Raghuram Shankar
← Bayesian Optimization

Part 4 of 8

The Maths II: Acquisition Functions

The posterior hands us μn(x) and σn(x); an acquisition function turns those two numbers into a decision. This section derives the classic by hand, tours the zoo, and covers the two upgrades a modern practitioner actually ships: LogEI and batch acquisition. Throughout, f*n = maxi yi is the best value observed so far and z = (μn(x) − f*n) / σn(x).

  1. Improvement, and its expectation
  2. The acquisition zoo
  3. Numerics and batches: LogEI and qEI
  4. The knowledge gradient: valuing what you will believe
  5. Paying for the knowledge gradient
  6. Quiz: 5 questions