Fig. 6: The effects of provisioning budgets on computational funnels’ performance.
From: A multi-fidelity machine learning approach to high throughput materials screening
Normalised Regret vs. cost expenditure for ideally provisioned, under provisioned and over provisioned computational funnels alongside TVR-EI applied to the materials discovery challenges (a) Alexandria, (b) HOPV-15 and (c) Chen. The ideally provisioned funnel is defined as the funnel with the lowest possible budget that is able to achieve 0 median regret while the under-provisioned funnel is assigned half this budget and the over provisioned funnel is assigned twice this budget. In contrast to Fig. 4 these funnel results show the change in regret as the funnel uses its budget. Median regret values are plotted from 15 optimisations with different random seeds. Shading shows the interquartile range of the runs.
