Matthew Bailey: Understanding ASL with Computers

As a computer science student at Lander, I have a variety of different computing and software topics, both in class and beyond. Knowing how software and hardware are used together to make a finished product is something I’ve done and understand well. But as a computer science student, I have always suffered from imposter syndrome; feeling like I don’t belong or that I’m as deserving as my colleagues.

One way I battle imposter syndrome is to build projects so I can feel accomplished, so when chance to further my experience in the one field I wasn’t as experienced in came, I felt compelled to do it. At Lander, I’ve had the opportunity to work with software, hardware, APIs, backend work, and even frontend work, but never machine learning.

What is machine learning? According to IBM, “[it] is a branch of artificial intelligence and computer science which focuses on the use of data and algorithms to imitate the way that humans learn, gradually improving its accuracy”. Essentially, using data and algorithms to teach a computer to look at data and make conclusions and decisions like a person would. Specifically, I’d eventually end up working with computer vision, which is doing the same things as machine learning, but with images and video. Teaching the computer to “see” in a way.

The opportunity was a machine learning seminar hosted by Dr. Gilliean Lee, who personally asked me if I’d be interested. From then on, two other students and I had weekly seminars on Fridays where we would present on machine learning. Every week we were given a chapter of information to present, getting ready to pick a project to do for submission to the SC Upstate Research Symposium. I pursued this because I wanted to experience something I hadn’t before, attempting to maximize my learning while at Lander.

Once we finally got to a point where we were confident with machine learning, we each picked our projects, mine being an American Sign Language translator. We were advised to build our own machine learning model, while a lot of projects normally use prebuilt ones. The models that we were building are the mathematical representation of how our program detects and recognizes patterns. Learning to create our own models was really rewarding and made me feel more confident in my abilities.

The goal behind doing the ASL translation was because I wanted to work towards solving accessibility issues while using computer vision, a subset of machine learning. Attempting to patch the communication between those who are deaf and those who don’t know ASL is something I felt could be assisted with technology. Doing a project like this would be a great learning experience because there has been some research done in this field, but not too much information available to keep it somewhat challenging.

While working with Dr. Lee, I learned a lot both academically and professionally. My quick thinking was challenged several times, with one being the short time given to complete the whole project (around 1.5 months), including research and pre-requisite information I needed to complete the project. There was a strict timeline, and if I wasn’t quick and effective with what I did, I would’ve fallen behind.

Another good example is the day my poster was due, where I was trying to get model statistics after one last revision. Training is where you have the model go through the data and “learn” from it and test itself on what it “learned”. During this training, the Lander machine learning server that we were using, crashed. This left me in a tight spot, having to submit a poster within an hour, with no one available to work on the server and bring it back online.

With just an hour left, I had to act quickly to get the statistics needed to put on my poster and knowing my laptop had barely any comparable power to the server, it did have a NVIDIA graphics card. Meaning that it had machine learning training capabilities. I learned quickly from my textbooks and online research how to enable the training capabilities. Training my full model would’ve taken about 2 hours, so I chopped my model down a bit, so it only took about 20 minutes, while still being just as accurate as it was previously, just trained on less data. This was one of the biggest tests of my quick thinking and acting in my life.

As it got closer to the day of presenting, it was sad to see the other students drop out last minute due to issues that arose for them. They were unable to finish due to internal problems they faced, and it helped me learn that no matter how much work you put into something, it can always be taken right out from under you.

When it came to the day for presenting, I was humbled by the amazing work of those around me. For example, while I was setting up my poster, I met the group next to me, who used machine learning to detect fraud in financial settlements. People had assured me I was doing something cool but meeting all these people doing such incredible things made me feel so much more like an imposter. The only thing that helped was the slight validation from knowing that I was accepted, just as they all were.

When presenting, I had to adapt quickly to the multitude of questions that were asked of me, all varying levels of expertise that were needed. This kept me on my toes and prepared me for the Lander Computer Science Symposium, where I was asked questions about the inner workings of my project.

Experiencing a symposium like the SC Upstate Academic Symposium was amazing, as I got to ask about others’ work, I had people ask about my work which felt good. At the end of the day, while I did not win any awards, this was still a really rewarding experience. I was able to prove to myself that I could learn and do whatever I wanted if I put my mind and dedication into it. I was able to get beneficial public speaking experience in a non-classroom setting.

Additionally, I discovered my passion for Data Science and Machine Learning and introduced new career paths to me. This project also showed me the importance of having a good mentor to learn under. My breakaway wasn’t like other students who completed internships or followed more traditional approaches, but I still believe this breakaway experience was life-changing for me.

After my breakaway, my options for careers expanded and my eyes were opened to the extent of what I could do in the future. This breakaway ensured my confidence about working in the field of data science and machine learning.

Matthew Bailey is from Lexington, SC and will graduate in May 2023 with a Bachelor of Science in Computer Information Systems. He spent the Spring 2023 semester working under Dr. Gilliean Lee doing machine learning, submitting a project for the SC Upstate Academic Symposium. After graduating, he plans to pursue a career in technology, which no favor of what specific field.