NSF–CSIRO Collaborative Research Project

Fair and trustworthy AI for forecasting infectious disease spread.

We examine how biases in mobility, demographic, case-reporting, and pathogen-genomic data shape epidemic forecasts. We also develop learning methods and simulation environments to measure, mitigate, and correct their effects.

From incomplete observations to fairer forecasts Mobility, demographic, case, and genomic observations feed an AI forecasting model. Simulated ground truth and a fairness evaluation loop help identify and correct bias. MOBILITY movement and contact DEMOGRAPHICS population context CASES testing and reporting GENOMICS pathogen evolution THRUST 1 AI forecasting models THRUST 2 SIMULATED GROUND TRUTH THRUST 3 Measure. Correct. OBSERVABLE DATA FAIRER DECISIONS
We compare biased observations with controlled ground truth to understand where forecasting systems fail and how to improve them.

Funded by

U.S. National Science Foundation
CSIRO

Academic partners

UNSW Sydney
George Mason University
Emory University
Arizona State University
3connected research thrusts
23research outputs
4critical data streams

Australia ↔ United States

Why this matters

Better forecasts require a clearer view of bias.

Infectious-disease forecasting increasingly relies on data that are incomplete by design and in practice.

Access to testing varies. Mobility samples represent only part of the population. Census participation is uneven. Pathogen sequencing is geographically imbalanced. When these gaps propagate unnoticed through an AI pipeline, they can produce unreliable forecasts and inequitable decisions.

By comparing complete simulated ground truth with realistically biased observations, we can trace how biases enter a model, quantify their effects, and evaluate strategies for building fairer, more robust forecasting systems.

Research programme

Three connected thrusts,
from prediction to correction.

01

Spatiotemporal AI for disease-spread prediction

We develop transferable models that learn from heterogeneous and sparsely labelled signals, including human mobility, confirmed cases, pathogen genomics and phylogenetics, ecological context, and auxiliary open-source data.

  • Spatiotemporal learning
  • Heterogeneous graphs
  • Phylogenetics
  • Causal inference
02

Agent-based simulation for understanding bias

We build scalable synthetic populations and disease-spread simulations that provide complete ground truth alongside observable datasets collected under controlled bias.

  • Agent-based modelling
  • Synthetic mobility
  • Contact networks
  • Differential privacy
03

Measuring, mitigating, and correcting bias

We connect the simulated sandbox to predictive models to study how known biases affect performance across regions and populations and to evaluate correction strategies throughout the modelling pipeline.

  • Fairness evaluation
  • Bias correction
  • Robustness
  • Explainability
01Ground-truth worldComplete synthetic population
02Biased observationsSampling, reporting, privacy
03Forecasting modelsLearning across signals
04Measure and correctFairness, robustness, and trust

Featured work

Recent research across the programme.

All publications

Publications

Methods, systems, and evidence.

23 research outputs

2026 ACM SIGSPATIAL 2026 · Best Paper Runner-Up · Application Track accepted

Inferring Locations and Correcting Bias in Infectious Disease Data Using Human Mobility

Owen Tucker, Lily Wang, Ruochen Kong, Haley Stone, Matthew Scotch, Jing Du, Flora D. Salim, Taylor Anderson, Chandini Raina MacIntyre, Andreas Züfle

A human-mobility-informed approach to recovering missing locations and correcting spatial bias in infectious-disease surveillance data.

  • human mobility
  • location inference
  • bias correction
2026 The Web Conference 2026 · Web4Good Special Track accepted

Genomic-Informed Heterogeneous Graph Learning for Spatiotemporal Avian Influenza Outbreak Forecasting

Jing Du, Haley Stone, Yang Yang, Ashna Desai, Hao Xue, Andreas Züfle, Chandini Raina MacIntyre, Flora D. Salim

A heterogeneous graph learning framework that brings ecological context, mobility, case histories, and pathogen genomics together for H5N1 outbreak forecasting.

