Prof. Sara Shashaani
University of Oxford
Andrew Wiles Building
Radcliffe Observatory Quarter
Woodstock Road
Oxford
OX2 6GG
Near submission (for the full list of working papers see my CV):
- “Convergence and Complexity Analyses of Adaptive Sampling Stochastic Trust-Region in Random Subspaces — Part I: via Supermartingales”. Dzahini, Shashaani, Cartis.
- “Convergence and Complexity Analyses of Adaptive Sampling Stochastic Trust-Region in Random Subspaces — Part II: Pathwise Analysis and Dimension Dependence”. Shashaani, Dzahini, Cartis.
- "Adaptive Sampling and Pathwise Analysis for Stochastic Optimization". Shashaani.
Recently published or under review (for the full list see my CV):
- “Adaptive Regularization within Trust Region Methods for Stochastic Nonconvex Optimization,” Under review. Ha, Shashaani, Tran-Dinh. https://arxiv.org/abs/2604.15457
- “Extrapolation-based Direct Search for Nonsmooth Stochastic Zeroth-Order Optimization,” Under review. Palmieri, Rinaldi, Shashaani. https://arxiv.org/abs/2607.29408
- “Stratified Adaptive Sampling for Derivative-free Trust Region Optimization: Complexity and Applications,” Computers and Operations Research (major revision). Amici, Shashaani, Jain. https://arxiv.org/abs/2604.00178
- “Adaptive Sampling Trust Region Optimization for Derivative-free Stochastic Functions and Deterministic Equality Constraints,” Proceedings of the 2026 Winter Simulation Conference, IEEE (to appear). Felice, Roberts, Shashaani. https://arxiv.org/abs/2608.15894
- “Large Language Models Can Invent and Analyze Solvers Autonomously” Under review. Wang, Eckman, Shashaani, Henderson.
- “Complexity of Zeroth- and First-order Stochastic Trust-Region Algorithms,” SIAM Journal on Optimization 35(3), 2025. Ha, Shashaani, Pasupathy. https://doi.org/10.1137/24M1664484
Sara Shashaani (North Carolina State University, Industrial and Systems Engineering) is visiting the Mathematical Institute in Michaelmas 2026 as Oliver Smithies Visiting Lecturer at the Balliol College. She works on stochastic derivative-free optimization, especially adaptive-sampling algorithms where oracle calls are set by the algorithm’s own progress. Her current work with Coralia Cartis scales these methods within a trust-region optimization framework to high dimensions with random subspaces. It establishes almost-sure convergence and sample-complexity guarantees through supermartingale and pathwise arguments, gaining efficiency with methods that reuse function evaluations or variance reduction or adaptive regularization within an iteration to save the computational budget. She is also a co-developer of SimOpt, an open-source testbed for simulation-optimization solvers. Her other interest lies in optimization for stochastic systems, for example, the digital twins where a computational model is kept in step with a physical system through streaming data as conditions change. She will give a Smithies Lecture, ‘Research Aspects of Digital Twins’, at the Institute on 18 November, and a CMA seminar on 5 November.