Analyzing Feed-Forward Blocks in Transformers through the Lens of Attention Maps
This work addresses the interpretation of Transformer components for researchers, but it is incremental as it focuses on a less-studied part without major breakthroughs.
The paper tackled the problem of interpreting feed-forward blocks in Transformers by analyzing their effects on attention maps, revealing that these blocks modify input contextualization to emphasize linguistic compositions and show redundancy with surrounding components.
Transformers are ubiquitous in wide tasks. Interpreting their internals is a pivotal goal. Nevertheless, their particular components, feed-forward (FF) blocks, have typically been less analyzed despite their substantial parameter amounts. We analyze the input contextualization effects of FF blocks by rendering them in the attention maps as a human-friendly visualization scheme. Our experiments with both masked- and causal-language models reveal that FF networks modify the input contextualization to emphasize specific types of linguistic compositions. In addition, FF and its surrounding components tend to cancel out each other's effects, suggesting potential redundancy in the processing of the Transformer layer.