Artificial intelligence and machine learning (AI/ML) can do a lot of things, but can it predict how well people will like a movie? During the PEARC24 conference, contestants in an ACCESS mini hackathon event hosted by ACCESS Support and the SCIPE/CIP teams were given the task of testing this theory. Using Python and Sklearn, participants were given 50,000 movie reviews from IMDB to create a supervised ML environment designed to predict positive or negative sentiment with regard to specific movies.
The hackathon was designed to bring in all types of participants, experienced or brand new to AI/ML. Participants were shown introductory material via presentations that gave concrete examples of how to solve the problem. Then, participants were given three challenge levels as options for submitting a proposed solution. Evan Jaffe, a machine-learning engineer from the Ohio Supercomputing Center (OSC), explained, “The challenges included: 1. Experiential, which focuses on concepts and problem-solving without requiring coding, 2. Best performance on the sentiment analysis task and 3. Most interesting/creative approach to the task. The performance and creativity challenges required more coding and ML background and were designed to be open-ended to accommodate advanced practitioners. We hope we were able to provide something of value to attendees at all levels of participation.”
The event was well-received, and three winners came out on top. Maryam Berijanian, a Ph.D. student at Michigan State University, took home a prize for “Best Originality/Creativity” in the Advanced category. Berijanian has been researching AI and deep learning, including exploring natural language processing (NLP) and computer vision, which is a specialty within AI research focused on training computers to interpret visual input, like photos or video.
“Attending the AI Hackathon was an amazing experience,” said Berijanian. “It pushed me to put my AI knowledge to the test for the task of sentiment analysis on movie data. I’m really thankful for the chance to participate, not just because it was fun, but also because I got to learn alongside so many talented people. I’m excited to take what I’ve learned and apply it in the future.”
Akua Sekyiwaa Osei-Nkwantabisa, a graduate student at the University of Texas Rio Grande Valley, is also studying AI/ML. She won in the Advanced category and took home the Best Performance award. She found the hackathon an exciting opportunity to try out the skills she’d been learning as she worked toward her degree.
“Being at the ACCESS Support-SCIPE/CIP AI Mini-Hack was an incredible experience that gave me the opportunity to apply my knowledge in a practical setting,” Osei-Nkwantabisa said. “Working on sentiment analysis for movie data challenged me to dive deep into model tuning. I am especially grateful for the chance to enjoy this experience with other talented participants. I also enjoyed the insightful sessions, from the Introduction to AI presentation to the hands-on Mini-Hack.”
While Osei-Nkwantabisa was ready for the challenge, she was still surprised by her strong results. “I never expected to be a winner,” she said. “As someone who has not taken extensive projects on sentiment analysis, this hackathon gave me the opportunity to learn more about the multilayer perceptron and how to make it more efficient. There were countless times when I waited nearly an hour for the model to be trained, only to find it performing at an average level. I’m glad I did not give up. Through perseverance, I continued training the model and finally got the best performing model.”
In the Experiential category, the prize went to Jane Elizabeth Herriman, who currently works at Lawrence Livermore National Lab. Her work focuses on HPC education and HPC user enablement. “I was excited to attend the AI Hackathon at PEARC24 because understanding the basics of AI/ML is important for users’ needs as ML becomes more important to their scientific workflows,” she said. Herriman was particularly impressed with the contest’s approachability for people with varying experience levels.
“This hackathon did a great job of providing tutelage/background on AI and engaging the audience with activities,” she said. “The opening presentation provided a framework for thinking about AI, including helpful imagery that clarified the relationships between different techniques and concepts in AI. The hackathon had both conceptual and hands-on components that made it possible for attendees of various skill levels to participate. I’m really glad I had the chance to attend and hope to see this content at other conferences!”
Winners received a $3,000 travel grant. ACCESS Support hosts several events throughout the year that offer opportunities for those interested in HPC. To keep abreast of new opportunities, subscribe to the ACCESS Advance newsletter or join the CSSN.