PABBO: Preferential Amortized Black-Box Optimization

Xinyu Zhang *, Daolang Huang *, Samuel Kaski, Julien Martinelli · 2025

Read PABBO: Preferential Amortized Black-Box Optimization by Xinyu Zhang *, Daolang Huang *, Samuel Kaski, Julien Martinelli (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-00924.txt.

Description

Xinyu Zhang and co-authors make preference-based optimization fast enough to chat with. Their 2025 paper works on preferential Bayesian optimization, which learns what a user likes by asking them to compare pairs of designs. The classic version refits a heavy statistical model after every answer, far too slow for interactive use with a person waiting.

Their method, PABBO, amortizes the whole loop: a transformer neural process is meta-trained to act as both the surrogate model and the chooser of the next pair. Reinforcement learning plus tailored extra losses shape that training. On synthetic and real-world benchmarks the system ran orders of magnitude faster than the Gaussian process standard. It also often won on accuracy, not just speed. The authors aim the work at design tools where taste, not a formula, is the objective.

Catalog record

CREATORXinyu Zhang *, Daolang Huang *, Samuel Kaski, Julien Martinelli
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-00924

Browse the full library · About the Tower of Babel