DeltaBO achieves a regret bound of order O(sqrt(T (T/N + gamma_delta))) for Bayesian optimization with knowledge transfer by modeling uncertainty on the difference function delta between source and target tasks.
For Bohachevsky functions, lengthscale of source, target, and difference kernel is 1.6, 0.8, and 1.0 withσ2 0 = 0.24, σ2 = 0.06and τ 2 = 0.32
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Provable Accelerated Bayesian Optimization with Knowledge Transfer
DeltaBO achieves a regret bound of order O(sqrt(T (T/N + gamma_delta))) for Bayesian optimization with knowledge transfer by modeling uncertainty on the difference function delta between source and target tasks.