New research explores active learning for conditional generative compressed sensing

Researchers have developed a new framework for conditional generative compressed sensing, specifically for image recovery from subsampled Fourier measurements using prompt-conditioned generative models. This approach distinguishes between the prompt used for sampling distribution design and the prompt used for the recovery model. The study provides stable recovery bounds for ReLU and Lipschitz conditional generators, indicating that prompt-matched sampling maintains optimal complexity while prompt mismatch introduces a penalty. Experiments with Stable Diffusion demonstrate that prompts can effectively shape sampling distributions and impact image recovery. AI

影响 Introduces a novel method for image recovery that leverages prompt conditioning in generative models, potentially improving signal reconstruction from limited data.

排序理由 This is a research paper detailing a new framework for image recovery using generative models and prompt conditioning. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

• arXiv
• Lipschitz
• ReLU
• Stable Diffusion

• 论文
• 其他

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →