Influence Patterns for Explaining Information Flow in BERT

Part of Advances in Neural Information Processing Systems 34 pre-proceedings (NeurIPS 2021)

Paper Supplemental

Bibtek download is not available in the pre-proceeding


Kaiji Lu, Zifan Wang, Piotr Mardziel, Anupam Datta


While attention is all you need may be proving true, we do not know why: attention-based transformer models such as BERT are superior but how information flows from input tokens to output predictions are unclear. We introduce influence patterns, abstractions of sets of paths through a transformer model. Patterns quantify and localize the flow of information to paths passing through a sequence of model nodes. Experimentally, we find that significant portion of information flow in BERT goes through skip connections instead of attention heads. We further show that consistency of patterns across instances is an indicator of BERT’s performance. Finally, we demonstrate that patterns account for far more model performance than previous attention-based and layer-based methods.