arXiv：cs.LG· Xuanchen Wang, Heng Wang, Weidong Cai·· 3 小时前AI 评分38
LatticeSMC：分块序列生成器该在哪里投入推理时计算
LatticeSMC: Where to Spend Inference-Time Compute in Chunked Sequence Generators
AI 导读
LatticeSMC 是一种基于 Feynman-Kac 模型、在 chunk 索引与去噪步二维格点上采样的推理时引导方法，用于音乐、动作和视频等分块长序列生成。
正文
Abstract:Long-form generators for music, motion and video produce sequences chunk by chunk, with each chunk generated by iterative denoising while rewards are defined over the full sequence. Existing inference-time steering methods typically act on one axis at a time: best-of-N at the end, Feynman-Kac steering across denoising steps, or streaming pruning across chunks, and are often compared under unmatched compute or different return rules. We introduce budget-matched chunked steering and propose LatticeSMC, a sampler derived from a Feynman-Kac model on the two-dimensional lattice of chunk index and denoising step. Two telescoping results make its design exact: for chunk-additive rewards, the two axes induce identical weights, so resampling should occur where lookahead is cheapest; for terminal rewards, any prefix score defines an exact intermediate potential, making prefix-evaluable rewards twists with no estimation or extra denoiser calls. LatticeSMC resamples on these potentials at chunk boundaries and, when scoring is free, within chunks, returning either a weighted draw or the best particle. Under matched compute, on music-to-dance diffusion and 40-second text-to-music generation, it raises beat alignment from 0.234 to 0.441 (best-of-N: 0.354) and prompt adherence from 0.470 to 0.560 at 32 particles, while preserving held-out quality. It also retains its advantage on long-range rewards and is preferred by human raters in 60-77 percent of pairwise comparisons. Finally, we show that commitment strength should follow the information in the current potential, while the value of lookahead is predicted by the within-set predictability of future reward.
| Comments: | 29 pages, 7 figures |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.02774 [cs.LG] |
| (or arXiv:2610.02774v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.02774 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Xuanchen Wang [view email]
[v1]
Fri, 2 Oct 2026 03:59:53 UTC (519 KB)
来源：arXiv：cs.LG · arxiv.org