Total Variation Similarity
A similarity measure derived from the total variation distance between probability distributions, defined as one minus the total variation distance. It measures the mass shared by two probability distributions and is suited to comparing distributions on the simplex.
Timeline (last 90 days)
The similarity measure is changed from cosine similarity to Total Variation similarity to suit probability distribution comparison on the simplex.
The computational complexity of Total Variation similarity is of the same order as that of cosine similarity.
Whether the simplex constraint limits the expressive freedom of embeddings, and the training stability under Total Variation similarity, still require more extensive experiments.
All Facts (3)
The similarity measure is changed from cosine similarity to Total Variation similarity to suit probability distribution comparison on the simplex.
50%UnverifiedThe computational complexity of Total Variation similarity is of the same order as that of cosine similarity.
50%UnverifiedWhether the simplex constraint limits the expressive freedom of embeddings, and the training stability under Total Variation similarity, still require more extensive experiments.
50%