Retrofitting Transformation
Dirk explains the process of inducing retrofitting from training data to unseen test data by learning a translation matrix between the original and retrofitted embedding spaces. He discusses how this process increases in-class similarity and potentially improves linear separability within classes, aiming for trivial separability through iterative retrofitting.In this clip
From this podcast
NLP Highlights
76 - Increasing In-Class Similarity by Retrofitting Embeddings with Demographics, with Dirk Hovy
Related Questions