{
  "id": 4023,
  "url": "https://arxiv.org/abs/2605.19804v1",
  "title": "Stitched Value Model for Diffusion Alignment",
  "summary": "For practical use, diffusion- or flow-based generative models must be aligned with task-specific rewards, such as prompt fidelity or aesthetic preference. That alignment is challenging because the reward is defined for clean output images, but the alignment procedure requires value function estimates at noisy intermediate latents. Existing methods resort to Tweedie-style or Monte Carlo approximations, trading off estimator bias against computational cost: Tweedie estimates are efficient but bias",
  "authors": "Hyojun Go, Hyungjin Chung, Prune Truong, Goutam Bhat, Li Mi, Zhaochong An et al.",
  "category": "research",
  "topics": "bias-fairness,safety-alignment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-19T13:02:50.000Z",
  "fetched_at": "2026-07-14T16:30:41.582Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/4023",
  "original_url": "https://arxiv.org/abs/2605.19804v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}