NeurIPS 2020

All Word Embeddings from One Embedding

Meta Review

The paper proposed a novel mechanism to generate word embedings by a function rather than store the whole embeddings for each word, which leads to a high compression of space needed for word embeddings. The idea is novel and can be used broadly. The effectiveness is verified by the experiments. Reviewers suggest to conduct experiments in larger network and more comparison with baselines to show the saving of computation time, energy, etc.