AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Influence Patterns for Explaining Information Flow in BERT

Kaiji Lu, Zifan Wang, Piotr Mardziel, Anupam Datta

NeurIPS 2021 · Architectures and efficiency › Attention and transformers · approach: Sparse and structured attention · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
keep only embedding-level influence-pattern nodes (GPR) vs. random nodes of the same count (ablation)BERT ablated to pattern nodes · SVA/RA/SA, all subtasksablated accuracy: SVA Obj. .55 -> .96; SA .52 -> .97Improved
keep GPR embedding patterns instead of attention-weight patterns (ablation)BERT ablated to pattern nodes · SVA/RA/SAablated accuracy: SVA WSC .49 -> 1.0; SA .47 -> .97Improved
keep GPR embedding patterns instead of internal-influence patterns (ablation)BERT ablated to pattern nodes · SVA/RA/SAablated accuracy: SVA WSC .50 -> 1.0; SA .49 -> .97Improved
keep GPR embedding patterns instead of conductance-based patterns (ablation)BERT ablated to pattern nodes · SVA/RA/SAablated accuracy: RA NA .71 -> .98; SVA Obj. .93 -> .96Improved
conductance baseline is strong, close to GPR on SA (ablation)conductance patterns vs GPR · SA SST-2ablated accuracy: SA .94 -> .97No clear effect
replace each skip node in the attention-level pattern with all A attention heads (ablation)Pi^a_+ · SVA/RA/SAablated accuracy: SVA Obj. .70 -> .50; SA .86 -> .48Hurt
attention-level GPR pattern versus random attention-level nodes (ablation)BERT ablated to pattern nodes · SVA/RA/SAablated accuracy: RA GA .50 -> .89; SVA Obj. .50 -> .70Improved
ablating to Pi^e_+ nodes versus the full unablated model (ablation)BERT-small and BERT-base · RA NA / RA GAaccuracy: RA NA .83 -> .98; RA GA .73 -> .99Improved
ablating to Pi^e_+ nodes versus the full unablated model on SVA (ablation)BERT-small · SVA Obj./Subj./WSC/APPaccuracy: SVA Obj. .96 -> .96; WSC 1.0 -> 1.0No clear effect
random-node baseline variability over 50 runs (ablation)Pi^e_rand · SA SST-2ablated accuracy: SA .52 +/- .030No clear effect
skip connections versus attention heads in attention-level patternsPi^a · all tasksshare of nodes in patterns: 75.4% of nodes in Pi_a are skipImproved
word like versus love/hate as clausal verb in SVA-Obj (SP)BERT-small · SVA-Obj (SP)accuracy: like .82; hate .99; love 1.0Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy