Seohwa Hwang
Seohwa Hwang
Postdoctoral Researcher  ·  Statistics

How do we turn everything hidden in the data into better decisions?

My research centers on multiple testing under dependence — developing procedures that exploit, rather than ignore, the correlation structure present in data. When observations are not independent, dependence is often treated as a nuisance; I am interested in using it as a resource for more powerful inference.

At the same time, I care about accessibility. Accounting for dependence often makes a method more complex — yet I strive to keep my methods simple enough that they can actually be used in practice.

Currently a postdoctoral researcher at the Department of Statistics, Seoul National University.

scroll for cv
Curriculum Vitae
Background
Download full CV (PDF)
Positions
2026–
Postdoctoral Researcher
Dept. of Statistics, Seoul National University
2020–2022
Registered Nurse
Seoul National University Hospital · Surgical ICU · OB/GY
Education
2022–2026
Ph.D. in Statistics
Seoul National University · Advisor: Prof. Junyong Park
Thesis: Online False Discovery Rate Control with Dependent Data Streams
2020–2022
M.S. in Public Health
Bioinformatics · Seoul National University · Advisor: Prof. Seongho Won
Thesis: The Application of PSM on SCCS and the Effect of Time-Dependent Confounding
2012–2018
B.A. in Nursing
Minor: Statistics & Bioethics · Seoul National University
Thesis: 입원 아동 보호자의 병원 내 환경적 요인에 대한 주관적 인식과 수면의 질의 관계 — S병원 어린이병원을 중심으로
Grants & Funding
2026–2029
Online FDR Control for High-dimensional Data Streams Achieving Structural Adaptation and Computational Feasibility
Post-doc. 성장형 연구지원 (단독연구 유형) · National Research Foundation of Korea
Awards & Honors
2026
Silver Reviewer
ICML 2026
2023
대학원생 논문발표상 3등
한국통계학회 학술논문발표회
2015
우수상
입원 아동 보호자의 병원 내 환경적 요인에 대한 주관적 인식과 수면의 질의 관계 — S병원 어린이병원을 중심으로
서울대학교 간호과학연구소 · 2015학년도 간호대학생 연구발표회
Etc
2018
Teacher's Certificate (Level II) — Health
Ministry of Education, Republic of Korea
2020
Registered Nurse License
Korea Health Personnel Licensing Examination Institute
Selected Papers
ICML 2026
Box Thirding: Anytime Best Arm Identification under Insufficient Sampling
Hwang, S. and Park, J.
Biometrical Journal
Two-Stage Multiple Test Procedures Controlling FDR with Auxiliary Variable
Hwang, S. et al.
Research
Efficient & Easy Statistical
Decision Making

How do we make statistically valid decisions as efficiently as possible, while keeping the method simple enough to use?

Efficient & Easy —
Dependence
Rather than treating correlation across observations as a nuisance to be corrected, I study how the dependence structure itself can be used to sharpen inference. This spans arbitrary dependence, stationary ergodic processes, and spatially or temporally correlated signals.
Streaming Data
Streaming data are never fully observed, which makes standard procedures hard to apply — one is forced either to assume independence or to fall back on overly conservative tests. In practice such data are typically dependent, and I aim to use that dependence to recover statistical power.
online inferenceanytime-valide-processes
Covariate-Assisted Inference
Using side information — covariates carried alongside each hypothesis — for more precise statistical inference without sacrificing validity.
side informationauxiliary variable
Assumption-Lean
The more assumptions a method requires, the more a user has to check and worry about before applying it. I aim for procedures that rest on as few assumptions as possible, so they can be used without much deliberation.
minimal assumptionsease of use
Decision making —
Multiple Testing
Testing many hypotheses at once while keeping the false discovery rate under control. My work develops FDR procedures that remain valid under dependence, often built on e-values and local-significance (LIS-type) constructions rather than independence assumptions.
Sequential Testing
Making testing decisions as data arrive, rather than only after collection ends. This includes converting offline procedures to online ones and controlling error across multiple streams that may themselves be dependent.
online FDRbanditsbest-arm identification
Papers

Papers

research at a glance — keyword frequency
2026
ICML 2026
Box Thirding: Anytime Best Arm Identification under Insufficient Sampling
Hwang, S. and Park, J.
2026
Biometrical Journal
Two-Stage Multiple Test Procedures Controlling FDR with Auxiliary Variable and their Application to Set4Δ Mutant Data
Hwang, S., Ramos, M.L., Park, D., Park, J., Lim, J. and Green, E.
2022
Clinical Gastroenterology and Hepatology
Nonalcoholic Fatty Liver Disease for the Incidence of Drug-Induced Liver Injury
Hwang, S., Won, S. and Lee, S.
—
Submitted
Online FDR Controlling Procedures for Statistical SIS Model
Hwang, S. and Park, J.
—
Submitted
FDR Control via Neural Networks under Covariate-Dependent Symmetric Nulls
Kim, T., Hwang, S. and Park, J.
—
Submitted
Online Anomaly Detection: Revisiting Storey's Procedure with Adaptive Quantile Sketching
Hwang, S. and Park, J.
Presentations

Presentations

Conference
2026
Online Testing with Dependent Streaming Data
EcoSta 2026 · Invited Talk, Organized Session "Statistical Inference and Learning"
2023
Two-Stage Multiple Test Procedures Controlling FDR with Auxiliary Variable
Korean Statistical Society Conference · 한국통계학회 학술논문발표회
Invited Seminars
2026
Online FDR Control with Dependent Data Streams
Invited Seminar · Academia Sinica · Taipei, Taiwan
2026
Online FDR Control with Dependent Data Streams
Invited Seminar · Inha University · Incheon, Korea
Workshop
2026
Post-hoc Selection of Significance Level
Workshop on Robust Inference for High-Dimensional Complex Data · SNU HDMT Lab · Siheung, Korea