Multifidelity optimization reduces hyperparameter evaluation cost by using cheaper objective approximations at different fidelity levels. It belongs to the three main automatic HPO algorithm families, alongside model-free and Bayesian optimization. section 5.2.2 section 5.5
The treatment stays at a systems-oriented overview rather than deriving algorithm mathematics. It provides enough background to select or set up an HPO system, while directing deeper mathematical study elsewhere. section 5.2.2
Use multifidelity optimization as part of an HPO workflow that:
- Balances objective-evaluation cost against the need to identify promising hyperparameter configurations. section 5.2.2
- Selects an algorithm according to task constraints rather than assuming one universally best HPO method. section 5.5
- Fits library-based experimentation for small teams or service-based execution for scalable production HPO. section 5.5
