Realistic healthcare prediction tasks using AgentDS datasets.
New Program
Data Challenge Program @ IEEE ICHI 2026 - AgentDS Healthcare
Human-AI collaboration in healthcare data science.
Overview
The Data Challenge Program at IEEE ICHI 2026 focuses on advancing human-AI collaboration in healthcare data science. This program showcases results from the AgentDS Healthcare benchmark, where participants developed reproducible solutions for realistic healthcare prediction tasks using collaborative human-AI workflows.
The session brings together researchers and practitioners to:
- Present accepted papers from the Data Challenge track
- Share insights into human-AI collaboration in real-world healthcare settings
- Discuss design principles for reliable, reproducible, and high-impact healthcare data science systems
Theme
Human-AI Collaboration in Healthcare Data Science
Reproducible pipelines and transparent evaluation.
Practical insights into human-AI collaboration workflows.
Across the program, speakers will share lessons for designing next-generation AI-assisted data science systems.
Program
From Clinical Concepts to Multimodal Features: Human-AI Collaborative Feature Engineering for Healthcare Prediction
Hai Li, Rui Guo and Chenglu Li
Human-Guided Agentic AI for Multimodal Clinical Prediction: Lessons from the AgentDS Healthcare Benchmark
Lalitha Pranathi Pulavarthy, Raajitha Muthyala, Aravind Kuruvikkattil, Zhenan Yin, Rashmita Kudamala and Saptarshi Purkayastha
HMITL: Manager-Governed LLM Iteration with Guardrails and Rollback for Reproducible Healthcare Machine Learning Pipelines
Jialuo Chen, Haijing Wang and Siyu Shao
Designing and Auditing Healthcare Tabular Models for Human Oversight: An Empirical Study from the AgentDS-Benchmark Healthcare Domain
James Sargsyan
Healthcare Data Analysis in Low-Data Environments: A Multi-Task Benchmark Study
Aaron Zou
Human-AI Collaborative Modeling for Healthcare Prediction Tasks in AgentDS
Vedangi Deepak Deshpande and Prathmesh Santosh Choudhari
About AgentDS Healthcare
The program is based on the AgentDS Healthcare benchmark, which provides:
- Realistic, domain-specific healthcare datasets
- Multimodal and structured prediction tasks
- Standardized evaluation protocols
Program Chairs
Jie Ding, PhD
University of Minnesota, USA
An Luo
University of Minnesota, USA
Yu Huang, PhD
Indiana University, USA
Program Committee
Yunling Ge
University of Minnesota, USA
Zhenghan Li
University of Minnesota, USA