5/27/2026 Cassandra Smith
University of Illinois ECE doctoral candidate Hyungyo Kim has been named a 2026 ML and Systems Rising Star by MLCommons, an international machine learning performance benchmarking organization. Working through the IBM-Illinois Discovery Accelerator Institute (IIDAI), Kim has developed groundbreaking research in large language model (LLM) efficiency, GPU-CPU heterogeneous computing, and memory-efficient AI training — all aimed at reducing the cost of deploying and training large language models.
Written by Cassandra Smith
Training and operating today’s most powerful AI models costs a fortune. One Ph.D. candidate within The Grainger College of Engineering at the University of Illinois Urbana-Champaign is working to change that and was internationally recognized for his work.
Hyungyo Kim is a Ph.D. candidate in the Electrical and Computer Engineering department. His advisor is professor Nam Sung Kim, the W.J. Jerry Sanders III Advanced Micro Devices, Inc. Endowed Chair in electrical and computer engineering.
Hyungyo was selected as a 2026 ML and Systems Rising Star, awarded by MLCommons. The award recognizes emerging researchers working in efficient machine learning and LLM systems.
Kim’s research is about making large language models (LLMs) cheaper and faster. He worked on this topic through several projects conducted through the IBM-Illinois Accelerator Institute. “All of them were centered around how to make LLMs, how to train and how to deploy LLMs cost-efficiently,” said Kim.
The first project is regarding heterogeneous GPU-CPU computing leveraging the latest AI accelerators on CPUs. “All of the traditional deployment of LLMs mainly use GPUs only, but we explored to leverage recent CPU’s on-chip AI accelerator in conjunction with GPUs during LLM inference. Doing so makes the hardware cost much lower.”
The second project is regarding LLM compression. They losslessly compressed LLM parameters to reduce memory consumption. They used CPU’s on-chip decompressions accelerator when needed for fast decompression. The result is significantly improved memory efficiency.
The third project was initiated during last summer’s IBM internship, according to Kim. It addresses the massive GPU memory demands of training LLMs. “During training, you always have this memory issue,” said Kim. “Basically, the memory that’s required for training these models is so large that you must use enormous amount of GPUs.” They are trying to leverage Nvidia’s UVM technology to offload memory needs to CPU memory.
Kim said his work with IBM has set him up for success. “IBM has provided so much infrastructure, mentorship and guidance in shaping the direction of this research,” said Kim. “They provided a lot of very insightful comments on where the research should be heading and what it should be solving.”
Collaboration is the key to Kim’s achievement and his earning of the Rising Star title. “This is all thanks to the collaboration with IBM and my advisor, Nam Sung Kim. I really want to recognize their tremendous guidance and all the support that they gave.” Kim continued to say his advisor taught him “how to research, how to think and how to conduct experiments.”
As Kim continues to push the boundaries of what’s possible in AI efficiency, his path offers a glimpse of where the field may be headed—toward a future where powerful AI systems do not solely belong to big organizations. As Kim presses towards expanding his IBM projects, his Rising Star designation acts as a springboard to help him further his goal of making technology better and more accessible for everyone.
Grainger Engineering Affiliations
Nam Sung Kim is an Illinois Grainger Engineering professor of electrical and computer engineering in the Department of Electrical and Computer Engineering and the Siebel School of Computing and Data Science. He is affiliated with the Coordinated Science Laboratory and the National Center for Supercomputing Applications. Kim holds the W.J. ‘Jerry’ Sanders III – Advanced Micro Devices, Inc. Endowed Chair Professor appointment.