NAISS
SUPR
NAISS Projects
SUPR
Stochastic Control and Optimization for Robust Generative Modeling under Uncertainty
Dnr:

NAISS 2026/4-780

Type:

NAISS Small

Principal Investigator:

Siyi Wang

Affiliation:

Kungliga Tekniska högskolan

Start Date:

2026-04-23

End Date:

2027-05-01

Primary Classification:

10210: Artificial Intelligence

Allocation

Abstract

This project aims to revolutionize Generative Optimization by framing the generation process as a stochastic optimal control problem. Unlike standard generative models that often ignore environmental or structural uncertainties, our approach integrates risk-averse measures and stochastic control theory to ensure robustness. We will leverage distributed optimization techniques to scale these complex control-theoretic formulations across multi-GPU nodes. The core objective is to develop a framework where the "generation" is an optimally controlled trajectory that remains stable and efficient even under significant system uncertainty.