Personalization as Inverse Planning: Learning Latent Design Intents for Agentic Slide Generation via Structural Denoising
中文标题: 逆向规划实现个性化：通过结构去噪学习潜在设计意图的智能幻灯片生成
英文摘要
This paper addresses Page-level Slide Personalization (PSP) by formulating it as an inverse planning problem to infer latent design intents without assuming any specific presentation tool. The proposed SPIRE framework creates a verifiable task by corrupting visual structures of clean slides and training two agents to denoise them via reinforcement learning. The authors prove that structural denoising is a consistent surrogate for PSP and that the multi-agent formulation reduces policy gradient variance. Experiments show SPIRE outperforms existing template- or instruction-based methods.
中文摘要
本文将页面级幻灯片个性化定义为一个逆向规划问题，旨在不依赖任何特定演示工具的情况下推断潜在设计意图。提出的SPIRE框架通过破坏干净幻灯片的视觉结构，创建一个可验证的结构去噪任务，并利用强化学习训练两个智能体协同优化可执行设计。作者证明了结构去噪是PSP的一致代理任务，且多智能体公式严格降低了策略梯度方差。实验表明SPIRE优于现有基于模板或指令的方法。
关键要点
Page-level Slide Personalization (PSP) is framed as an inverse planning problem to capture latent design intents without tool-specific knowledge.
将页面级幻灯片个性化定义为逆向规划问题，无需特定工具知识即可捕捉潜在设计意图。
SPIRE uses structural denoising as a consistent surrogate, with two RL agents collaboratively refining designs and reducing policy gradient variance.
SPIRE以结构去噪作为一致代理，两个强化学习智能体协同优化设计，并降低策略梯度方差。
Experimental results demonstrate SPIRE's superiority over existing methods that rely on templates or verbose instructions.
实验证明SPIRE优于依赖模板或冗长指令的现有方法。