Research

Papers, manuscripts, and theses on machine learning for clinical and other high-stakes data. A complete list is on Google Scholar.

01Publications

2026

EMNLP Findings

Two-Sided State-Space Models for Sequential Recommendation with Non-Random Multimodal Review Feedback

Ziwen Pan*, Zihan Liang*, Ruoxuan Xiong

  • Recommender Systems
Paper Code

Jointly models how user preferences and item states evolve in sequential recommendation, treating whether and how users write multimodal reviews as non-random feedback.

2026

ACL Findings

DART: Mitigating Harm Drift in Difference-Aware LLMs via Distill-Audit-Repair Training

Ziwen Pan*, Zihan Liang*, Jad Kabbara, Ali Emami

  • LLM Safety
Paper Code

Finds that fine-tuning safety-tuned LLMs for difference-aware classification raises accuracy but makes their rationales more harmful, and introduces a distill–audit–repair pipeline that improves accuracy while keeping explanations safe.

2026

ACL Findings

Learning Dynamic Representations and Policies from Multimodal Clinical Time-Series with Informative Missingness

Zihan Liang*, Ziwen Pan*, Ruoxuan Xiong

  • Clinical ML
Paper Code

Learns patient states from multimodal ICU time series by treating what gets recorded, and when, as signal, and uses these states for both outcome prediction and offline treatment-policy learning.

2026

ICASSP

Oral

MambaDATG: Domain-Adaptive Tri-Plane-Gated Pre-training for 3D Abdominal Segmentation

Yaomin Shen*, Dongming Jiang*, Zihan Liang*, Yangbo Wei*, Wenkai Yang, Xiaoxin Sun, Zhen Huang, Suhua Wang, Qingsong Yao

  • Medical Imaging
Paper

Pre-trains a state-space model for 3D abdominal CT segmentation that fuses axial, coronal, and sagittal views through voxel-level gating, improving small-organ segmentation at no extra inference cost.

2026

TRB Annual Meeting

Dynamic Policy Design for Autonomous Taxi Adoption: A Hierarchical Game-Theoretic Framework

Zihan Liang, Ziwen Pan

  • Transportation Policy
Paper

Models autonomous-taxi adoption as a three-tier game among government, platform, and workers, and derives dynamic policies that balance innovation incentives with social welfare.

2025

EMNLP Main

Causal Representation Learning from Multimodal Clinical Records under Non-Random Modality Missingness

Zihan Liang*, Ziwen Pan*, Ruoxuan Xiong

  • Clinical ML
Paper Code

Treats which modalities are missing from clinical records as signal, and corrects the bias that this non-random missingness introduces into prediction.

2025

AAAI ICWSM Workshop

1st in Task 5

CareLab at #SMM4H-HeaRD 2025: Insomnia Detection and Food Safety Event Extraction with Domain-Aware Transformers

Zihan Liang*, Ziwen Pan*, Sumon Kanti Dey, Azra Ismail

  • Health NLP
Paper Code

Domain-aware transformer systems for insomnia detection in clinical text and food-safety event extraction, ranking first on Task 5 of the shared task.

02Manuscripts

2025–

Under review · IEEE T-ITS

Adaptive Spatiotemporal Graph Neural Networks with Trend-Aware Prediction and Validation-Gated Calibration for Traffic Flow Forecasting

Zihan Liang*, Ziwen Pan*, Shuyang Yu

  • Spatiotemporal Forecasting

An adaptive spatiotemporal graph network for traffic forecasting, with a calibration step that is applied only when it improves validation error.

03Theses and Dissertations

2026

Emory Honors Thesis

Highest Honors

Causal Representation Learning under Informative Missingness for Clinical Multimodal Prediction and Offline Decision-Making

Zihan Liang | Advised by Prof. Ruoxuan Xiong

  • Clinical ML
Paper Slides Poster

Brings the clinical work above into one causal representation learning framework that models informative missingness at two scales, which modalities a patient has and how closely they are monitored, for both prediction and offline decision-making.

04Collaborative Research Projects

2024–2026

Sep. 2024 – May 2026

Research Assistant at Collective Action & Research for Equity (CARE) Lab, Emory University

  • Data for Social Good
  • Maintain a WhatsApp education bot (Python, Twilio) for MakerGhat, an India-based organization, that helps 15,000+ teachers document lesson plans and teaching activities, reaching 600,000+ students and 10,000+ educational projects.
  • Analyzed nearly 1,000 user data points in an interactive dashboard (JavaScript, Python, HTML) and built an offline-first Android app for classroom session recording with Google Sheets validation, offline caching, and secure Drive sync.

2025

Jun. 2025 – Aug. 2025

Research Assistant at Polymath Jr REU Summer Program

  • Epidemic Modeling
  • Developed and analyzed ordinary differential equation models of the community transmission of Clostridioides difficile (C. difficile), implementing simulations in MATLAB and R.
  • Worked with the team to evaluate mitigation strategies and identify optimal interventions for controlling community spread.

* These authors contributed equally to this work.