Simin Chen Takes Home David Daniel Thesis Award for Research on Trustworthy Machine Learning

Before earning the esteemed David Daniel Thesis Award at The University of Texas at Dallas, Dr. Simin Chen began his academic journey in a completely different field: civil engineering. With no formal training in computer science, he was driven by a deep curiosity about machine learning and a strong commitment to intellectual growth. That decision ultimately led him to complete a PhD in the UT Dallas Computer Science Department and to be named one of the University’s top doctoral graduates. The David Daniel Thesis Award, presented annually to two outstanding recipients from the Erik Jonsson School of Engineering and Computer Science and the School of Natural Sciences and Mathematics, recognized Chen’s dissertation Toward More Trustworthy and Efficient Machine Learning Software for its exceptional combination of technical sophistication, practical utility and interdisciplinary insight.

Chen’s work challenges fundamental assumptions about artificial intelligence systems, addressing inefficiencies and hidden vulnerabilities in the software infrastructure that powers machine learning. His research produced tools and techniques that improve both the speed and reliability of AI applications in real-world scenarios.

“I began asking, ‘Can we trust that the model we trained is the same one being used in the real world?’” Chen said. That question guided his exploration of the entire machine learning software stack, encompassing data pipelines, model design, compilers and runtime systems.

His work resulted in a suite of tools designed to enhance both trust and efficiency in machine learning. Among his contributions are tools such as EfficFrog, a system for detecting malicious or poorly curated training data that could trigger slowdowns; NMTSloth and DeepPerform, which identify inefficiencies in adaptive models; and DyCL, a compiler optimization tool that enables machine learning models to run more securely and efficiently. He also developed NNReverse, a system designed to uncover hidden vulnerabilities in compiled AI binaries.

One of his most notable discoveries was that large language models, such as ChatGPT, can sometimes “over-reason,” generating unnecessarily long or irrelevant responses when presented with imperfect inputs. “That behavior wastes time, consumes resources and can degrade user experience,” Chen says. His early identification of this issue helped shape efforts to improve the reliability and cost-effectiveness of generative AI.

Chen attributes much of his development to the mentorship of Dr. Wei Yang, who welcomed him into his lab despite his nontraditional background and guided him throughout his PhD journey. This support played a pivotal role in shaping his approach to research. “His mentorship became the cornerstone of my development,” Chen said. “He supported me through foundational learning and encouraged me to pursue deep, practical research.”

Reflecting on his time at UT Dallas, Chen recalls the challenges of transitioning into computer science, especially without a traditional background in the field. “The early stages were particularly challenging,” he noted. “but I wanted to explore machine learning, a field that fascinated me for its growing impact in the real world.” His dedication to addressing practical problems and creating tools for real-world deployment helped him bridge that knowledge gap and find his voice as a researcher.

His PhD journey was marked by moments of growth both technical and personal. He describes late-night deadlines, including one memorable evening when he raced against the clock with his advisor to finalize a paper. “When we finally submitted the paper at 5 a.m., I stepped outside and felt a deep sense of pride, not just from finishing the paper but also from the entire journey.”

When asked what he wished he knew at the start of his PhD journey, Chen reflected, “I would tell myself: “You don’t need to have all the answers — just keep asking good questions. You grow by navigating uncertainty, embracing failure and staying connected to the purpose behind the work.”

During setbacks, he focused on progress in small steps. “Even finishing an experiment or writing a clean paragraph was enough to keep me going,” he said. Outside the lab, Chen found an unexpected source of balance and inspiration in playing Dota 2, a competitive strategy game he often turned to as a mental reset during intense research periods. “It gave me a break from research and helped reset my brain,” he said. Some of his best ideas, he recalled, came after stepping away from work to play a few rounds. The game also became a shared point of connection with his advisor Yang who enjoyed watching professional tournaments. “We bonded over matches, especially during major tournaments,” Chen added. This blend of academic rigor and lighthearted downtime helped him stay motivated and avoid burnout over the long course of his PhD work.

Today, Chen is continuing his research at Columbia University as a postdoctoral scholar. His current work focuses on optimizing the deployment of large language models and ensuring the reliability of AI-generated code. He is particularly interested in advancing certifiable machine learning systems that can be validated for safe and predictable performance.

His perspective on AI’s future remains grounded in real-world challenges. “My mission remains the same,” he said. “I want to ensure AI systems are not only powerful but also predictable, safe and grounded in the real world.”

Chen encourages aspiring researchers to pursue work that has lasting value. “Ask yourself: “Will this work still be meaningful in a few years?” he advised. “Don’t be afraid to explore unconventional paths. That’s often where the most impactful ideas come from.”

For Chen, receiving the David Daniel Thesis Award was not only an academic milestone but also was a reflection of a broader journey, a journey defined by perseverance, mentorship and a commitment to solving significant, practical problems. His story illustrates how interdisciplinary thinking and intellectual resilience can shape the future of trustworthy artificial intelligence.