arXiv Computation and Language November 4, 2026 By Karl Li

Sparse Autoencoder Attribution Graphs Are Decided at the Margin

Read the original on arXiv Computation and Language →

The paper investigates attribution graphs for sparse autoencoders (SAEs) by examining how zeroing a feature affects subsequent layer activations. It shows that in TopK SAEs, removing a feature often causes another feature to cross the activation cutoff, and these crossings account for most of the response’s energy. The authors demonstrate that a margin‑based graph, which adds first‑order changes to preactivations, achieves low error (≤0.033) while accurately capturing the impact of feature ablations.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Computation and Language.

benchmarks

arXiv Machine Learning

Sep 14

Where Decoder Cosine Similarity Fails for SAE Feature Flow Discovery

arXiv:2609. 12591v1 Announce Type: new Abstract: Foundation models are increasingly adapted through fine-tuning, model editing, and alignment procedures while retaining previously acquired capabilities.

By Hendrik Droste, Christian Medeiros Adriano, Kathrin Korte, Holger Giese

fine-tuning benchmarks safety

arXiv Machine Learning

Sep 30

GraphVQ: Structure-Aware Autoregressive Decoding over Context-Quantized Graph Tokens

arXiv:2609.37604v1 Announce Type: new Abstract: Graph foundation models need a discrete token representation, but casting a graph as a generatable token sequence faces a structural obstacle: edges sp...

By Yuxiang Yao, Zijun Zhao

nlp efficiency

arXiv AI

Oct 16

Representation Redistribution in Sparse Autoencoders

The paper introduces the Redistribution Sensitivity Score (RSS) to detect consistent changes in sparse feature activations of Sparse Autoencoders (SAEs) when prompts are transformed. RSS identifies large, directionally consistent feature responses and evaluates their impact on downstream readouts across 27 pretrained SAEs from eight language models. Experiments show that combining RSS support erosion with the clean margin predicts recurrent safety readout failures with high AUROC, demonstrating RSS’s utility in quantifying representation changes without conflating them with failure.

By Ziye Mei, Haotian Ma

llms safety

arXiv Machine Learning

Aug 14

Where You Measure Decides What You Measure: Position Selection in Ablation-Based SAE Evaluation

arXiv:2608. 13337v1 Announce Type: new Abstract: Sparse autoencoders are meant to name the things a language model computes, and the usual way to check that a latent matters is to switch it off and see what changes.

By Valentin No\"el

llms benchmarks

arXiv Machine Learning

Jul 24

Are Single-Token Sparse Autoencoder Features Causally Necessary? Layer-Depth and SAE-Family Effects

arXiv:2607. 20596v1 Announce Type: new Abstract: Sparse autoencoder (SAE) features are used to interpret and steer large language models, yet whether a feature's causal role is stable across SAE families remains untested.

By Seonglae Cho, Zekun Wu, Kleyton Da Costa, Rishi Kalra, Ilham Wicaksono, Adriano Koshiyama

llms benchmarks safety

arXiv AI

Jun 11

Unstable Features, Reproducible Subspaces: Understanding Seed Dependence in Sparse Autoencoders

arXiv:2606. 12138v1 Announce Type: cross Abstract: Sparse autoencoders (SAEs) are widely used to interpret neural network representations, but their utility depends on whether the learned features are reproducible across training runs.

By Gleb Gerasimov, Timofei Rusalev, Nikita Balagansky, Daniil Laptev, Vadim Kurochkin, Daniil Gavrilov

One email a morning, machine-written

Related stories

Where Decoder Cosine Similarity Fails for SAE Feature Flow Discovery

GraphVQ: Structure-Aware Autoregressive Decoding over Context-Quantized Graph Tokens

Representation Redistribution in Sparse Autoencoders

Where You Measure Decides What You Measure: Position Selection in Ablation-Based SAE Evaluation

Are Single-Token Sparse Autoencoder Features Causally Necessary? Layer-Depth and SAE-Family Effects

Unstable Features, Reproducible Subspaces: Understanding Seed Dependence in Sparse Autoencoders