arXiv · 2607.00407
Personalization as Inverse Planning: Learning Latent Design Intents for Agentic Slide Generation via Structural Denoising
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.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Tianci Liu, Zihan Dong, Linjun Zhang, Haoyu Wang, Jing Gao, Emre Kiciman, Ranveer Chandra, Wei-Ting Chen. 2026-08-12. Personalization as Inverse Planning: Learning Latent Design Intents for Agentic Slide Generation via Structural Denoising. https://arxiv.org/abs/2607.00407
Cite the original work for its findings. Save a collection to share your selection of sources.