Metraipolis › archive › dc_63f8eca12ec7

Good Memory Has ECC: Evaluating the Memory of Vision-Language Models Beyond Accuracy — arXiv:2609.00103

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.00103
License (as archived)
CC0-1.0
Archived
2026-10-17T06:32:47.277Z
SHA-256 of the archived text
a36b4cc3b5730bf76e2a674438757c91ffe5451ed30bf5cfb582a1ef34e29624
Stable URL of this page
https://metraipolis.com/documento/dc_63f8eca12ec7

How to cite

Good Memory Has ECC: Evaluating the Memory of Vision-Language Models Beyond Accuracy — arXiv:2609.00103 — arXiv (en), https://arxiv.org/abs/2609.00103 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_63f8eca12ec7 (Metraipolis archive dc_63f8eca12ec7, sha256 a36b4cc3b5730bf76e2a674438757c91ffe5451ed30bf5cfb582a1ef34e29624)
BibTeX
@misc{dc63f8eca12ec7,
  title = {Good Memory Has ECC: Evaluating the Memory of Vision-Language Models Beyond Accuracy — arXiv:2609.00103},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.00103},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_63f8eca12ec7; sha256 a36b4cc3b5730bf76e2a674438757c91ffe5451ed30bf5cfb582a1ef34e29624},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.00103. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Memory is widely viewed as an important unsolved problem for LLMs and VLMs, and current benchmarks typically evaluate it by testing accuracy over long text or video. However, accuracy alone misses properties that matter for real long-horizon tasks. We introduce ECCBench, a benchmark and evaluation protocol that measures memory beyond a system's capacity--its raw accuracy at a specific budget--via three axes we call ECC: efficiency--the computation, in FLOPs, needed to answer from memory; compression--whether compressible inputs are remembered more accurately or efficiently; and calibration--whether the system abstains in response to its own uncertainty and the cost of an error. We find that pretrained VLMs compress their memory over text but not video and are poorly calibrated on both. Among a broader set of memory backbones, several non-Transformer architectures achieve better compression-calibration tradeoffs than RoPE Transformers, suggesting they may be useful components for agents operating over long horizons.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-02 in cs.AI).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.00103 — 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 2026-10-17T06:15:12.743Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-02?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).