Metraipolis › archive › dc_b87f07b6ddc9

Towards Identifying the Dataset Biases Causing Phantom Transfer — arXiv:2609.14449

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.14449
License (as archived)
CC0-1.0
Archived
2026-10-16T00:10:40.697Z
SHA-256 of the archived text
ed8184ea7c78221dadf6219d9583852909363c7827541ebd65caf0a4ccfa01b7
Stable URL of this page
https://metraipolis.com/documento/dc_b87f07b6ddc9

How to cite

Towards Identifying the Dataset Biases Causing Phantom Transfer — arXiv:2609.14449 — arXiv (en), https://arxiv.org/abs/2609.14449 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_b87f07b6ddc9 (Metraipolis archive dc_b87f07b6ddc9, sha256 ed8184ea7c78221dadf6219d9583852909363c7827541ebd65caf0a4ccfa01b7)
BibTeX
@misc{dcb87f07b6ddc9,
  title = {Towards Identifying the Dataset Biases Causing Phantom Transfer — arXiv:2609.14449},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.14449},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_b87f07b6ddc9; sha256 ed8184ea7c78221dadf6219d9583852909363c7827541ebd65caf0a4ccfa01b7},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.14449. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Recent work has shown that a teacher model can transfer a bias to a student through a dataset from which every explicit reference to that bias has been filtered out, and that none of the tested data-level defenses reliably removes or detects such a bias, even when the defender knows what to look for. To shed light on the hidden traces these biases leave, we embed a dataset's completions with Sentence-BERT, subtract the embeddings of clean reference completions, and compare the result to an open vocabulary of candidate topics. This simple signature identifies the topic of the bias with a Matthews correlation coefficient of 0.83 when the attacker's teacher model is known, and 0.46 when it is not. We also observe that different teacher models appear to express the same bias through different vocabulary.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-15 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.14449 — 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 2026-10-15T23:54:14.697Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-15?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).