Professor Eun Ryung Lee of the Department of Statistics at Sungkyunkwan University received the 2025 2nd  Korea Statistical Researcher of the Year in recognition of her paper “Efficient Functional Lasso Kernel Smoothing for High-Dimensional Additive Regression”, which was published in Annals of Statistics in August 2024. The award ceremony was held on August 28, 2025, at the 14th National Statistics Development Forum in Seoul. This award, presented by Statistics Korea, honors outstanding researchers who have made significant contributions to the development of statistics.

The paper addresses a fundamental challenge in modern data analysis: how to identify truly important variables and accurately estimate their nonlinear effects when the number of variables is much larger than the sample size. In ultra-high-dimensional settings, it is difficult to achieve variable selection, flexible modeling, computational feasibility, and statistical inference at the same time. This study provides a new solution to that problem.

Professor Lee and her collaborators developed a new kernel-based methodology that combines functional Lasso with smooth backfitting. The proposed method can automatically select important variables while flexibly estimating their effects through nonlinear functions, and it is supported by both computationally efficient algorithms and rigorous theoretical analysis. In addition, the study introduces a debiased inference procedure, making it possible not only to improve prediction accuracy but also to construct confidence intervals and conduct hypothesis testing.

The proposed method was further applied to large-scale gene expression and anticancer drug response data from cancer cell lines, where it showed strong empirical performance and successfully identified biologically meaningful genes associated with drug response. The study is expected to have broad impact in bioinformatics, precision medicine, finance, environmental science, and other fields where high-dimensional data are increasingly common. This award recognizes both the originality and the practical importance of Professor Lee’s contribution to modern statistical methodology.

▲ A three-stage graphical abstract — "input → method → output"

summarizing the core idea of fLasso-SBF

• Left (Input) A grid of scatter plots for many candidate covariates, with the total number allowed to be much larger than the sample size. Most covariates (grey) carry essentially no information about the response and behave like pure noise, while only the three highlighted ones (red, green, blue) are truly active with genuine nonlinear effects. This depicts the high-dimensional sparse additive regression setting that motivates the paper.
• 
  

• Middle (Method) The proposed fLasso-SBF method minimizes an objective that combines kernel-based smooth backfitting with a functional Lasso penalty. Its solution is obtained by iteratively applying a simple "soft-threshold + projection" update, which adds only a single thresholding step to the standard smooth backfitting algorithm and keeps both the implementation and the theoretical analysis clean.
• 
  
• Right (Output) Overlay of the estimated component functions produced by fLasso-SBF. Only the three genuinely active components are recovered as smooth curves, while the estimates for the remaining inactive covariates are automatically shrunk to zero. Variable selection and nonparametric function estimation are thus carried out simultaneously in a single procedure, and the accompanying debiased version further enables pointwise confidence intervals and hypothesis testing.