Latent Posterior Factors (LPF) bridge neural representations with structured probabilistic reasoning by converting VAE posteriors into factors for Sum-Product Networks.
arXiv · March 18, 2026 · 2603.15670
The Takeaway
This allows for tractable, exact probabilistic reasoning over unstructured data (like images or text) with calibrated uncertainty estimates, significantly outperforming deep learning and LLM baselines in evidential reasoning.
From the abstract
Real-world decision-making, from tax compliance assessment to medical diagnosis, requires aggregating multiple noisy and potentially contradictory evidence sources. Existing approaches either lack explicit uncertainty quantification (neural aggregation methods) or rely on manually engineered discrete predicates (probabilistic logic frameworks), limiting scalability to unstructured data.We introduce Latent Posterior Factors (LPF), a framework that transforms Variational Autoencoder (VAE) latent p