unavailable in many applications. Instead, only aggregate failure-time data are collected by actual users, due to technical and/or economic reasons. When dealing with such data for reliability estimation, practitioners often face the challenges of selecting the underlying failure-time distributions and the corresponding statistical inference methods. So far, only the exponential, normal, gamma and inverse Gaussian distributions have been used in analyzing aggregate failure-time data, due to these distributions having closed-form expressions for such data. However, the limited choices of probability distributions cannot satisfy extensive needs in a variety of engineering applications.more »PHase-type (PH) distributions are robust and flexible in modeling failure-time data, as they can mimic a large collection of probability distributions of non-negative random variables arbitrarily closely by adjusting the model structures. In this article, PH distributions are utilized, for the first time, in reliability estimation based on aggregate failure-time data. A Maximum Likelihood Estimation (MLE) method and a Bayesian alternative are developed. For the MLE method, an Expectation-Maximization algorithm is developed for parameter estimation, and the corresponding Fisher information is used to construct the confidence intervals for the quantities of interest. For the Bayesian method, a procedure for performing point and interval estimation is also introduced. Numerical examples show that the proposed PH-based reliability estimation methods are quite flexible and alleviate the burden of selecting a probability distribution when the underlying failure-time distribution is general or even unknown.« less
  4. Need/Motivation (e.g., goals, gaps in knowledge) The ESTEEM implemented a STEM building capacity project through students’ early access to a sustainable and innovative STEM Stepping Stones, called Micro-Internships (MI). The goal is to reap key benefits of a full-length internship and undergraduate research experiences in an abbreviated format, including access, success, degree completion, transfer, and recruiting and retaining more Latinx and underrepresented students into the STEM workforce. The MIs are designed with the goals to provide opportunities for students at a community college and HSI, with authentic STEM research and applied learning experiences (ALE), support for appropriate STEM pathway/career, preparationmore »and confidence to succeed in STEM and engage in summer long REUs, and with improved outcomes. The MI projects are accessible early to more students and build momentum to better overcome critical obstacles to success. The MIs are shorter, flexibly scheduled throughout the year, easily accessible, and participation in multiple MI is encouraged. ESTEEM also establishes a sustainable and collaborative model, working with partners from BSCS Science Education, for MI’s mentor, training, compliance, and building capacity, with shared values and practices to maximize the improvement of student outcomes. New Knowledge (e.g., hypothesis, research questions) Research indicates that REU/internship experiences can be particularly powerful for students from Latinx and underrepresented groups in STEM. However, those experiences are difficult to access for many HSI-community college students (85% of our students hold off-campus jobs), and lack of confidence is a barrier for a majority of our students. The gap between those who can and those who cannot is the “internship access gap.” This project is at a central California Community College (CCC) and HSI, the only affordable post-secondary option in a region serving a historically underrepresented population in STEM, including 75% Hispanic, and 87% have not completed college. MI is designed to reduce inequalities inherent in the internship paradigm by providing access to professional and research skills for those underserved students. The MI has been designed to reduce barriers by offering: shorter duration (25 contact hours); flexible timing (one week to once a week over many weeks); open access/large group; and proximal location (on-campus). MI mentors participate in week-long summer workshops and ongoing monthly community of practice with the goal of co-constructing a shared vision, engaging in conversations about pedagogy and learning, and sustaining the MI program going forward. Approach (e.g., objectives/specific aims, research methodologies, and analysis) Research Question and Methodology: We want to know: How does participation in a micro-internship affect students’ interest and confidence to pursue STEM? We used a mixed-methods design triangulating quantitative Likert-style survey data with interpretive coding of open-responses to reveal themes in students’ motivations, attitudes toward STEM, and confidence. Participants: The study sampled students enrolled either part-time or full-time at the community college. Although each MI was classified within STEM, they were open to any interested student in any major. Demographically, participants self-identified as 70% Hispanic/Latinx, 13% Mixed-Race, and 42 female. Instrument: Student surveys were developed from two previously validated instruments that examine the impact of the MI intervention on student interest in STEM careers and pursuing internships/REUs. Also, the pre- and post (every e months to assess longitudinal outcomes) -surveys included relevant open response prompts. The surveys collected students’ demographics; interest, confidence, and motivation in pursuing a career in STEM; perceived obstacles; and past experiences with internships and MIs. 171 students responded to the pre-survey at the time of submission. Outcomes (e.g., preliminary findings, accomplishments to date) Because we just finished year 1, we lack at this time longitudinal data to reveal if student confidence is maintained over time and whether or not students are more likely to (i) enroll in more internships, (ii) transfer to a four-year university, or (iii) shorten the time it takes for degree attainment. For short term outcomes, students significantly Increased their confidence to continue pursuing opportunities to develop within the STEM pipeline, including full-length internships, completing STEM degrees, and applying for jobs in STEM. For example, using a 2-tailed t-test we compared means before and after the MI experience. 15 out of 16 questions that showed improvement in scores were related to student confidence to pursue STEM or perceived enjoyment of a STEM career. Finding from the free-response questions, showed that the majority of students reported enrolling in the MI to gain knowledge and experience. After the MI, 66% of students reported having gained valuable knowledge and experience, and 35% of students spoke about gaining confidence and/or momentum to pursue STEM as a career. Broader Impacts (e.g., the participation of underrepresented minorities in STEM; development of a diverse STEM workforce, enhanced infrastructure for research and education) The ESTEEM project has the potential for a transformational impact on STEM undergraduate education’s access and success for underrepresented and Latinx community college students, as well as for STEM capacity building at Hartnell College, a CCC and HSI, for students, faculty, professionals, and processes that foster research in STEM and education. Through sharing and transfer abilities of the ESTEEM model to similar institutions, the project has the potential to change the way students are served at an early and critical stage of their higher education experience at CCC, where one in every five community college student in the nation attends a CCC, over 67% of CCC students identify themselves with ethnic backgrounds that are not White, and 40 to 50% of University of California and California State University graduates in STEM started at a CCC, thus making it a key leverage point for recruiting and retaining a more diverse STEM workforce.« less
  5. Cuntz, Hermann (Ed.)
    Cellular barcoding methods offer the exciting possibility of ‘infinite-pseudocolor’ anatomical reconstruction—i.e., assigning each neuron its own random unique barcoded ‘pseudocolor,’ and then using these pseudocolors to trace the microanatomy of each neuron. Here we use simulations, based on densely-reconstructed electron microscopy microanatomy, with signal structure matched to real barcoding data, to quantify the feasibility of this procedure. We develop a new blind demixing approach to recover the barcodes that label each neuron, and validate this method on real data with known barcodes. We also develop a neural network which uses the recovered barcodes to reconstruct the neuronal morphology from themore »observed fluorescence imaging data, ‘connecting the dots’ between discontiguous barcode amplicon signals. We find that accurate recovery should be feasible, provided that the barcode signal density is sufficiently high. This study suggests the possibility of mapping the morphology and projection pattern of many individual neurons simultaneously, at high resolution and at large scale, via conventional light microscopy.« less