Dr Helen Sandoval

A lower FID is evidence only when the comparison survives scrutiny of the measurement protocol. Helen Sandoval makes that question the centre of her research and of SLOP8412.

Research and background#

Helen holds a continuing academic appointment in Slop University’s School of Continuous Improvement. Her background is in computational statistics; her work moved from estimating distributional distances to asking when those estimates support a model ranking.

She leads the Measurement Reliability Group, an internal research group that studies how sampling and implementation choices change the conclusions drawn from generative-model benchmarks. Her current interests include finite-sample bias and uncertainty in comparisons whose reported scores differ only slightly. The group’s question is practical: what evidence should accompany the number before another researcher relies on it?

Role in SLOP8412#

Helen convenes the course and leads the formal lectures. The Week 2 derivation establishes what the Gaussian distance measures; Week 5 asks what happens when its moments must be estimated. She returns to that distinction in the final report: a proposed replacement metric must face the same scrutiny as FID.

Supervision interests#

Potential research directions include uncertainty-aware benchmark reporting and experiments that separate estimator behaviour from model differences. A useful project question names a comparison, a suspected failure and an experiment that could contradict the hypothesis. These are supervision interests within the fictional course setting, not an advertised PhD opening.

For implementation questions and reproducible examples, see Tomasz Wierzba. Use the course forum for questions whose answers would help the whole class; keep individual matters private.