Newest papers

Strength Controls Change How Interventions Compare Across Sparse Autoencoder Codes

Jinchen Zhang, Zheyuan Liu, Yihan Zhu, Meng Jiang · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26489 · cited by 0 works (OpenAlex) · open-access PDF

Korean

스파르스 오토코더(SAE) 코드를 설정하는 것은 개입 방향과 크기 모두를 변경할 수 있으므로 개입 효과를 비교하는 것은 균등하게 유지되는 것을 지정하는 강도 제어를 필요로합니다. 표준 일치 및 출력 캘리브레이션은 다른 컨트롤을 제공하지만, 강도 일치 후 개입이 코드 간에 다른 효과를 제공하는지 여부는 불분명합니다. Gemma와 Llama의 고정된 기능 그룹 (NNLS)에 대한 이러한 컨트롤에 따라 개입 효과를 비교하고 다음을 관찰합니다 : (1) 전체 패치 스케일링. 입력 당 1 개의 요소와 함께 전체 출력 Kullback-Leibler (KL) 캘리브레이션은 코더 기반 참조 개입의 평균 효과 격차를 덜 남깁

Machine translation (Meta M2M100); the English original follows.

Abstract

Changing sparse autoencoder (SAE) codes can alter both intervention direction and magnitude, so comparing intervention effects requires a strength control specifying what is held equal. Norm matching and output calibration provide different controls, but whether interventions yield different effects across codes after strength matching remains unclear. We compare intervention effects under these controls for fixed feature groups in Gemma and Llama and observe: (1) Whole-patch scaling. With one coefficient per input, full-output Kullback-Leibler (KL) calibration leaves smaller mean effect gaps from the encoder-based reference intervention than total-norm matching for nonnegative least-squares (NNLS) refits. On new Llama inputs, total-norm matching loses 11 reference transfers, whereas full-output calibration preserves every transfer decision. (2) Positionwise scaling. On a small matched gender-agreement subset, positionwise norm matching leaves smaller observed mean gaps than full-output calibration. (3) Answer exclusion. Compared with unscaled NNLS, answer-excluded calibration reduces gaps in the measured subject-verb agreement comparisons but increases them for gender refits. Full-output KL calibration also reduces larger gaps for norm-matched constructed candidates, although matching coverage is incomplete in several settings. Since calibration uses the evaluated input's outputs, it serves as a diagnostic of residual disagreement.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Matching (statistics), Calibration, Mathematics, Autoencoder, Residual, Norm (philosophy)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange