CV Download CV (PDF)

LLM Benchmark Auditing — Surface-Feature Leakage (TruthfulQA-476)

• Showed that the binary-choice TruthfulQA benchmark is separable by answer style alone: a question-blind six-feature model (SURFACE6) hits 68.9% accuracy (AUC 0.715) — above the best model on the llm-stats TruthfulQA leaderboard (66.9%); auditing 13 additional benchmarks finds similar leakage, most severe on HaluEval QA (AUC 0.973).
• Designed Audit-Prune, a classifier-cleaner algorithm, and released TruthfulQA-476, a cleaned split that cuts audit AUC to near-chance (0.528) while preserving model rankings (Spearman ρ = 0.915) (code).
• Paper: Judging by the Cover: Cleaning LLM Truthfulness Benchmarks to Avoid Surface-Level Feature Leakage — arXiv 2026, under review at NeurIPS 2026 (arXiv:2609.13003).

LLM Activation Steering and Robustness under Dataset Corruption

• Prototyped verbosity/tone steering in Llama 3.1 and Qwen-2.5 via residual-stream injections (layer/α sweeps; unpublished).
• Built a guardrailed LLM-judge evaluation pipeline with JSON-schema-validated outputs, rejecting malformed judge responses.
• Made steering resilient to noisy and adversarial dataset corruption (Llama 3.2, Mistral 7B, OLMo 2; six behaviors) via robust high-dimensional mean estimation, in collaboration with Martian AI.
• Released three public contrast datasets on Hugging Face (conciseness–verbosity, positivity–negativity, formal–informal) for steering and evaluation.
• Paper: Understanding and Mitigating Dataset Corruption in LLM Steering — 2026 (arXiv:2603.03206).

Efficient Multi-Dimensional Two-Sample Testing (dKS)

• Extended the Kolmogorov–Smirnov distance to multiple dimensions: a unit-invariant metric (IPM) with a near-linear O(n log n) ε-approximate algorithm.
• Shipped as an open-source C++/pybind11 library — ~76,000× faster than exact O(n²) computation at ~1M points in 2D (code, project page).
• Paper: Efficient and Stable Multi-Dimensional Kolmogorov–Smirnov Distance — submitted to Foundations of Data Science; authors listed alphabetically (arXiv:2504.11299).

MotionPI — Privacy-First Wearable Health Sensing Platform

• Sole developer of the mobile app, backend, and offline-first sync layer (smartphone app → API → database) within a multi-institution study team; wristband firmware by Ohio State collaborators — capturing longitudinal, in-the-wild participant data end to end.
• Streamed wristband ENMO, accelerometry and survey signals to the phone with offline-first, schema-validated sync, with raw high-rate PPG/IMU archived on-device; sustained ~7.7M records/day with zero malformed writes.
• Built an ENMO threshold-calibration visualizer for rapid activity-trigger tuning and data-quality review (code); filed and diagnosed an upstream Flutter SDK packaging bug, triaged P2 by the Flutter team (issue).
• Built the participant-monitoring report the study team uses for weekly compliance follow-ups; typical per-participant wristband records rose from ~400–600 to 700–800, with some participants reaching 1,000 (study-team reported), recovering data that would otherwise be silently missing.
• Paper: Designing a Secure and Resilient Distributed Smartphone Participant Data Collection System — EAI SmartSP 2025, Springer LNICST vol. 699, pp. 283–293 (doi, arXiv:2510.19938).

Anomaly Detection — Region-Aggregated Spatial Scan Statistics

• Replaced SaTScan-style centroids with 20–50 points sampled per region: higher detection power, scanning 3,711 region polygons spanning all 3,108 continental U.S. counties in 0.33 s (~3,000× faster than FlexScan, scan step only) (code).
• Paper: Sampling for Region-Aggregated Spatial Scan Statistics — ACM SIGSPATIAL 2026, accepted as a full paper (58/248, 23.4%) (arXiv:2607.01451).

Computer Vision — High-Dimensional Spectral–Spatial Change Detection

• Created a dual-stream 3D/2D CNN with SE attention; accuracy >96%, κ > 0.9, and lower false positives vs. baselines.
• Paper: A Hyperspectral Change Detection (HCD-Net) Framework Based on Double Stream Convolutional Neural Networks and an Attention Module — Remote Sensing 16(5):827, 2024 (DOI).

Ph.D. in Computer Science

• Advisor: Prof. Jeff Phillips.
• Expected graduation: May 2028.
• GPA: 3.9/4.0
• Focus: LLM interpretability; high-dimensional geometric data analysis.
• Relevant coursework
  • Machine Learning
  • Probabilistic Machine Learning
  • Deep Learning
  • Data Mining
  • Visualization

M.Sc. in Computer Science (Algorithms & Computation)

• GPA: 17.74/20
• Thesis: Graph-theoretic modeling of bushfire propagation.
• Relevant coursework
  • Advanced Algorithms
  • Approximation Algorithms
  • Randomized Algorithms
  • Quantum Algorithms & Computation
  • Graph Algorithms
  • Network Science
  • Internet Algorithms
  • Distributed Systems

Graduate Fellowship

Top-Talent Scholarship (merit)

M.Sc. University Entrance Exam — Rank 29/20,000 (top 0.15%)

Talk — Designing a Secure and Resilient Distributed Smartphone Participant Data Collection System

Poster — Efficient and Stable Multi-Dimensional Kolmogorov–Smirnov Distance (dKS)