Spectextor Published IEEE Research Paper

In the summer of 2021, I participated in a remote REU program with a focus on AI at the University of Maryland, Baltimore County. Over this time, I worked on 2 projects. The first was a cryptocurrency assitant chatbot designed to respond to questions about cryptocurrencies. Three weeks into the ten week program, I told the program head, Nirmalya Roy, that I felt unengaged and unfulfilled with the work, as I was not on the AI side of the project and the mismanagement of the project meant I could not even speak to the professor in charge of the project.

I wanted to learn about and develop zero-shot learning chatbots, not a frontend user interface. In addition, the team I was working with clearly saw the project more as a potential business endeavor rather than academic research (research being the other primary experience I was hoping to get out of the program).

After describing my displeasure with my REU experience so far, Professor Roy was attentive and put me in his lab under the guidance of graduate student, Indrajeet Ghosh. Over the last seven weeks of the program, I started, developed, and wrote a paper ready for publication about a GenAI model designed to take sports video as input and output dense captions describing the video.

The project began with a survey on state-of-the-art video-to-text technology, benchmark testing methodologies, and sports video captioning datasets. This was followed by designing, developing, training, and testing various models. All findings can be found at the paper linked below.

Due to losing three weeks of the program to my first project, the timeline for this paper was truncated significantly. It addition to the paper, by week nine of the program (six weeks into this project) I already had results and created a poster presentation to present at the end of the program. After I left the program, the team submitted the paper for publication to the 2022 IEEE International Conference on Smart Computing. It was accepted.