Return on investment (ROI) measures profit or benefit per unit of investment cost. In multifidelity optimization, it divides the information gained about an objective's optimum by the cost of querying a data point. section 9.3.2
Use the ROI acquisition score to balance information and expense:
- Prefer the query with greater information gain when candidate queries cost the same. section 9.3.2
- Prefer the cheaper query when candidate queries provide the same information. section 9.3.2
- Use informative low-fidelity queries while they remain valuable, then switch to high-fidelity queries when necessary. section 9.3.2
Choosing a fidelity
A multifidelity MES policy may prefer low-fidelity points when their cost is low and their information gain is useful. It may prefer high-fidelity points when low-fidelity observations are insufficient or nearly as expensive.
