New CRG award to develop scalable methods for partially observed stochastic optimal control

About

The KAUST Stochastic Numerics group is pleased to announce the award of a Competitive Research Grant (CRG) for the project: “Efficient Hierarchical Approximations for Partially Observed Stochastic Optimal Control with Applications.”

The project is funded at $900,000 for 36 months, starting April 1, 2026.

Team and partners

The project is led at KAUST by Prof. Raúl Tempone (PI) and Prof. Ibrahim Hoteit (co‑PI). It includes international collaboration supported through subawards to:

• EPFL (Switzerland) — Prof. Fabio Nobile
• WIAS Berlin (Germany) — Prof. Christian Bayer

Together, the KAUST and partner teams will connect new theory with scalable algorithms and validated application testbeds. 

Research focus

Many real‑world decision problems are partially observed: the true system state is not directly accessible, and decisions must be made using noisy and intermittent data. This project will develop efficient hierarchical approximations that make such problems computationally tractable—linking filtering (inference), optimal control, and goal‑oriented sensing within a unified numerical framework. 

Key methodological ingredients include hierarchical estimators and surrogates (e.g., MLMC/MIMC, sparse‑tensor ideas, projections/regularization), as well as modern stochastic optimization tools for nested objectives and gradients—supporting robust, risk‑aware decision making under uncertainty. 

Application drivers and outputs

The project will be validated through a reproducible algorithm‑and‑software stack and demonstrated on representative use cases, including:

• Ship / underwater‑vehicle routing under partial ocean–atmosphere observations
• Oil‑spill response, including risk‑aware boom placement and decision support
• Energy system operations under forecast uncertainty and partial state information