Current browse context:
cs.LG
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: Unified Interpretation of Softmax Cross-Entropy and Negative Sampling: With Case Study for Knowledge Graph Embedding
(Submitted on 14 Jun 2021 (v1), last revised 16 Mar 2022 (this version, v4))
Abstract: In knowledge graph embedding, the theoretical relationship between the softmax cross-entropy and negative sampling loss functions has not been investigated. This makes it difficult to fairly compare the results of the two different loss functions. We attempted to solve this problem by using the Bregman divergence to provide a unified interpretation of the softmax cross-entropy and negative sampling loss functions. Under this interpretation, we can derive theoretical findings for fair comparison. Experimental results on the FB15k-237 and WN18RR datasets show that the theoretical findings are valid in practical settings.
Submission history
From: Hidetaka Kamigaito [view email][v1] Mon, 14 Jun 2021 09:07:02 GMT (55kb)
[v2] Sat, 3 Jul 2021 07:12:21 GMT (55kb,D)
[v3] Mon, 11 Oct 2021 19:38:51 GMT (56kb,D)
[v4] Wed, 16 Mar 2022 16:49:26 GMT (56kb,D)
Link back to: arXiv, form interface, contact.