Metraipolis › archive › dc_23c03567bd7d

Buried in Textual Debt: Context Pruning with Visual Evidence Preservation for MLLM Agents — arXiv:2608.22963

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.22963
License (as archived)
CC0-1.0
Archived
2026-10-18T07:41:44.990Z
SHA-256 of the archived text
8b72f00b4703931bbba4b41f006cf0fa8f9bb5faa42b582b3a96e726368b5b66
Stable URL of this page
https://metraipolis.com/documento/dc_23c03567bd7d

How to cite

Buried in Textual Debt: Context Pruning with Visual Evidence Preservation for MLLM Agents — arXiv:2608.22963 — arXiv (en), https://arxiv.org/abs/2608.22963 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_23c03567bd7d (Metraipolis archive dc_23c03567bd7d, sha256 8b72f00b4703931bbba4b41f006cf0fa8f9bb5faa42b582b3a96e726368b5b66)
BibTeX
@misc{dc23c03567bd7d,
  title = {Buried in Textual Debt: Context Pruning with Visual Evidence Preservation for MLLM Agents — arXiv:2608.22963},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.22963},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_23c03567bd7d; sha256 8b72f00b4703931bbba4b41f006cf0fa8f9bb5faa42b582b3a96e726368b5b66},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.22963. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Multimodal Large Language Models (MLLMs) are increasingly deployed as multi-step agents, where explicit reasoning supports task decomposition and tool coordination but also accumulates self-generated text. Over long trajectories, this text can dominate the context and suppress visual evidence, creating textual debt. We observe that reasoning becomes redundant once task-relevant visual evidence is grounded, while stale hypotheses can misguide later inference when grounding remains uncertain. Pruning must therefore remove redundant text without discarding visual evidence. We propose SPARE, a Kullback-Leibler (KL)-guided framework for pruning accumulated reasoning in multimodal tool-use agents. SPARE uses a compact task-state summary as privileged diagnostic context. For each candidate segment, it replays the same model under the original and summary-conditioned contexts. Reverse-KL divergence from on-policy self-distillation (OPSD) then tests whether the summary sufficiently covers the segment without disrupting future reasoning. We further fine-tune the summarizer with supervised fine-tuning (SFT), enabling more compact summaries, broader coverage, and more aggressive pruning. Across multi-step visual tool-use benchmarks, SPARE achieves the highest average accuracy among pruning methods while removing 37.89-64.58\% of reasoning tokens. This favorable accuracy-context trade-off shows that reducing textual dominance restores reliance on visual evidence and mitigates over-conditioning on self-generated language.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-25 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2608.22963 — 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 2026-10-18T07:37:52.753Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-25?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).