Metraipolis › archive › dc_dd05590a34ee

GradLev: Token-Parallel Test-Time Training Via Costate Prediction — arXiv:2609.34174

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.34174
License (as archived)
CC0-1.0
Archived
2026-10-07T21:02:13.151Z
SHA-256 of the archived text
a9332b87c1638646b12904d6d31cb452766e38ff2fcd8807a74b35149c9d54b8
Stable URL of this page
https://metraipolis.com/documento/dc_dd05590a34ee

How to cite

GradLev: Token-Parallel Test-Time Training Via Costate Prediction — arXiv:2609.34174 — arXiv (en), https://arxiv.org/abs/2609.34174 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_dd05590a34ee (Metraipolis archive dc_dd05590a34ee, sha256 a9332b87c1638646b12904d6d31cb452766e38ff2fcd8807a74b35149c9d54b8)
BibTeX
@misc{dcdd05590a34ee,
  title = {GradLev: Token-Parallel Test-Time Training Via Costate Prediction — arXiv:2609.34174},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.34174},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_dd05590a34ee; sha256 a9332b87c1638646b12904d6d31cb452766e38ff2fcd8807a74b35149c9d54b8},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.34174. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Test-time training (TTT) allows a model to improve its predictions at inference time by updating weights after every observed token. However, sequential gra- dient writes make parallel training difficult. We observe that, given layer inputs and activation gradients (costates), online gradient descent admits exact parallel scans for both forward evaluation and reverse backpropagation. GradLev lever- ages this duality: a causal auxiliary network predicts costates across all tokens in parallel; associative scans compute the adapted weights and forward activations and propagate gradients backward; and the resulting gradient targets supervise the predictor via a consistency loss. Exact consistency guarantees exact recovery of the sequential online learner. At deployment, the auxiliary predictor is discarded, and the model updates natively via token-by-token forward and backward passes.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.34174 — 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 2026-10-07T20:29:02.883Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).