Researchers have developed a new framework for training order-agnostic autoregressive models on incomplete datasets. This approach reinterprets these models through the lens of missing data, showing that standard training implicitly handles imputation. The new method, called MO-ARM, directly trains on incomplete data and uses its density estimation capabilities for active information acquisition, outperforming existing imputation baselines on real-world benchmarks. AI
Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →
IMPACT Introduces a novel framework for handling missing data in generative models, potentially improving their applicability in real-world scenarios with incomplete observations.
RANK_REASON This is a research paper published on arXiv detailing a new modeling framework.