AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

SPICE: Simple Polysemantic feature Interpretation via Clustering-based Explanations

Sehyun Lee, Dahee Kwon, Damin Lee, Jaesik Choi

ECCV 2026 · Architectures and efficiency › RNNs and normalization · approach: Mechanistic circuit interpretability

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
add adaptive per-neuron cluster discovery instead of fixed K=2 (Input x Gradient) (ablation)SPICE · ViT-B/16 B11separability: 1.027 -> 1.308Improved
add per-dimension attribution normalization on top of adaptive K (ablation)SPICE · ViT-B/16 B11separability: 1.308 -> 1.577Improved
replace CRP with Input x Gradient under fixed K=2 (ablation)PURE-style clustering · ViT-B/16 B11separability: 1.036 -> 1.027No clear effect
use constant K in {2,3,4} instead of adaptive K (ablation)SPICE · ViT-B/16 B11separability: 1.577 -> 1.027/1.025/1.026Hurt
add feature-wise normalization (ablation)SPICE · ResNet-50 L4.2separability: 1.017 -> 1.293Improved
add feature-wise normalization (ablation)SPICE · ViT-B B2separability: 1.065 -> 1.137Improved
add feature-wise normalization (ablation)SPICE · ViT-B B11separability: 1.507 -> 1.577Improved
use LeGrad instead of Input x Gradient attribution (ablation)SPICE · ViT-B/16 B11separability: 1.577 -> 1.493No clear effect
use static global tau>=0.9 instead of adaptive tau (ablation)SPICE · ViT-B B6separability / coverage: score 1.25 -> 2.18, coverage 100/100 -> 7/100Mixed
use SPICE instead of LE in deep layersLE baseline · ViT-B B11separability: 1.357 -> 1.577Improved
use SPICE instead of LE in early layersLE baseline · ViT-B B2separability: 1.199 -> 1.137Hurt
use SPICE instead of LE in DenseNet L4.16LE baseline · DenseNet L4.16separability: 1.191 -> 1.033Hurt

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