Matrix Completion

We explain how visual attribute learning may benefit from multi-task/transfer learning approaches that carefully select what to share among whom. We show that such methods can combat noisy correlations among attributes, and variations among attribute appearances across object categories.