A probabilistic model that learns to make predictions by conditioning on observed examples, useful for few-shot learning and uncertainty estimation.