SuPER-Mamba 框架提升 Mamba 模型在长上下文上的性能

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-15 04:00

研究人员推出 SuPER-Mamba，一个旨在增强 Mamba 等状态空间模型在处理扩展上下文时的性能的新框架。这种无需训练的方法利用一种称为“意外性”（Surprisal）的模型内在信号来优先处理长距离状态路由。通过计算逐位置的意外性并聚合响应，SuPER-Mamba 在无需查询依赖评分或额外训练的情况下，指导全局通道更新。在 LongBench-E 基准测试上的评估显示出显著的改进，SuPER-Mamba 将平均得分比基础模型提高了 94.7%，并比之前的最先进水平提高了 13.1%。 AI

影响增强了状态空间模型的长上下文处理能力，可能提高需要广泛文本分析的应用的性能。

排序理由详细介绍一种改进状态空间模型新方法的论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 17 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 详细介绍一种改进状态空间模型新方法的论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, infra; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.AI TIER_1 English(EN) · Zhetong Zhang, Gayan K Kulatilleke · 2026-10-15 04:00

SuPER-Mamba：用于 Mamba 中扩展上下文路由的惊奇度优先排序

arXiv:2610.16636v1 Announce Type: cross Abstract: State space models provide efficient linear-complexity processing for long sequences, but their performance frequently degrades as context length grows. We propose SuPER-Mamba, a training-free framework that uses Surprisal as a mo…

本文如何被排名

报道来源 [1]

SuPER-Mamba：用于 Mamba 中扩展上下文路由的惊奇度优先排序

相关实体

相关话题