The reviewed record of science sign in
Pith

arxiv: 2004.05916 · v2 · pith:AGQVS5L3 · submitted 2020-04-10 · cs.LG · cs.CL

Telling BERT's full story: from Local Attention to Global Aggregation

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:AGQVS5L3record.jsonopen to challenge →

classification cs.LG cs.CL
keywords attentionlocalbehaviordistributionsmixingpatternsanalyzeattribution
0
0 comments X
read the original abstract

We take a deep look into the behavior of self-attention heads in the transformer architecture. In light of recent work discouraging the use of attention distributions for explaining a model's behavior, we show that attention distributions can nevertheless provide insights into the local behavior of attention heads. This way, we propose a distinction between local patterns revealed by attention and global patterns that refer back to the input, and analyze BERT from both angles. We use gradient attribution to analyze how the output of an attention attention head depends on the input tokens, effectively extending the local attention-based analysis to account for the mixing of information throughout the transformer layers. We find that there is a significant discrepancy between attention and attribution distributions, caused by the mixing of context inside the model. We quantify this discrepancy and observe that interestingly, there are some patterns that persist across all layers despite the mixing.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.