I had a kick-off meeting with Dr. Bruce Maxwell on June 14th, and it was a productive start to the project. Interestingly, he graduated from Swarthmore College, which is not far from my alma mater, Franklin and Marshall College (F&M). We connected on this shared geographical background, as he mentioned his experience swimming at F&M, which made me feel a sense of familiarity.

During the meeting, we established a regular meeting time for the remainder of the project period, which is every Thursday afternoon at 2 pm. Dr. Maxwell introduced me to his ongoing research projects, all centered around the evaluation of different input data types for computer vision tasks. These projects include object recognition, color constancy, adversarial images, super-resolution, and intrinsic imaging. Deep learning plays a significant role in all of these projects, which immediately caught my interest. Object recognition, in particular, resonates with me as it is closely related to my daily life, from the facial recognition feature on my phone to the technology used in veterinary medicine for identifying farm animal ear tags.

One intriguing concept that I learned about for the first time is adversarial attacks. Dr. Maxwell explained that instead of embedding a plaintext message in an input image, adversarial attacks involve embedding a noise vector purposely designed to deceive deep learning models. This noise vector aims to confuse the models and serves as a way for hackers to bypass certain filters. It’s fascinating how such techniques can manipulate the behavior of machine learning systems.

While we didn’t delve into the specifics of my research project during this meeting, Dr. Maxwell advised me to familiarize myself with the basics of computer vision. I would like to build a solid foundation before delving deeper into the project, considering my limited background in computer vision and machine learning. To facilitate this, he recommended a website for me to gain a basic understanding of convolutional neural networks: Towards Data Science. Additionally, he suggested exploring Jason Brownlee’s blog, Machine Learning Mastery, which offers a more accessible approach to understanding machine learning concepts. Unlike traditional machine learning courses that can be dry and math-heavy, Brownlee’s blog focuses on practical tutorials with working code.

In addition, Dr. Maxwell emphasized that I should become proficient in PyTorch, as it will be the primary library and framework used. I plan to dedicate the second week to self-teaching and familiarizing myself with PyTorch. In addition to the recommended online resources, Dr. Maxwell also pointed out valuable references such as the computer vision textbook written by Richard Szeliski, available at szeliski.org, as well as Dr.Maxwell his own lecture notes and ongoing textbook-in-writing, “Fundamentals of Computer Vision.”

Looking ahead to the second week, my primary focus will revolve around acquiring the aforementioned foundational knowledge in computer vision and setting up the necessary URLs required by the DREAM program.

I am excited about the journey ahead and the opportunity to expand my knowledge in the field of computer vision.