Email: sam@cs.uoregon.edu

Curriculum Vitae: PDF (April 2019)

Sam's Publications, Grants, and Presentations are noted here.

Abstract: Machine learning techniques, while well established for many observation types, have only recently come onto the scene for graphs and other combinatorial objects. Further, the use and efficacy of machine learning techniques in predicting computationally difficult invariants on discrete combinatorial objects is next to unknown. This thesis outlines methodologies useful for articulating discrete structures in the paradigm of many machine learning algorithms. Moreover, we examine several NP-hard problems in different articulations. We then report on the results of various techniques and methodologies in solving certain families of these problems.

Abstract: Matrix factorization under the reals is a solved problem. Decompositions, such as LU and QR, are taught in foundational Linear Algebra courses. Computationally, the algorithms to find a factorization of a real-valued matrix are polynomial in their time complexity. Once we restrict ourselves to the elements and operations in the Boolean semiring, however, the computational complexity to find any decomposition of a matrix becomes NP Complete. This presentation will discuss ways in which we can articulate the Boolean matrix factorization problem as a partially complete image. We will then show how machine learning techniques, such as random forests and neural networks, can be applied to this representation. In the processes of doing so, we will motivate how we can find polynomial time decompositions in the amortized case.

Abstract: Much of what we do as advanced undergraduates and graduate students in STEM revolves around writing programs to solve acute research problems. Once the research problem is ameliorated by our code-fu skills, however, we are still left with the non-trivial task of publishing our software for use by others. This presentation discusses generalized practices for researchers which can facilitate this process.

Abstract: In the modern era, typewritten work is the norm in every discipline and mathematics and statistics are no exception. Nevertheless, many undergraduate institutions provide little to no training to their students on how to typeset mathematical documents. Indeed, many instructors gladly accept handwritten work even in upper-level division mathematics courses. Contrasted with many upper division humanities classes, where a handwritten essay would likely be rejected upon receipt, mathematics and statistics departments nationwide do a poor job in preparing their students for the reality that both academia and industry expect professional mathematics to be typed. Given that a non-negligible number of incoming graduate students have no experience in using the most common framework used to typeset mathematical expressions, we present a ready-made workshop framework which will ameliorate the initially steep learning curve of LaTeX.