arXiv subjects

Harry Bendekgey

Publications and source records attributed to Harry Bendekgey.

2 recordsLinked to original sources

arXiv2024

A Systematic Literature Review of Undergraduate Data Science Education Research

The presence of data science has been profound in the scientific community in almost every discipline. An important part of the data science education expansion has been at the undergraduate level. We conducted a systematic literature review to (1) portray current evidence and knowledge gaps in self-proclaimed undergraduate data science education research and (2) inform policymakers and the data science education community about what educators may encounter when searching for literature using the general keyword 'data science education.' While open-access publications that target a broader audience of data science educators and include multiple examples of data science programs and courses are a strength, significant knowledge gaps remain. The undergraduate data science literature that we identified often lacks empirical data, research questions and reproducibility. Certain disciplines are less visible. We recommend that we should (1) cherish data science as an interdisciplinary field; (2) adopt a consistent set of keywords/terminology to ensure data science education literature is easily identifiable; (3) prioritize investments in empirical studies.

stat.OT↗

arXiv2023

Unbiased Learning of Deep Generative Models with Structured Discrete Representations

By composing graphical models with deep learning architectures, we learn generative models with the strengths of both frameworks. The structured variational autoencoder (SVAE) inherits structure and interpretability from graphical models, and flexible likelihoods for high-dimensional data from deep learning, but poses substantial optimization challenges. We propose novel algorithms for learning SVAEs, and are the first to demonstrate the SVAE's ability to handle multimodal uncertainty when data is missing by incorporating discrete latent variables. Our memory-efficient implicit differentiation scheme makes the SVAE tractable to learn via gradient descent, while demonstrating robustness to incomplete optimization. To more rapidly learn accurate graphical model parameters, we derive a method for computing natural gradients without manual derivations, which avoids biases found in prior work. These optimization innovations enable the first comparisons of the SVAE to state-of-the-art time series models, where the SVAE performs competitively while learning interpretable and structured discrete data representations.

cs.LG↗