AI Paper Atlas

Research map / Probabilistic and scientific modeling / Bayesian optimization

Generative Pretraining for Black-Box Optimization

Satvik Mehul Mashkaria, Siddarth Krishnamoorthy, Aditya Grover

ICML 2023 · Probabilistic and scientific modeling › Bayesian optimization · approach: Offline black-box optimization with surrogates · 2 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use BONET instead of the strongest listed baselineoffline black-box optimization · Design-Bench 100th percentile mean rankmean rank: 3.7 -> 2.4Improved
Condition the generated suffix on a low evaluation regret budget (ablation)BONET on Branin · Braninbest function valueImproved
Provide a longer prefix of offline history (ablation)BONET on Branin · Braninbest function valueImproved
Use a prefix history rather than fully independent rollouts (ablation)BONET · Braninrollout stabilityImproved
Keep the evaluation regret budget fixed rather than updating it during the suffix (ablation)BONET · Braninbest function valueNo clear effect
Use reweighting when sampling offline trajectory points (ablation)BONET trajectory construction · D'Kittybest function valueImproved
Sort sampled points by function value when constructing trajectories (ablation)BONET trajectory construction · D'Kittybest function valueMixed
Use one bin instead of multiple bins for reweighting (ablation)BONET trajectory construction · Branin and Design-Bench tasksbest function valueHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy