AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Temporal Sparse Autoencoders: Leveraging the Sequential Nature of Language for Interpretability

Usha Bhalla, Alex Oesterling, Claudio Mayrink Verdun, Himabindu Lakkaraju, Flavio Calmon

ICLR 2026 · Architectures and efficiency › RNNs and normalization · approach: Sparse dictionary representations · 13 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
add temporal contrastive loss to the SAE (ablation)Temporal SAE (Pythia-160m) · MMLU sparse probing k=5semantics and context probe accuracy: removing it: semantics -0.07, context -0.1Improved
add temporal contrastive loss to the SAE (ablation)Temporal SAE (Pythia-160m) · Pythia-160m smoothness (high)smoothness (lower is smoother): removing it: +0.07Improved
add temporal loss to SAE trainingMatryoshka SAE · Reconstruction (FVE, Pythia-160m)FVE: 0.95 -> 0.94No clear effect
add temporal loss to SAE trainingMatryoshka SAE · Autointerp (Pythia-160m)autointerp score: 0.83 -> 0.81No clear effect
add temporal loss to SAE trainingMatryoshka SAE · High-level feature smoothness (Gemma2-2b)smoothness (lower is smoother): 0.15 -> 0.10Improved
contrast with a random earlier token instead of the previous token (ablation)Temporal SAE (Pythia-160m) · MMLU sparse probing k=5context vs syntax probe accuracy: context +0.11, syntax -0.10, semantics -0.02Mixed
increase high-level split from 20% to 50% (ablation)Temporal SAE (Pythia-160m) · MMLU sparse probing k=5semantics, context and syntax probe accuracy: semantics +0.02, context +0.09, syntax -0.08Mixed
decrease high-level split from 20% to 10% (ablation)Temporal SAE (Pythia-160m) · Pythia-160m ablationfraction alive, probes: fraction alive -0.07; semantics -0.01, context +0.01No clear effect
steer with T-SAE high-level features instead of baseline SAE featuresGemma2-2b steering · Steering success vs coherence (30 features)steering Pareto frontierImproved

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