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

Indonesian

Perubahan kode autoencoder sparse (SAE) dapat mengubah kedua arah intervensi dan magnitud, sehingga membandingkan efek intervensi membutuhkan kontrol kekuatan yang menentukan apa yang dipegang sama. Perbandingan norma dan kalibrasi output menyediakan kontrol yang berbeda, tetapi apakah intervensi menghasilkan efek yang berbeda di antara kode setelah perbandingan kekuatan tetap tidak jelas. Kami membandingkan efek intervensi di bawah kontrol ini untuk kelompok fitur tetap di Gemma dan Llama dan memperhatikan: (1) scaling seluruh patch. Dengan satu koefisien per input, kalibrasi Kullback-Leibler (KL) output penuh mempertahankan kesenjangan efek rata-rata yang lebih kecil dari intervensi referensi berbasis encoder daripada total-norm matching untuk non-negative least-quarters (NNLS) refits. Pada inputs Llama baru, total-norm matching kehilangan 11 referensi eksklusif

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