SURFACE: Preference Distribution Matching on Pareto Fronts via Scalarization Steering
arXiv 于 2611.17484v1 公布新论文 SURFACE,研究如何用标量化引导(scalarization steering)在帕累托前沿上匹配偏好分布。论文摘要指出,构建有限的帕累托表示需要把数量有限的解分配到整条帕累托前沿(PF)上。目前公开的是新投稿(Announce Type: new)的摘要,完整方法与实验细节尚未在材料中披露。
How this story unfolded
Key points5 items Model summary · ordered by when the evidence was published
- Point 01 11/16, 13:00
论文 SURFACE 以 arXiv:2611.17484v1 发布,属新投稿。
Based on 1 - Point 02 11/16, 13:00
论文聚焦在帕累托前沿上做偏好分布匹配,方法为标量化引导。
Based on 1 - Point 03 11/16, 13:00
摘要提出,构建有限帕累托表示需在整条帕累托前沿上分配有限数量的解。
Based on 1 - Point 04 11/18, 13:00
同期另一篇 arXiv:2611.21669v1 研究掩码扩散生成中的 token 承诺控制面。
Based on 1 - Point 05 11/19, 13:00
arXiv:2605.09439v2 为替换更新,主题是条件分布匹配的逆向设计。
Based on 1
ReportsBy publish time · newer further down
- November 16, 2026 · Mon
- Start
SURFACE: Preference Distribution Matching on Pareto Fronts via Scalarization Steering
Original summary arXiv:2611.17484v1 Announce Type: new Abstract: Constructing a finite Pareto representation requires allocating a limited number of solutions across the Pareto front (PF). A desire
arXiv cs.LG Original ↗ - November 18, 2026 · Wed
- +2d 0h
Token Commitment as a Control Surface for Steering Masked Diffusion Generation
Original summary arXiv:2611.21669v1 Announce Type: new Abstract: Masked diffusion models repeatedly predict provisional tokens before a sampler decides which predictions become permanent. Existing
arXiv cs.CL Original ↗ - November 19, 2026 · Thu
- +3d 0h
Inverse Design for Conditional Distribution Matching
Original summary arXiv:2605.09439v2 Announce Type: replace-cross Abstract: Generative models are powerful tools for sampling from a learned distribution $\mathcal{P}(Y \mid X)$, and inverse-design
arXiv stat.ML Original ↗ - Still developing · later reports will keep extending this axis