Metraipolis › archive › dc_1768ef071608

SAE Interprets MoE: Sparse AutoEncoders Reveal That MoE Experts Exhibit Human-Interpretable Fine-Grained Specialization — arXiv:2611.08089

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.08089
License (as archived)
CC0-1.0
Archived
2026-11-06T14:50:51.873Z
SHA-256 of the archived text
ee82f28e543d7383236e1e89ba012dbd6f47b9b146969449e2b2c62d6d0826f2
Stable URL of this page
https://metraipolis.com/documento/dc_1768ef071608

How to cite

SAE Interprets MoE: Sparse AutoEncoders Reveal That MoE Experts Exhibit Human-Interpretable Fine-Grained Specialization — arXiv:2611.08089 — arXiv (en), https://arxiv.org/abs/2611.08089 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_1768ef071608 (Metraipolis archive dc_1768ef071608, sha256 ee82f28e543d7383236e1e89ba012dbd6f47b9b146969449e2b2c62d6d0826f2)
BibTeX
@misc{dc1768ef071608,
  title = {SAE Interprets MoE: Sparse AutoEncoders Reveal That MoE Experts Exhibit Human-Interpretable Fine-Grained Specialization — arXiv:2611.08089},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.08089},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_1768ef071608; sha256 ee82f28e543d7383236e1e89ba012dbd6f47b9b146969449e2b2c62d6d0826f2},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.08089. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Mixture-of-Experts (MoE) models route each token to a few of many experts, yet what an individual expert is responsible for remains unsettled. Existing analyses count how often an expert is selected on domain-labeled corpora, an instrument that fixes the taxonomy in advance, stays coarse, and records selection rather than contribution. We instead study specialization in feature space by training a sparse autoencoder (SAE) on the output of an MoE layer. Since this output is a router-weighted sum of expert outputs and the SAE encoder is affine, the pre-activation of every latent decomposes exactly over the experts. Aggregated over a corpus, this gives each expert a dominant set of latents whose activation it principally drives. On layers 4, 30 and 46 of Qwen3-30B-A3B, contribution is far more concentrated than gating, most latents have only one or two dominators, and the overlap between dominant sets concentrates on a few expert pairs. Auto-interpreting these sets reveals a division of labor below coarse domains, with distinct experts for Java and front-end web code, or for geometry and for arithmetic and polynomials, while the coarse domains themselves take very unequal numbers of experts. Our SAEs and code are available at this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2611.08089 — 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 2026-11-06T14:21:42.642Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).