Metraipolis › archive › dc_fbc644f5cd0b

Gradients for Interventions and Activations for Detection: Targeted Feature Learning in Language Models — arXiv:2609.32355

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.32355
License (as archived): CC0-1.0
Archived: 2026-10-07T21:50:04.429Z
SHA-256 of the archived text: 1269c45ddb65dbc399166028da074c672a384f828c1715997c49aeeea2b8cd61
Stable URL of this page: https://metraipolis.com/documento/dc_fbc644f5cd0b

How to cite

Gradients for Interventions and Activations for Detection: Targeted Feature Learning in Language Models — arXiv:2609.32355 — arXiv (en), https://arxiv.org/abs/2609.32355 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fbc644f5cd0b (Metraipolis archive dc_fbc644f5cd0b, sha256 1269c45ddb65dbc399166028da074c672a384f828c1715997c49aeeea2b8cd61)

BibTeX

@misc{dcfbc644f5cd0b,
  title = {Gradients for Interventions and Activations for Detection: Targeted Feature Learning in Language Models — arXiv:2609.32355},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.32355},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fbc644f5cd0b; sha256 1269c45ddb65dbc399166028da074c672a384f828c1715997c49aeeea2b8cd61},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.32355. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Model-internal features can be studied through both their ability to identify a specified concept and their causal effect when manipulated, e.g., through steering or weight editing. A prominent approach to feature learning is Sparse Autoencoders (SAEs), which learn broad feature dictionaries whose relation to particular concepts is typically identified post hoc. However, many interpretability questions are instead hypothesis-driven and concern a concept specified in advance. We study this setting as targeted feature learning, where a single feature is constructed for such a predefined concept. We present a controlled comparison across three model signals (activation values, activation gradients, and parameter gradients) and two estimators (contrastive mean and a learned one-dimensional encoder-decoder), yielding six targeted methods, with CAA and GRADIEND as existing instances and four new methods covering the remaining combinations. We compare these methods against pretrained SAEs across 15 tasks and three language models, evaluating both detection and causal intervention. Across models, the strongest detection performance is achieved by contrastive activation value methods, whereas the strongest intervention performance is achieved by gradient-based methods. Overall, our results show that targeted feature quality depends jointly on the model signal and estimator, with detection and intervention capturing complementary properties.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.32355 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T21:27:08.413Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).