Metraipolis › archive › dc_ef7818bc448e

HLA-WM: Hybrid Linear Attention for Long-Horizon Video World Models — arXiv:2610.05739

Source: arXiv
Original: https://arxiv.org/abs/2610.05739
License (as archived): CC0-1.0
Archived: 2026-10-06T05:27:08.381Z
SHA-256 of the archived text: 07b01c390420421fb7ff882ad6f5ab07651e4c839fb499cbfe1308810bae5e13
Stable URL of this page: https://metraipolis.com/documento/dc_ef7818bc448e

How to cite

HLA-WM: Hybrid Linear Attention for Long-Horizon Video World Models — arXiv:2610.05739 — arXiv, https://arxiv.org/abs/2610.05739 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ef7818bc448e (Metraipolis archive dc_ef7818bc448e, sha256 07b01c390420421fb7ff882ad6f5ab07651e4c839fb499cbfe1308810bae5e13)

BibTeX

@misc{dcef7818bc448e,
  title = {HLA-WM: Hybrid Linear Attention for Long-Horizon Video World Models — arXiv:2610.05739},
  howpublished = {arXiv},
  url = {https://arxiv.org/abs/2610.05739},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ef7818bc448e; sha256 07b01c390420421fb7ff882ad6f5ab07651e4c839fb499cbfe1308810bae5e13},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.05739. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Long-horizon video world models require persistent memory to preserve scene consistency over extended rollouts. Softmax attention retains the full generation history through a growing KV cache, whereas recurrent linear attention compresses history into fixed-size states with substantially lower memory cost. However, we identify severe long-range forgetting in Gated DeltaNet (GDN), where information from distant but relevant scenes is progressively attenuated by subsequent state updates. To address this limitation, we propose HLA-WM, a training-free hybrid linear-attention framework that combines coarse-grained geometry-guided retrieval with fine-grained recurrent linear-state computation. HLA-WM exploits the affine structure of GDN to cache compact chunk-wise transition summaries, retrieve scene-relevant historical chunks using camera geometry, and recompose them into query-specific recurrent states. On the $60$-second SANA-WM-Bench, HLA-WM improves all six aggregate revisit-consistency and camera-control metrics of the base autoregressive generator without additional training, including a $0.74$ dB PSNR gain and a $28.5\%$ reduction in rotation error. The improvements persist after downstream refinement and generalize to MBench-A, where HLA-WM consistently improves all three revisit-consistency metrics across all four subsets and all evaluated inference modes over $547$ samples. At a $60$-second context, HLA-WM reduces historical-state memory by $12\times$ relative to full KV caching while incurring at most a $1.6\%$ reduction in inference throughput. These results demonstrate that selectively addressable recurrent memory can improve long-range scene recall while preserving the efficiency advantages of GDN. Project page: this https URL

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/2610.05739 — 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 2026-10-06T05:24:06.065Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).