Student: Tyler French
Project: Solvent-cast 3D Printing with Different Molecular Weight Polymers
Poster: Vertical (PDF) | Horizontal (PDF)
Institution: Lehigh University
Major: Materials Science and Engineering
Advisor: Lesley Chow
Abstract
Subspace clustering is the task of partitioning data that belongs to a collection of unknown subspaces into groups (i.e., clusters) such that all members of each group belong to the same subspace. The data to be clustered is unlabeled (i.e., the subspaces to which they belong are not assumed to be known in advance) and therefore must be learned, and the different subspaces are allowed to be of different dimensions. There are many real world applications that can benefit from advances in subspace clustering such as video segmentation and face clustering under different lighting conditions in computer vision. We used Julia, a fast dynamic language, to successfully cluster data from multiple randomly generated subspaces. To solve the subspace clustering problem, we considered an optimization modeling formulation closely related to the Lasso optimization problem to compute coefficient vectors that encode the “similarity” between pairs of data points. The Lasso problem, defined as the “Least Absolute Shrinkage and Selection Operator” problem, aims to regularize the coefficients computed by “shrinking” the least important coefficients to zero. The set of coefficient vectors are collected into a sparse matrix, which is then used to compute the clustering of the data. We tested our algorithm using synthetic data also created in Julia.
About Tyler French
Tyler French, a junior majoring in materials science and engineering with a biotechnology minor studying at Lehigh University. Works under the guidance of Dr. Lesley Chow focusing on characterizing the effect of polymer molecular weight on the mechanical properties of solvent-cast 3D-printed scaffolds. Member of the Lehigh men’s cross country team, track and field team, and Flight 45 serving on the Student Athlete Council.