Fenchel Tilting: Weighted Correction for Efficient Finetuning of Generative Models

Excerpt

arXiv:2609.40030v1 Announce Type: new Abstract: Adapting a pretrained generative model to an arbitrary preference expressed as a utility function underlies reward alignment, guided design, and constraint satisfaction, enabling diverse applications. Existing fine-tuning methods trade off generality against computational cost: they either restrict the family class of supported preferences to keep optimization simple or preserve generality at the expense of efficiency. We introduce Fenchel Tilt Flo