How do you teach artificial intelligence to forget? That’s the question driving Justin Lee, a Computer Science and Engineering major at The Ohio State University. After participating in the Undergraduate Research Apprenticeship Program (URAP) last summer under the mentorship of Harry Chao, Lee continued exploring to one of AI’s most complex problems: preventing models from generating harmful or copyrighted content.
“My research focuses on image generation models and unlearning algorithms,” Lee explains. “These algorithms train models to avoid generating certain concepts. The challenge I studied is what happens when these algorithms are applied repeatedly the model’s ability to generate unrelated images can degrade, sometimes to the point where it can’t generate anything at all.”
Lee’s work doesn’t stop at identifying the problem. His paper proposes methods to reduce this degradation, ensuring models can unlearn multiple concepts while preserving their creative capabilities.
For Lee, research is about more than results, it’s about discovery.
“The most rewarding part is reaching that point where everything clicks and I truly grasp what’s going on,” he says. “From there, exploring creative solutions becomes exciting.”
His experience in URAP gave him the foundation to tackle these challenges and taught him a critical lesson: assumptions can be misleading.
“You don’t really know how something works until you test it thoroughly,” Lee reflects. “Research constantly reminds me to avoid jumping to conclusions and to let the full results guide my understanding.”
