Skip to content

Attention Is Not Explanation

Authors: Jain & Wallace Year: 2019 ArXiv/Link: https://arxiv.org/abs/1902.10186

Summary

Critiques the assumption that attention weights explain model decisions, highlighting limitations of attention visualization.

Key Concepts

  • Attention critique
  • Interpretability limits
  • Explanation assumptions
  • Attention patterns
  • Model transparency

Impact

Cautions against over-interpreting attention weights

Category

Interpretability