Ladder Loss
Ladder loss is proposed as a loss function for learning to rank tasks where certain orders or relations are retained while computing an embedding or ranking.
Proposed as an extension of triplet loss in information retrieval, ladder loss simultaneously maintain a series of distance orders by optimizing a continuous relevance degree with a chain of distance inequalities. It is claimed that such extension ensures additional proximity orders among "irrelevant" candidates and leads to performance enhancements of visual-semantic embedding in learning to rank tasks.
