Visualizing and Measuring the Geometry of BERT
TL;DR
小品文 paper，做的事情有：
- Attention weight 帶有 dependency 資訊
- Tree embedding: 要用 squared Euclidean distance 來看 (證明+視覺化)
- Word sense embedding 分離現象
另有精美 blog 網站 https://pair-code.github.io/interpretability/bert-tree/
Slide:
Please wait a minute for the embedded frame to be displayed. Reading it on a computer screen is better.