Independent Research in Medical Image Computing (Co-Authored Publication)

First Author; Under Review at ICASSP 2026

• Developed MambaDATG, a domain-adaptive tri-plane-gated framework modeling anisotropy in 3D CT with linear-time efficiency

• Improved segmentation accuracy on BTCV benchmark from 80.16% to 81.95% mean Dice (+1.79%), achieving state-of-the-art performance on small organs and vascular structures such as pancreas and hepatic vessels

• Created a lightweight domain adaptation stage for cross-scanner robustness without inference overhead

Supervised by Dr. Ruoxuan Xiong, Emory University

First Author; Under Review at KDD 2026

• Proposed the TMC-MNR framework with cross-modal attention, category transfer, and learnable temporal re-weighting to address ultra-sparse recommendation and modality non-random missingness (MNAR)

• Developed an MNAR-debiasing training scheme combining modality selection heads and inverse probability weighting, enabling robust multimodal fusion without additional inference cost

• Achieved SOTA on Amazon Review 2023 (R@20=0.1176, P@20=0.0080, NDCG@20=0.0459; +29.2%/+63.3%/+18.3%)

Supervised by Dr. Ruoxuan Xiong, Emory University

First Author; Accepted at EMNLP 2025 Main

• Developed CRL-MMNAR, a causal multimodal framework that models modality missing-not-at-random (MMNAR) to learn robust patient representations from structured data, CXR, notes, and radiology reports

• Designed an MMNAR-aware fusion module with missingness embeddings plus cross-modal reconstruction and contrastive objectives, and a multitask predictor with a rectifier for bias correction across cohorts (MIMIC-IV, eICU—208 hospitals)

• Delivered consistent SOTA gains: AUC +8.4% (readmission), +13.1% (ICU admission), +4.7% (mortality) on MIMIC-IV; +13.8% (readmission) and +0.5% (mortality) on eICU, with lower Brier scores

Independent Research in Game Theory (Co-Authored Publication)

First Author; Under Review at TRB Annual Meeting 2026

• Constructed a three-tier hierarchical Stackelberg game integrating government policy, platform R&D, and labor re-skilling, enabling equilibrium analysis of autonomous taxi adoption

• Designed and implemented GPU-accelerated Bellman iterations with neural function approximation to solve an 11-dimensional dynamic game under uncertainty

• Conducted large-scale simulations and Monte Carlo robustness tests, generating tractable policy guidance on balancing innovation, employment, and public trust