<ul data-eligibleForWebStory="true"><li>Researchers introduce Spatial Reasoning Models (SRMs) for reasoning over sets of continuous variables using denoising generative models.</li><li>SRMs infer continuous representations on unobserved variables based on observations on observed variables.</li><li>Current generative models like diffusion and flow matching models can lead to hallucinations in complex distributions.</li><li>The study includes benchmark tasks to evaluate the quality of reasoning in generative models and quantify hallucination.</li><li>SRMs highlight the importance of sequentialization in generation, the associated order, and sampling strategies during training.</li><li>The framework shows that the order of generation can be predicted by the denoising network itself, leading to significant accuracy improvements.</li><li>The project website offers additional resources including videos, code, and benchmark datasets.</li><li>The SRM framework enhances accuracy in specific reasoning tasks from less than 1% to over 50%.</li></ul>

Spatial Reasoning with Denoising Models

Discover more