Causal AI Lab
Dr. Md Osman Gani, Department of Information Systems.
Our lab develops knowledge-guided and temporal causal discovery methods that integrate domain knowledge with observational data to identify causal structure and estimate treatment effects, with primary applications in healthcare and critical care decision-making. Our methods extend classical causal discovery (constraint-based, score-based, and optimization-based approaches) to incorporate structured domain knowledge as constraints, and to handle autocorrelated, non-stationary time-series data common in clinical settings. Current projects include causal discovery for ICU treatment-timing decisions (e.g., antibiotic and oxygen-therapy timing in sepsis) and benchmarking against deep-learning and reinforcement-learning baselines for individualized treatment-effect estimation.
In parallel, our lab conducts research in pervasive and ubiquitous computing, including image- and sensor-based systems for accessibility, indoor localization, and human activity recognition. Current work includes developing computer-vision and sensor-fusion models to characterize road-surface conditions relevant to wheelchair accessibility, toward a crowd-sensed accessibility map for personalized navigation, alongside continued work on activity recognition and localization using wireless and sensor data.
This research requires substantial computational resources: GPU acceleration for training deep-learning treatment-effect baselines, computer-vision and sensor models, and large-scale synthetic benchmarking experiments (varying graph size, edge density, and noise conditions) to validate causal discovery algorithms against ground-truth structure; high-memory CPU nodes for large-scale graph search and cross-cohort analysis across multiple clinical datasets; and reliable storage for de-identified clinical data and sensor/image data processing pipelines.