The Cost of Robustness: Tighter Bounds on Parameter Complexity for Robust Memorization in ReLU Nets
We analyze the parameter bounds for robust memorization as a function of the robustness ratio.
🫧My name is Yujun Kim, and I am a Ph.D. candidate in the Optimization & Machine Learning (OptiML) group at KAIST AI. I am interested in optimization and deep learning theory. I joined KAIST AI as a Master/Ph.D. integrated program in Fall 2024. Previously, I completed my undergraduate program in Mathematical Sciences with a double major in School of Computing at KAIST.
Education
M.S./Ph.D. integrated in Artificial Intelligence
Korea Advanced Institute of Science and Technology (KAIST)
B.S. in Mathematical Sciences, B.E. in School of Computing
Korea Advanced Institute of Science and Technology (KAIST)
Korea Science Academy of KAIST (KSA)
Research Interest
I am interested in optimization and deep learning theory. I am currently researching on parameter-free optimization and other recently proposed matrix (and tensor) structured optimizers.
Featured Publications
Recent News
Continuing studies as Ph.D. integrated
I am continuing my studies as a first-year Ph.D. student from Fall 2025
Best Paper Awards for KAIA 2025
I got best paper awards for KAIA 2025 for the paper “The Cost of Robustness: Tighter Parameter Complexity for Robust Memorization in ReLU Nets”
The Cost of Robustness: Tighter Bounds on Parameter Complexity for Robust Memorization in ReLU Nets
We analyze the parameter bounds for robust memorization as a function of the robustness ratio.
Incremental Gradient Descent with Small Epoch Counts is Surprisingly Slow on Ill-Conditioned Problems
We study Incremental Gradient Descent in the small epoch regime and show that it exhibits severe slowdown especially in the presence of nonconvex components.
Recent Publications
(2025).
The Cost of Robustness: Tighter Bounds on Parameter Complexity for Robust Memorization in ReLU Nets.
In NeurIPS 2025.
(2025).
The Cost of Robustness: Tighter Bounds on Parameter Complexity for Robust Memorization in ReLU Nets.
In HiLD, ICML 2025.
Recent & Upcoming Talks
