Marformer: A Transformer for Predicting Missing Data Distributions
What it is
Marformer is a Transformer model designed to predict the conditional marginal distributions of missing variables given observed values. It uses an attention mechanism to iteratively refine hidden vector representations for each variable, allowing it to make predictions in a single forward pass without modeling the full joint distribution.
The problem it solves
Real-world decisions often require information about variables that are not observed, and existing methods for predicting these missing values are either slow, require specific domain knowledge of the data-generating process, or model the full joint distribution unnecessarily.
What you could use it for
- Computing Bayes risk under incomplete information
- Estimating Value of Information (VOI) for decision-making
- Predicting missing data in Bayesian networks and Gaussian distributions
- Handling structured annotation data with missing entries
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.