Desi R. Ivanova

Desi R. Ivanova is a research fellow at the University of Oxford's Department of Statistics under the Florence Nightingale Bicentennial Fellowship. Her work bridges probabilistic machine learning, Bayesian experimental design, and LLM evaluation frameworks.

She holds a DPhil in Statistics from Oxford's StatML CDT program (2020-2024) and an MMORSE in Mathematics from University of Warwick (2011-2016) with Erasmus exchange at LMU Munich.

• Research spans causal machine learning and uncertainty quantification
• Developed CO-BED and Step-DAD frameworks
• Focus on LLM evaluation methodology and calibration
• Expert in real-time adaptive experimental systems

Her publications demonstrate expertise in Bayesian self-consistency methods, neural data compression, and privacy-preserving dataset merging. Key contributions include improving amortized inference efficiency and developing gradient-based causal experimental designs.

Current work emphasizes rigorous statistical evaluation of language models, advocating for appropriate uncertainty quantification when analyzing performance across small datasets. She critiques CLT-based methods for LLM evaluation and proposes more robust frequentist and Bayesian alternatives.