Apoorva Bisht
Author: Apoorva Bisht | Majors: Computer Science and Physics | Semester: Summer 2022
Deep-learning and AI hold immense significance in identifying objects and classifying them. AI based models have been incorporated into many devices like smartphones, cameras etc. More often we see the applications of deep-learning at macro scale. In this project we propose to use deep-learning at nano scale i.e., to detect 2D flakes of materials and identify their thickness. In material research, many times the first majorstep is to identify useful flakes of the material which can be used for making devices. These flakes are characterized by their size, thickness and thickness gradient. Traditionally, 2D flake “hunting” process has been done via physical methods which are extremely time consuming and expensive. We recognize that microscope images of 2D flakes show distinct colors due to changes in thickness. We plan to exploit this property using a deep-learning based application that will be trained using a large dataset to learn how to identify the flakes in a microscope image and then determine their thickness based on their color variation.
Choosing this project for my honors computer science research came quite naturally. As a double major in physics and computer science, I was looking for a project that would align well with interests from both of my majors. The physics lab that I have been a part of focuses on quantum photonics and materials and is also part of the MonArk Quantum Foundry (an NSF program that is jointly led by Montana State University and the University of Arkansas with the overarching mission to accelerate two-dimensional materials research for quantum technologies in the US). So, I was looking for a project that would deal with computer science applications in physics, specifically optics/material. Then I came across CVIU (Computer Vision and Image Understanding) lab supervised by Dr. Khoa Luu who is one of the collaborators in the MonArk Quantum Foundry. As I went through the project, it immediately seemed to connect my physics and computer science interests. Although not very old concepts, but AI and deep-learning have seen their foothold in many areas. They are now seeing a spread in micro and nano scale applications. This to me is very exciting since at such small scales, it is usually very hard to quickly find what we are looking for, whether it be cells with specific mutations or 2D materials flakes with specific characteristics. All these applications require hours of taxing labor and expertise in analyzing the sample under microscope. And it so happens, that we can make AI based applications work well with repetitive data that has a trend. Hence, this project seemed promising and interesting and I was ready to venture into a new area!
I appreciate the guidance of Xuan Bac Nguyen and our supervisor Dr. Khoa Luu. I was completely new to AI when I came to the lab, and gradually I have been learning about different processes involved in developing a successful model. Regular meetings with lab members and collaborators of this project have helped me to understand the problem better and present what I have learnt. As part of the project, I had the opportunity to present the progress made in one semester at the MonArk NSF Virtual Site. At the only undergraduate student to present at the site visit, it was encouraging to converse with the experts in the field from different universities.
The entire process so far from data collection and countless hours spent on AFM (Atomic Force Microscopy) to seeing how the trained model is able to detect and identify thickness of samples is very rewarding. I look forward to continue this project next year as I complete my honors thesis. This entire experience has allowed me to develop an understanding of what I want to do in future and what type of projects I would like to work on.