Measure, learn, choose again
NIST describes sequential Bayesian experiment design as a loop: measurements update a probability distribution over unknown model parameters, then that updated knowledge helps select the next setting. The goal is to reduce uncertainty without spending measurements on settings expected to teach little. [1]
- Update: what do the results imply?
- Choose: which setting teaches more?
- Measure: return with new evidence
Two knobs for a spin
The documentation’s simulated spin-control example adjusts both pulse duration and frequency. A pulse must have suitable settings to manipulate a spin accurately. The demonstration tracks how parameter estimates change as simulated measurements arrive; it is not a report of a new laboratory device. [2]
The model still matters
The software requires a model connecting adjustable settings and unknown parameters to predicted observations. Its documentation puts responsibility for a reliable model and reasonable settings on the user. Choosing informative measurements cannot make an unsuitable model trustworthy. [3]
Go a little deeper
Optional reading · about 1 more minute
Not the same as finding the highest value
The documentation distinguishes parameter estimation from Bayesian optimization, which seeks a maximum or minimum of an unknown function. It also says the package primarily guides new measurements rather than analyzes a finished dataset. [3]
Read the example at its actual scale
NIST’s project page was created in 2019, updated in March 2025 and is marked completed. Its quantum-control example is a simulation. This is an explanation of an established measurement strategy, without a claim of a new September 2026 result or a universal speedup. [1]
Original sources
Attributed synthesis, not original reporting. Examples labeled hypothetical or illustrative are explanatory. Reviewing a source does not independently validate its findings.
- NIST: sequential Bayesian experiment design ↗
2019 project page updated March 26, 2025, completed. Description and simulated quantum-control example read September 29; no performance ratio adopted.
- NIST OptBayesExpt: demonstration examples ↗
Version 1.2.0 documentation; Tuning a pi pulse and Figure 5 read September 29. Explicitly simulated data.
- NIST OptBayesExpt: overview ↗
Documentation dated December 22, 2022. What is it for and What does it do read September 29; model responsibility and optimization distinction.