AAISM Domain 2 Practice Test

The idea being tested is how NAS assesses many candidate architectures quickly and cost-effectively. That role belongs to the performance estimation strategy. It’s about training and validating each candidate in a lightweight way to get a comparable measure of performance without fully training every model to convergence. By using proxies, partial training, weight sharing, early stopping, or learned surrogates, you can estimate how good an architecture will be while controlling compute and time, which is essential when you’re comparing many options. The other components serve different purposes. The search strategy determines how you explore the space of architectures to propose candidates. The training procedure defines how a given architecture is trained. The model selection step happens after evaluation, deciding which architecture to deploy based on the estimated performance.