跳到正文
HuggingFace Daily Papers(社區熱門論文)·· 2026-10-14AI 評分43

稀疏自動編碼器的數據邊際收益遞減:功能表現不保證特徵可重現性

Diminishing Returns of Data in Sparse Autoencoders: Functional Behavior does not Guarantee Feature Reproducibility

AI 導讀

研究顯示,稀疏自動編碼器(SAE)的重建表現往往可用較少激活 token 接近全數據參照值,但特徵可重現性仍隨數據增加而提升。研究以 5M 至 500M 激活 token 的數據預算,訓練 4 種 SAE 架構,並在 3 個語言模型上比較功能表現與特徵一對一重現。Gemma-2-2B BatchTopK 的嚴格匹配特徵比例由 5M token 時的 2% 升至 50M 時的 29%。

正文

Published on Oct 14

Authors:

,

,

Abstract

Sparse autoencoders can be trained with far fewer activation tokens when data budgets align with evaluation goals, though feature-level reproducibility demands larger scales.

Sparse autoencoders (SAEs) are typically trained on hundreds of millions to billions of activation tokens, yet little work has tested whether this scale is necessary. We study how much activation data SAEs need by training four SAE architectures on three language models with budgets ranging from 5M tokens to a 500M-token reference. We measure two notions of convergence: functional behavior on standard evaluations and one-to-one reproducibility of learned features. Functional data requirements are strongly metric-dependent: reconstruction often approaches the full-data reference at substantially smaller budgets, whereas feature reproducibility continues to improve with data. For Gemma-2-2B BatchTopK, the fraction of strictly matched features rises from 2\% at 5M tokens to 29\% at 50M. Data selection also matters: random sampling better preserves the original activation distribution and generally favors reconstruction, whereas diversity-based sampling frequently improves AutoInterp. A theoretical analysis separates statistical consistency of subsampled training from the stronger identifiability and optimization conditions required for one-to-one feature recovery. Our results show that SAE training can often be made far more data-efficient when budgets are matched to the goal, and feature-level claims are backed by explicit reproducibility checks, paving the way toward cheaper and more reliable SAE research.

View arXiv page View PDF Add to collection

Get this paper in your agent:

hf papers read 2610.20680

Don't have the latest CLI?

curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2610.20680 in a model README.md to link it from this page.

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2610.20680 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2610.20680 in a Space README.md to link it from this page.

Collections including this paper 0

No Collection including this paper

Add this paper to a collection to link it from this page.

來源:HuggingFace Daily Papers(社區熱門論文) · huggingface.co