LISA: Likelihood Score Alignment for Visual-condition Controllable Generation

  • 类型：arxiv
  • 标识：2606.27192
  • 链接：https://arxiv.org/abs/2606.27192
  • 主分类：engineering
  • 形态：method
  • 被引：2
  • 被引来源：Semantic Scholar
  • S2被引：2
  • OpenAlex被引：0
  • 影响力被引：0
  • TLDR：Experiments demonstrated that LISA can not only consistently accelerate the training convergence and improve final synthetic results, but also encourage the side network's features to be more disentangled for conditional modeling with negligible additional training cost and zero extra inference cost.
  • OpenAlex ID：W7166139718
  • OpenAlex DOI：10.48550/arxiv.2606.27192
  • DOI：10.48550/arxiv.2606.27192
  • DOI来源：OpenAlex
  • 开放获取：green
  • 开放获取链接：https://doi.org/10.48550/arxiv.2606.27192
  • OpenAlex更新：2026-07-19
  • 副分类：risk
  • 待LLM分类：否
  • 标题中文：LISA：面向视觉条件可控生成的似然分数对齐
  • TLDR中文：实验表明，LISA 不仅能持续加速训练收敛并提升最终合成结果，还能促使侧网络特征在条件建模中更加解耦，且几乎无额外训练成本，推理成本为零。
  • 来源文件：
  • /inbox/tom/_candidates/2026-06-29-agent-memory-tool-use-candidates.json
  • /inbox/tom/_candidates/2026-06-28-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]