TX-Ray: Quantifying and Explaining Model-Knowledge Transfer in (Un-)Supervised NLP
Published in
Conference on Uncertainty in Artificial Intelli...(2019)External Links:
TL
TL;DR
The established computer vision explainability principle of 'visualizing preferred inputs of neurons' is modified to make it usable transfer analysis and NLP, and TX-Ray expresses neurons as feature preference distributions to quantify fine-grained knowledge transfer or adaptation and guide human analysis.