  • heterogeneous graphs
  • genomics
  • avian influenza
2025 KDD 2025 published

Simulated Infectious Diseases Datasets with Controlled Data Collection Bias

Ruochen Kong, Taylor Anderson, Matthew Scotch, David Heslop, Yonchanok Khaokaew, Hao Xue, Li Xiong, Chandini Raina MacIntyre, Flora D. Salim, Andreas Züfle

A reusable benchmark pairing complete simulated ground truth with observable datasets affected by controlled data-collection bias.

  • benchmark
  • agent-based simulation
  • data bias
2025 SIGSPATIAL 2025 published

A Probabilistic Framework for Imputing Genetic Distances in Spatiotemporal Pathogen Models

Haley Stone, Jing Du, Hao Xue, Matthew Scotch, David Heslop, Andreas Züfle, Chandini Raina MacIntyre, Flora D. Salim

A probabilistic approach to estimating missing genetic distances when pathogen sequencing is sparse or unevenly sampled across locations.

  • phylogenetics
  • imputation
  • pathogen models
2025 SIGSPATIAL 2025 published

STOAT: Spatial-Temporal Probabilistic Causal Inference Network

Yang Yang, Du Yin, Hao Xue, Flora D. Salim

A probabilistic causal network for learning spatiotemporal structure beyond purely correlational forecasting.

  • causal inference
  • spatiotemporal learning
  • forecasting
2026 Journal of Artificial Societies and Social Simulation · 29(1):7 published

All Models Are Wrong, but Can They Be Useful? Lessons from COVID-19 Agent-Based Models: A Systematic Review

Emma Von Hoene, Sara Von Hoene, Szandra Péter, Ethan Hopson, Emily Csizmadia, Faith Fenyk, Kai Barner, Timothy Leslie, Hamdi Kavak, Andreas Züfle, Amira Roess, Taylor Anderson

A systematic review identifying when COVID-19 agent-based models were useful and where their assumptions limited public health impact.

  • systematic review
  • agent-based models
  • COVID-19
2026 Submitted to SIGKDD 2026

An Infectious Disease Spread Simulation Based on Large Language Model Decision Making

Yonchanok Khaokaew, Ruochen Kong, Andreas Züfle, Hao Xue, Taylor Anderson, Chandini Raina MacIntyre, Matthew Scotch, Flora D. Salim, David Heslop

An extension of the simulation sandbox in which large language models drive individual agent decisions and behaviour.

  • LLM agents
  • behaviour
  • simulation
2026 SIGSPATIAL 2026 · Application Track accepted

CDGL-H3: H3-Grid Reported Case Intensity Forecasting for U.S. Avian Influenza Surveillance

Yang Lin, Jing Du, Haley Stone, Hao Xue, Flora D. Salim

An H3-grid-based case-intensity forecasting approach for spatiotemporal avian influenza surveillance across the United States.

  • avian influenza
  • H3 grids
  • case forecasting
2026 Preprint

From Flyways and Byways to Outbreak Prevalence

Haley Stone, Jing Du, Yang Yang, Ashna Desai, Rebecca Dawson, Hao Xue, David Heslop, Matthew Scotch, Andreas Züfle, Chandini Raina MacIntyre, Flora D. Salim

A heterogeneous graph transmission model for epidemiological reasoning with sparse ecological and spatiotemporal data.

  • heterogeneous graphs
  • ecology
  • transmission modelling
2025 NeurIPS 2025 published

Auto-Search and Refinement: An Automated Framework for Gender Bias Mitigation in Large Language Models

Yue Xu, Chengyan Fu, Li Xiong, Sibei Yang, Wenjie Wang

An automated search-and-refinement framework for identifying and mitigating gender bias in large language models.

  • gender bias
  • LLMs
  • mitigation
2025 SIGSPATIAL 2025 published

EpiScale: Large-Scale Simulation of Infectious Disease Based on Human Mobility

Ruochen Kong, Taylor Anderson, David Heslop, Matthew Scotch, Flora D. Salim, Chandini Raina MacIntyre, Andreas Züfle

A scalable engine for simulating infectious-disease dynamics across realistic urban populations using human mobility data.

