PMLE Scaling Prototypes into ML Models Practice Question

An ML team wants to use Vertex AI Hyperparameter Tuning to tune a custom training job. They have a budget of 50 trials and want to use an algorithm that balances exploration and exploitation. Which algorithm should they choose?

⚠ Common exam trap

A common pitfall is assuming that random search is the best default for balancing exploration and exploitation. However, random search lacks any exploitation mechanism, making Bayesian optimization the correct choice for efficient tuning within a constrained budget, as emphasized in Google PMLE.

Answer choices

Why each option matters

Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.

Correct answer & explanation

✓

Bayesian optimization (Vizier default)

Bayesian optimization (the default algorithm in Vertex AI Vizier) is the correct choice because it explicitly balances exploration and exploitation by building a probabilistic model of the objective function and using an acquisition function to select the next hyperparameter configuration. With a budget of 50 trials, this algorithm efficiently converges to optimal regions while still exploring uncertain areas, making it ideal for tuning custom training jobs where each trial is computationally expensive.

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

✗
Random search
Why it's wrong here
Random search samples configurations independently, so each trial ignores prior results and cannot exploit promising regions; it balances nothing adaptively. It is tempting because it handles high-dimensional spaces and parallel trials well, but Vertex AI's Bayesian optimisation is the algorithm that actually balances exploration against exploitation.
✗
Grid search
Why it's wrong here
Grid search exhaustively evaluates a fixed lattice of predefined values, so it explores uniformly and never exploits promising regions; with 50 trials it wastes budget on unpromising combinations. It is tempting for small, low-dimensional spaces where exhaustive coverage is feasible, but it cannot adaptively balance exploration and exploitation.
✓
Bayesian optimization (Vizier default)
Why this is correct
Bayesian optimization (Vizier's default) models the objective probabilistically, using prior trial results to pick promising configurations while still sampling uncertain regions — directly balancing exploration and exploitation. It converges within the 50-trial budget far more efficiently than grid or random search, satisfying the stated constraint.
✗
Manual search
Why it's wrong here
Manual search requires a human to set each configuration by hand, so it cannot autonomously balance exploration against exploitation across 50 trials. It is tempting because manual tuning suits small experiments where intuition guides a handful of runs, but Vertex AI's Bayesian optimisation is what delivers that balance automatically.

Go deeper

PMLE Scaling Prototypes into ML Models Practice Question

Why each option matters

Option-by-option breakdown

Related to this question