GFML: Gravity function for metric learning

Hoyoon Byun, Sungjun Lim, Kyungwoo Song

Abstract: Highlights•Gravity Function Metric Learning: an agile and model-agnostic metric learning method.•We validate that GFML is effective for recommender systems and image classification.•GFML is compatible with existing methods: matrix factorization and attention modules.•It also handles long-tailed items or data instances effectively.•We empirically show GFML helps learn robust, invariant features in vision task.