Personalization as Inverse Planning: Learning Latent Design Intents for Agentic Slide Generation via Structural Denoising
2026-07-01 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors address how to customize slide designs not just for the whole presentation but for each individual page, which current AI methods find difficult. They treat this as a puzzle where you try to figure out the hidden design goals from the final slides without needing to know the slide-making software used. To solve this, they created a system called SPIRE that intentionally messes up slide designs and trains two AI agents to fix them, learning good design habits through trial and error. Their experiments show that this approach works better than previous methods for personalizing slides at the page level.
Slide designPage-level personalizationInverse planningReinforcement learningMulti-agent systemsPolicy gradientStructural denoisingExecutable designsDesign intentAI agents
Authors
Tianci Liu, Zihan Dong, Linjun Zhang, Haoyu Wang, jing Gao, Emre Kiciman, Ranveer Chandra, Wei-Ting Chen
Abstract
Slide design requires personalizing both deck themes and page layouts. Yet, current AI agent-based methods struggle with fine-grained, page-level design. Solely relying on prespecified templates or user verbose instructions, they fail to capture latent design intents, leaving Page-level Slide Personalization (PSP) unresolved. To close this gap, this work formulates PSP as an inverse planning problem. We propose to learn a design intent without assuming any knowledge of the specific executing tools (e.g., PowerPoint, Beamer) being used. However, relinquishing control over these tools makes the problem intractable to optimize end-to-end. To overcome this, we propose SPIRE, a principled framework to solve PSP approximately. By intentionally corrupting the visual structures of clean slides, SPIRE creates a verifiable task to denoise the corruption, whereby two agents learn to collaboratively refine executable designs via reinforcement learning (RL). We present a proof that structural denoising is a consistent surrogate for PSP, and that the multi-agent formulation strictly reduces policy gradient variance in RL. Extensive experiments demonstrate the superiority of SPIRE.