Course contentsShow
← Machine Learning and Deep Learning
Lesson 511 of 3,53812. Model Validation and TuningFree lesson

Acquisition Functions in Bayesian Optimization

Balance exploration and exploitation using expected improvement and upper confidence bounds.

Acquisition Functions in Bayesian Optimization

What you'll learn: How acquisition functions help Bayesian optimization intelligently choose which hyperparameters to try next by balancing exploration of unknown regions with exploitation of promising areas.

The Core Challenge

In Bayesian optimization (which you learned in lesson 510), we build a probabilistic model of how hyperparameters affect performance. But how do we decide which hyperparameter combination to test next? That's where acquisition functions come in—they're the decision-making recipe that guides the search.

Balancing Exploration and Exploitation

Think of searching for the best restaurant in a new city:

  • Exploitation = returning to the great place you already found
  • Exploration = trying unknown restaurants that might be even better

Acquisition functions mathematically balance this trade-off:

Expected Improvement (EI)

Asks: "How much better than our current best could this point be?" It calculates the expected amount of improvement, considering both the predicted value and the uncertainty. High uncertainty in unexplored regions gets rewarded—encouraging exploration when there's potential for big gains.

Upper Confidence Bound (UCB)

Uses the formula: predicted value + (β × uncertainty)

The β parameter controls the balance. Higher β means more exploration (trust the uncertainty). Lower β means more exploitation (trust the prediction). UCB is like being optimistic about uncertain options—it assumes they might be better than predicted.

Why This Matters

These functions automatically adapt the search strategy: early on, they explore broadly; later, they exploit the best regions found. This makes Bayesian optimization far more sample-efficient than grid or random search for expensive evaluations.

Key Takeaway: Acquisition functions like Expected Improvement and Upper Confidence Bound guide Bayesian optimization by mathematically balancing the need to explore uncertain regions against exploiting known promising areas, making hyperparameter search intelligent and efficient.