Black-Box Optimization From Small Offline Datasets via Meta Learning with Synthetic Tasks

In three linesOptBias, a meta-learning framework, addresses offline black-box optimization with scarce data. It generates synthetic tasks via Gaussian process to learn reusable optimization bias, then fine-tunes the surrogate model on small target datasets. Outperforms baselines on continuous and discrete optimization benchmarks.

Summary generated by Claude — human-verified