Ngoc Minh Tri Nguyen (THIRD PLACE)

Instruments of scientific discovery (e.g., electron microscopes, scanning probe microscopes, and others) acquire vast collections of images that contain physical insight. Humans, however, struggle to search and draw correlations from enormous databases of images; thus, only a small fraction of the data collected is translated into knowledge. Researchers need an analytical toolbox that can create an image similarity manifold and allow visual interaction in an informative, comprehensive, and intuitive way. Here, we develop machine learning algorithms to create image similarity projections of microscopy images. We implement pre-trained VGG-16 convolutional neural network optimized on the ImageNet dataset to extract high dimensional features of size [1x1x4096]. We use t-Distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP), manifold unfolding techniques, to generate interpretable 2D projections of these features. We are currently creating an interactive image viewer where researchers could actively explore the data. While we demonstrate the aptitude of this approach on piezoresponse force microscopy images, this approach is amenable to other forms of microscopy and imaging techniques.

Tri Nguyen is a junior pursuing a B.S. in Materials Science and Engineering. He performs undergraduate research in Dr. Josh Agar’s lab through the support of Nano/Human Interface Presidential Engineering Research Initiative, and the National Science Foundation. Most recently, his research has focused on developing machine learning algorithms for image similarity in piezoelectric force microscopy. Outside of academics, Tri is a Rossin Junior Fellow and plays the clarinet in the Marching 97. He is also involved with LU Diplomats where he interviews international students via Skype and provides write-ups for Lehigh Undergraduate Admissions. After graduation, he plans to complete a PhD program in Materials Science and Engineering and pursue his goal of developing materials to enhance and enable devices for a sustainable future.