Metraipolis › archive › dc_005ee6b536db

Are We Really Doing Few-Shot Learning? A Critical Examination of Pre-Training Assumptions — arXiv:2609.10851

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.10851
License (as archived): CC0-1.0
Archived: 2026-10-16T05:24:42.008Z
SHA-256 of the archived text: 386c8282192eef8cc6bd76507fe424c0deb1cc674fef017803c7f6647621259e
Stable URL of this page: https://metraipolis.com/documento/dc_005ee6b536db

How to cite

Are We Really Doing Few-Shot Learning? A Critical Examination of Pre-Training Assumptions — arXiv:2609.10851 — arXiv (en), https://arxiv.org/abs/2609.10851 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_005ee6b536db (Metraipolis archive dc_005ee6b536db, sha256 386c8282192eef8cc6bd76507fe424c0deb1cc674fef017803c7f6647621259e)

BibTeX

@misc{dc005ee6b536db,
  title = {Are We Really Doing Few-Shot Learning? A Critical Examination of Pre-Training Assumptions — arXiv:2609.10851},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.10851},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_005ee6b536db; sha256 386c8282192eef8cc6bd76507fe424c0deb1cc674fef017803c7f6647621259e},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.10851. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Few-shot learning is commonly evaluated under protocols that pre-train a model on a large auxiliary set whose classes are disjoint from the target episodes yet drawn from the same visual domain. This paper examines whether such protocols truly reflect low-data learning. We systematically compare no pre-training, class-disjoint in-domain pre-training, supervised out-of-domain pre-training, and label-free out-of-domain pre-training across eight datasets, three few-shot architectures, and multiple way-shot settings. Our results show that class disjointness alone is insufficient to remove the influence of target-domain data. In-domain pre-training improves over no pre-training by 33.41 percentage points on average, whereas supervised out-of-domain pre-training yields 23.75 percentage points, revealing a 9.66-point optimistic bias associated with domain overlap. Although out-of-domain pre-training is more realistic in applications where target-domain data are scarce, its effectiveness depends strongly on the compatibility between source and target domains. We further show that labeled source data are not strictly required, with an augmentation-based label-free strategy reaching an average gain of 27.71 percentage points and closely matching supervised out-of-domain pre-training at 27.97 percentage points. Finally, we introduce a descriptor-based source-selection strategy that estimates source-domain suitability before pre-training, reaching a median gap of only 1.37 percentage points to oracle selection. These findings highlight the need to move beyond in-domain pre-training as the default few-shot evaluation protocol, since it can overestimate performance in realistic scenarios where target-domain data are scarce.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-11 in cs.AI).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.10851 — 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 2026-10-16T05:17:52.087Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-11?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).