Research
My research asks how learning systems can move from pattern recognition to mechanism-aware reasoning. I work across the causal learning stack and translate those methods into biomedical systems.
FOUNDATIONS
Deep causal learning
I study how deep representation learning can address high-dimensional variables, combinatorial structure search, hidden confounding, and heterogeneous treatment effects. This work produced a comprehensive framework connecting causal representation learning, causal discovery, and causal inference.
Deep Causal Learning: Representation, Discovery and Inference · ACM Computing Surveys, 2025. 2025 Impact Factor: 30.4 (ranked 1/146 in Computer Science Theory & Methods).
METHODS
Identifiable and adaptive causal discovery
Supervised causal learning is powerful but can inherit structural bias and fail under distribution shift. My recent work tackles both problems: SiCL predicts identifiable skeletons and v-structures, while TTT-SCL generates instance-aligned training data at test time for robust graph recovery.
TRANSLATION
Temporal causal precision medication
I develop causal models that separate underlying physiological state from treatment effects in longitudinal clinical records. The TCPM framework learns treatment-free physiological profiles, estimates counterfactual responses, and searches for personalized medication strategies across time.
The framework was evaluated across six cohorts spanning public intensive-care data and clinical datasets from partner hospitals, linking causal methodology to actionable biomedical decisions.
A generalist precision medication framework using temporal causal inference based on treatment-free physiological profiles · Nature Communications, 2026
BIOMEDICAL REPRESENTATION
Representation learning for neural dynamics
I study stable, biologically meaningful representations of neuronal systems. NeurPIR uses contrastive learning across activity segments to recover intrinsic properties that generalize across experimental conditions and unseen animals.
Causal learning toolkit
My research practice spans causal representation learning, graphical causal models, causal discovery, counterfactual estimation, heterogeneous treatment effects, temporal causal inference, and deep learning for biomedical systems. I work primarily in Python with PyTorch and commonly use YLearn, DoWhy, CausalML, EconML, and TrustworthyAI.