  • large-scale simulation
  • human mobility
  • digital twins
2025 Preprint preprint

Evaluating the Bias in LLMs for Surveying Opinion and Decision Making in Healthcare

Yonchanok Khaokaew, Flora D. Salim, Andreas Züfle, Hao Xue, Taylor Anderson, Chandini Raina MacIntyre, Matthew Scotch, David Heslop

An evaluation of demographic and social biases that can arise when LLMs are used to survey public opinion or support healthcare decision-making.

  • LLM bias
  • healthcare
  • decision support
2025 Winter Simulation Conference 2025 published

Evaluating the Transferability of a Synthetic Population Generation Approach for Public Health Applications

Emma Von Hoene, Aanya Gupta, Hamdi Kavak, Amira Roess, Taylor Anderson

An assessment of whether synthetic population methods remain useful when transferred across public health settings.

  • synthetic populations
  • transferability
  • public health
2025 SIGSPATIAL 2025 published

Generating and Understanding Large Datasets of Social Contact Patterns from Foot-Traffic Data

Hossein Amiri, James Song, Li Xiong, Andreas Züfle

A method for constructing and analysing large-scale social contact networks from aggregate foot-traffic observations.

  • contact networks
  • foot traffic
  • mobility
2025 IEEE MDM 2025 published

Geo-Llama: Leveraging LLMs for Human Mobility Trajectory Generation with Constraints

Siyu Li, Toan Tran, Haowen Lin, John Krumm, Cyrus Shahabi, Lingyi Zhao, Khurram Shafique, Li Xiong

A constrained approach based on large language models for generating plausible synthetic human-mobility trajectories.

  • LLMs
  • synthetic mobility
  • trajectory generation
2025 SIGSPATIAL 2025 published

Spatial Inference on Phylogenetic Trees

Owen Tucker, Andreas Züfle

Methods for reasoning about spatial relationships directly on phylogenetic trees.

  • phylogenetics
  • spatial inference
2025 PLOS ONE · 20(6), e0325166 published

Towards Using Tweet Sentiment for Infectious Disease Detection

James Stassinos, Taylor Anderson, Andreas Züfle

An evaluation of social-media sentiment as a complementary open-data signal for earlier infectious-disease detection.

  • social media
  • sentiment
  • early detection
2025 SIGSPATIAL 2025 published

Training Machine Learning Models on Human Spatio-temporal Mobility Data: An Experimental Study

Yueyang Liu, Lance Kennedy, Ruochen Kong, Joon-Seok Kim, Andreas Züfle

A systematic study of how mobility data volume, granularity, and sampling regimes shape machine-learning performance.

  • human mobility
  • representation learning
  • evaluation
2024 SIGSPATIAL 2024 published

An Infectious Disease Spread Simulation to Control Data Bias

Ruochen Kong, Taylor Anderson, David Heslop, Andreas Züfle

An early implementation of an infectious-disease simulator designed specifically to manipulate and study data bias.

  • simulation
  • controlled bias
  • infectious disease
2024 Proceedings of the VLDB Endowment · Volume 17 published

HRNet: Differentially Private Hierarchical and Multi-Resolution Network for Human Mobility Data Synthesization

Shun Takagi, Li Xiong, Fumiyuki Kato, Yang Cao, Masatoshi Yoshikawa

A hierarchical, multi-resolution method for synthesising useful human mobility data under differential privacy.

  • differential privacy
  • human mobility
  • data synthesis
2024 ACM Transactions on Spatial Algorithms and Systems published

Leveraging Simulation Data to Understand Bias in Predictive Models of Infectious Disease Spread

Andreas Züfle, Flora D. Salim, Taylor Anderson, Matthew Scotch, Li Xiong, Kacper Sokol, Hao Xue, Ruochen Kong, David Heslop, Hye-Young Paik, Chandini Raina MacIntyre

A methodological framework for using controlled simulation data to trace and evaluate bias in infectious-disease prediction pipelines.

