Erase-then-Delta Attention: Decoupling Erase and Write Addresses in Delta-Rule Linear Attention

  • 类型:arxiv
  • 标识:2606.26560
  • 链接:http://arxiv.org/abs/2606.26560v1
  • 主分类:llm-infra
  • 形态:method
  • 被引:2
  • 被引来源:Semantic Scholar
  • S2被引:2
  • OpenAlex被引:0
  • 影响力被引:0
  • TLDR:Erase-then-Delta Attention (EDA), a memory update rule that decouples where to erase from where to write, is proposed, suggesting that recurrent memory models should decide not only what to write, but also what stale information to erase and where.
  • OpenAlex ID:W7166170180
  • OpenAlex DOI:10.48550/arxiv.2606.26560
  • DOI:10.48550/arxiv.2606.26560
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://doi.org/10.48550/arxiv.2606.26560
  • OpenAlex更新:2026-07-19
  • 待LLM分类:否
  • 成熟度:research
  • 场景:linear attention、memory update rule、sequence modeling
  • 标题中文:Erase-then-Delta Attention:解耦 Delta 规则线性注意力中的擦除地址与写入地址
  • TLDR中文:提出 Erase-then-Delta Attention (EDA),一种将"在哪里擦除"与"在哪里写入"解耦的内存更新规则;研究表明循环记忆模型不仅应决定写入什么,还应决定擦除哪些陈旧信息以及擦除的位置。
  • 来源文件
  • /inbox/tom/_candidates/2026-06-27-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-06-26-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]