Metraipolis › archive › dc_b4e976ec80f8

What Makes Synthetic Hard Negatives Work in Vision-Language Pretraining? — arXiv:2610.09700

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.09700
License (as archived): CC0-1.0
Archived: 2026-10-08T11:11:22.420Z
SHA-256 of the archived text: e0eccd883d5af5f4eff48cbb5e70af825409620ec0013531c83c31e13746b312
Stable URL of this page: https://metraipolis.com/documento/dc_b4e976ec80f8

How to cite

What Makes Synthetic Hard Negatives Work in Vision-Language Pretraining? — arXiv:2610.09700 — arXiv (en), https://arxiv.org/abs/2610.09700 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_b4e976ec80f8 (Metraipolis archive dc_b4e976ec80f8, sha256 e0eccd883d5af5f4eff48cbb5e70af825409620ec0013531c83c31e13746b312)

BibTeX

@misc{dcb4e976ec80f8,
  title = {What Makes Synthetic Hard Negatives Work in Vision-Language Pretraining? — arXiv:2610.09700},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.09700},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_b4e976ec80f8; sha256 e0eccd883d5af5f4eff48cbb5e70af825409620ec0013531c83c31e13746b312},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.09700. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Synthetic hard negatives generated in the representation space have proven effective for unimodal self-supervised learning, but transferring this idea to vision-language pretraining is not straightforward. We analyze six representation-space synthesis strategies and identify two failure modes in their transfer to vision-language pretraining: cross-modal constructions that produce overly easy negatives or pull them toward the query, and intra-modal constructions that incorporate the matched positive. We also observe logit-scale saturation when training with synthetic hard negatives and a learnable temperature, and find that fixing the temperature improves downstream performance. Using this geometric analysis we propose SNAP, which generates intra-modal hard negatives that never involve the positive from either modality, avoiding both failure modes entirely. SNAP is model-agnostic, requires no external generative models, and adds less than 10% training time overhead. Evaluated on top of CLIP and FLIP across multiple architectures and datasets, SNAP delivers consistent improvements on zero-shot retrieval, zero-shot classification, and linear probe evaluation.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2610.09700 — 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 2026-10-08T10:14:06.118Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).