Tag
The paper introduces a multi-stage architecture for learning individuated utility functions from multi-modal data, showing that models accounting for heterogeneous preferences outperform universal utility models in subjective tasks like aesthetic judgments.