My research focuses on robust AI for sequential decision-making in environments under epistemic and aleatory uncertainty. I develop computational frameworks that integrate machine learning, uncertainty quantification, reinforcement learning, and large-scale distributed computing to close the reliability gap in autonomous systems—architecting agentic systems that maintain safety and performance guarantees in high-stakes, real-world deployment.
Sequential decision-making algorithms — including deep reinforcement learning — that learn optimal policies by interacting with complex, high-dimensional environments where actions carry long-horizon consequences.
Uncertainty quantification for decision-making in autonomy — supporting autonomous decisions under uncertainty by capturing and propagating uncertainty in both the input data and the model inference.
Calibration, hallucination detection, and evaluation that make large language models and the agents built on them dependable enough to act under real-world consequences.
Training models under formal privacy guarantees — differential privacy, secure aggregation, and federated learning on edge devices — so they learn useful patterns without exposing sensitive user data.
Forecasting methods for temporal and sequential data — modeling trends, seasonality, and regime shifts, with applications across financial markets and real-time system monitoring.
Machine learning applied to markets — building predictive signals, systematic trading strategies, and risk models that turn financial data into disciplined, data-driven decisions.
A selection below. Google Scholar
Probabilistic Engineering Mechanics (2022)
A Deep Interval Neural Network (DINN) that propagates input and parameter uncertainty through deep networks via interval analysis, producing guaranteed prediction enclosures with sharp bounds for safety-critical settings.
View PaperarXiv preprint (2021)
Introduces the Deep Interval Neural Network for quantifying input and parameter uncertainty in deep networks used in safety-critical applications, where reliable prediction bounds matter more than point estimates.
View Paper8th International Workshop on Reliable Engineering Computing (2018)
Addresses a major shortcoming to interval uncertainty approaches in computational mechanics by developing machine learning methods for interval field representation.
View Paper