[Adapt] 【ADAPT】Dimensionality Selection of Word Embedding
张盛瑶
sophie_zhang at sjtu.edu.cn
Tue Mar 5 23:10:56 CST 2019
Hi Adapters,
This week I'm going to give a talk on the paper "On the Dimensionality of Word Embedding" which is published in NeurIPS2018. This paper provides a theoretical understanding of word embedding and its dimensionality. It proposed the Pairwise Inner Product (PIP) loss to measure the similarity between word embeddings and reveals a fundamental bias-variance trade-off in dimensionality selection. By optimizing over the bias-variance trade-off of the PIP loss, the dimensionality selection for word embedding can be explicitly answered.
Best Regards,
Sophie
More information about the Adapt
mailing list