  • simulation
  • predictive bias
  • methodology
2024 SIGSPATIAL 2024 published

Urban Anomalies: A Simulated Human Mobility Dataset with Injected Anomalies

Hossein Amiri, Ruochen Kong, Andreas Züfle

A simulated mobility dataset with labelled, controllable anomalies for reproducible stress-testing and anomaly detection.

  • mobility data
  • anomaly detection
  • benchmark
2023 ECML PKDD 2023 published

Interpretation Attacks and Defenses on Predictive Models Using Electronic Health Records

Fereshteh Razmi, Jian Lou, Yuan Hong, Li Xiong

A study of attacks on model interpretations and corresponding defences for predictive systems built from electronic health records.

  • explainability
  • robustness
  • electronic health records

Open resources

Code, data, and reproducible research.

Repository links are maintained for individual publications and added once confirmed.

Dataset and benchmark

Controlled-bias infectious-disease datasets

Ground-truth and observable-data pairs for evaluating how collection bias propagates through predictive models.

Repository details forthcoming
Simulation

EpiScale

Large-scale, mobility-informed infrastructure for simulating infectious-disease spread across realistic urban populations.

Model family

Heterogeneous outbreak forecasting

Graph-learning methods that connect ecological, mobility, genomic and epidemiological signals.

Dataset

Urban Anomalies

Simulated human mobility with controlled, labelled anomalies for reproducible stress-testing.

Repository details forthcoming

Talks and events

From research to public health practice.

symposium3 February 2026 · 9:00 AM–5:00 PM AEDT

AI for Outbreak Forecasting: Are We Seeing the Full Picture?

This symposium connects advances in AI and large language models with practical methods for outbreak forecasting and simulation, public health translation, and government engagement.

People

An interdisciplinary Australia–U.S. collaboration.

Our team brings together computer scientists, simulation modellers, public health researchers, and epidemiologists working across the full forecasting pipeline.

Project leads

Andreas Züfle

U.S. Principal Investigator

Spatial and spatiotemporal data scienceEmory University

Flora D. Salim

Australian Principal Investigator

AI, time series and spatiotemporal forecastingUNSW Sydney

Investigators

C. Raina MacIntyre

Epidemiology and emerging infections

UNSW Sydney

David Heslop

Public health and integrated simulation

UNSW Sydney

Hao Xue

Deep learning, time series and mobility

UNSW Sydney

Li Xiong

Privacy-preserving spatiotemporal data

Emory University

Matthew Scotch

Genomic epidemiology and bioinformatics

Arizona State University

Taylor Anderson

Agent-based modelling and geosimulation

George Mason University

Key team members

Ashna Desai

Spatiotemporal epidemic forecasting and graph learning

UNSW Sydney

Haley Stone

Epidemic modelling, machine learning, and computational biology

UNSW Sydney

Jing Du

Graph representation learning, spatiotemporal modelling, and trustworthy AI

UNSW Sydney

Owen Tucker

Human mobility, spatial inference, and data privacy

Emory University

Ruochen Kong

Infectious-disease simulation, data bias, and machine learning

Emory University

Toan Tran

Machine learning, data privacy, and human-mobility modelling

Emory University

Yang Yang

Spatiotemporal graph learning and epidemic forecasting

UNSW Sydney

Yonchanok Khaokaew

Large language models, human-centred computing, and wearable computing

UNSW Sydney

Contributors

Aanya Gupta · Amira Roess · Chengyan Fu · Cyrus Shahabi · Du Yin · Emily Csizmadia · Emma Von Hoene · Ethan Hopson · Faith Fenyk · Fereshteh Razmi · Fumiyuki Kato · Hamdi Kavak · Haowen Lin · Hossein Amiri · James Song · James Stassinos · Jian Lou · John Krumm · Joon-Seok Kim · Kai Barner · Khurram Shafique · Lance Kennedy · Lily Wang · Lingyi Zhao · Masatoshi Yoshikawa · Rebecca Dawson · Sara Von Hoene · Shun Takagi · Sibei Yang · Siyu Li · Szandra Péter · Timothy Leslie · Wenjie Wang · Yang Cao · Yang Lin · Yuan Hong · Yue Xu · Yueyang Liu