AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Weight-sparse transformers have interpretable circuits

Leo Gao, Achyuta Rajaram, Jacob Coxon, Soham V. Govande, Bowen Baker, Daniel P Mossing

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Mechanistic circuit interpretability · 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
train weight-sparse model instead of dense model at matched pretraining lossGPT-2 style transformer on Python code · 20 binary tasksminimal circuit size: roughly 16-fold smaller circuitsImproved
decrease L0 (make weights sparser) at fixed total parameters (ablation)weight-sparse transformer · pretraining loss vs circuit sizecapability and interpretabilityMixed
increase model width d_model at fixed L0 (ablation)weight-sparse transformer · pretraining loss vs circuit sizecapability and interpretabilityImproved
scale total parameter count of sparse models (ablation)weight-sparse transformer · capability-interpretability frontierPareto frontierImproved
ablate the circuit nodes only, leaving rest of network intact (ablation)sparse circuits · task losstask lossHurt
lengthen list context to dilute the averaged bracket-depth featurebracket counting circuit · flat vs nested list closingp(correct)Hurt
steer sparse-model quote-type channel and map via bridges to dense modelbridged dense model · single_double_quote 4-layerP(single quote)Improved
steer sparse-model colon channel and map via bridges to dense modelbridged dense model · while_return_true 4-layerP(colon)Improved
use weight-sparse training instead of dense trainingtraining compute · comparable capabilitytraining and inference compute: 100-1000x more computeHurt

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