Where statistics meets AI
SOCAL Lab — Statistical Optimization & Collaborative AI Learning at Kyungpook National University (KNU).
Prof. Nathaniel Kang (강네이트)
Assistant Professor · Department of Computer Science and Engineering (CSE) · Kyungpook National University (KNU)
SOCAL Lab accepts M.S., Ph.D., and undergraduate researchers each cycle in federated learning, synthetic data, and AI alignment (LLM).
If you are interested, please visit Prospective students.
Principal investigator
About the PI
From Irvine and Seoul to Daegu — bridging rigorous statistics with secure, real-world AI.
I am an Assistant Professor in the Department of Computer Science and Engineering (CSE), Kyungpook National University, starting March 2026. Before KNU, I was a Postdoctoral Fellow in Statistics at Yonsei University and a Researcher at UCLA.
My research group works on federated learning (data privacy), synthetic data, and AI alignment (LLM) with statistical rigor. See Prospective students for funding and how to inquire.
연합학습·합성 데이터·AI 정렬(LLM)을 통계적 엄밀성과 함께 연구합니다. 재정·문의 방법은 Prospective students 페이지를 참고해 주세요.
PhD
Yonsei University
Statistics and Data Science
Statistics and Data Science
MS
University of California, Los Angeles
Master of Science in Engineering
Master of Science in Engineering
BS
UCLA & NTU
Computer Science / Statistics
Computer Science / Statistics
Research program
What we work on
Three threads — one goal: models and systems you can trust when data are messy, private, or rare.
Federated learning & privacy
Robust aggregation and learning when local data are non-IID and never leave the device.
See detailsImbalanced data
Calibration, synthesis, and regression when the tail of the distribution matters most.
See detailsAI alignment (LLM)
Auditing and privacy frameworks as LLMs and public-sector AI scale in the wild.
See detailsScholarship
Publications
Selected peer-reviewed work. Full bibliography on Google Scholar.
2026
Rare but Critical: Copula-Calibrated Augmentation for Imbalanced Tabular Regression
Conditional Diffusion Models for Imbalanced Tabular Regression
Orchestrating Federated Tabular Synthesis at Scale with Fault-Tolerant Pipeline Execution
Calibrated mixup for imbalanced regression on tabular data
2025
Optimizing federated learning: Addressing key challenges in real-world applications
Generalized additive models for mixed-data regression using informal data
Re-sampling calibrated SNN loss: A robust approach to non-IID data in federated learning
2024
Online news-based economic sentiment index
Working papers
Manuscripts under review or in active revision for resubmission — titles may change with revision.
Journals: venues name the current submission target and may change after revision or resubmission. Conferences: while a paper is under anonymous (double-blind) review, we do not list the conference or track name on this page.
Under review (journals)
Active journal submissions in editorial portals.
Prediction Intervals for Mixed-Frequency Nowcasts under Publication Delay
First-Release Nowcasts Need the Right Engine, Not Another Lag Scheme
Density-Stratified Distributionally Robust Optimization with Localized Risk Guarantees
Minimax-Optimal Interpolation in Multi-Source Regression Under Response Heterogeneity
CrossSynth: Scalable Federated Synthetic Data Generation with Single-Round Differentially Private Marginal Aggregation
Internal Auditing of Composed Privacy Loss Across the IoT Data Lifecycle
Not All Large Language Model Deployments Are Created Equal: A Taxonomy-Driven Survey of Security, Defense, and Governance
Rank-Informed Calibration of Text-Based Indices
What Survey Statisticians Can Teach Federated Learning about Non-IID Data
Neighbors Know Best: Manifold Augmentation for Tabular Regression
When Rare Targets Bend the Denoising Path: A Score-Energy View of Imbalanced Tabular Regression
Imputing the Present: How Streaming Diffusion Models Enable Real-Time Tabular Nowcasting
Transporting Head Mass to Rare Targets: Stratum-Marginal Schrödinger Bridges for Imbalanced Tabular Regression
Making Gradient Boosting Care About Rare Targets
Making Synthetic Tables Semantically Valid via LLM Constraints
One Privacy Budget to Rule Them All: Synthesizing Multi-Table Databases Under Differential Privacy
When Rare Targets Collapse Geometry: Bilevel Spectral Control for Deep Imbalanced Regression
Mending the Tail: Residual-Zero Students for Imbalanced Regression
From Local Mixing to Global Consistency via Barycentric Calibration
A Guarantee-Based Comparison of Authenticity Mechanisms for Generative Media
TailWeave: Radial-Angular Tail Synthesis for Imbalanced Tabular Regression
Doubly Robust Federated Learning under Selective Labeling
Rare Rows Need Anchors: Prototype-Guided Augmentation for Tabular Imbalanced Regression
Noise That Hides Heterogeneity: Information Limits of Federated Learning under Stochastic Measurements
TRG-RCP: A Tail-Risk-Gated Residual Correction Portfolio for Frozen Tabular Regressors
The Certificate, Not the Engine: Conformal Plausibility for Official-Statistics Synthesizers
Keeping Small Models Safe: A Privacy Survey
Measuring the Implementation Gap: A Framework for Auditing Policy Erosion in Public Sector Technology Procurement
Calibrating Non-IID Federated Models with Conditional Cryptographic Privacy
Under review (conferences)
Well-Calibrated Yet Unfair: How Standard Conformal Prediction Fails Along the Predicted-Value Range
The Interpolation Cliff: Phase Transitions in Personalized Federated Regression Under Label Skew
Federated Learning Is Blind to Rare Targets Without Label-Distribution Awareness
When Rare Becomes Common in Streaming Imbalanced Regression
Sequential Target-Conditional Distributionally Robust Optimization for Imbalanced Regression Streams with Moving Rarity
Join the lab
Prospective students
What we offer
Scoped projects, dedicated compute, and support for strong ML and application venues. Financial support may be available through the Brain Korea (BK) scholarship program and the PI’s personal research funding (tuition and living expenses where applicable, per university rules).
BK(Brain Korea) 장학금과 교수 개인 연구비로 재정 지원(규정 범위 내 등록금·생활비 등).
-
Financial support Brain Korea (BK) scholarship where eligible, plus PI personal research funding for tuition and living on funded projects.
-
Dedicated compute Lab nodes including high-memory NVIDIA RTX 5090 GPUs for model training.
-
Horizontal culture We keep the lab short on class. Respect is the default, not something juniors earn by staying quiet.