AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Ensembling Pruned Attention Heads For Uncertainty-Aware Efficient Transformers

Firas Gabetni, Giuseppe Curci, Andrea Pilzer, Subhankar Roy, Elisa Ricci, Gianni Franchi

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Theory of self-attention dynamics · 4 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
Use Hydra Ens (Circ) instead of Deep EnsemblesHydra Ensembles · ImageNet-1K OOD avgAUROC: 85.48 -> 86.29Improved
Merge three Taylor-pruned members into Hydra Ens (ablation)Taylor pruning · ImageNet-1Kaccuracy: 80.68 -> 81.20Improved
Add ensembling of three circuit-pruned members to CircAvg (ablation)CircAvg · SST-2 OOD avgAUROC: 75.04 -> 77.6Improved
Add ensembling of three pruned members to CircAvg (ablation)CircAvg · Zero-shot classification avgaccuracy: 71.91 -> 74.00Improved
Use circuit-based head selection instead of Taylor pruning, zero-shot (ablation)OpenCLIP ViT-B/32 pruned heads · Zero-shot classification avgaccuracy: 60.75 -> 71.91Improved
Zero-shot Taylor head pruning of the dense model (ablation)ViT-B/16 · ImageNet-1k zero-shot pruningaccuracy: 80.67 -> 66.46Hurt
Use circuit-based instead of Taylor head pruning, zero-shot (ablation)ViT-B/16 · ImageNet-1k zero-shot pruningaccuracy: 66.46 -> 73.61Improved
Use circuit pruning (CircOOD) instead of dense model for ID-OOD separation (ablation)ViT-B/16 · CIFAR-100 OOD CLS-token distanceEuclidean distance: 1.75 -> 2.26Improved
Extract the circuit for accuracy (CircAcc) instead of OOD (CircOOD) (ablation)ViT-B/16 · ImageNet-1Kaccuracy: 71.11 -> 73.41Improved
Extract the circuit for OOD (CircOOD) instead of accuracy (CircAcc) (ablation)ViT-B/16 · ImageNet-1K OOD avgAUROC: 80.80 -> 81.89Improved
Use Hydra Ens (Circ) instead of ViLUHydra Ensembles · Zero-shot ImageNet-1k OODAUROC: +1.3% AUROC, -3.5% FPR95, +4.0% AUPRImproved
Use Hydra Ensembles instead of Deep Ensembles on SST-2Hydra Ensembles · SST-2 OOD avgAUROC: AUROC (+2.8%), FPR95 (-7.6%), AUPR (+2.2%)Improved

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