Towards a Densing Law for User Representation Learning at Billion-Scale Capacity

  • 类型:arxiv
  • 标识:2608.23392
  • 链接:https://arxiv.org/abs/2608.23392
  • 主分类:llm-infra
  • 形态:method
  • 被引:0
  • 被引来源:Semantic Scholar + OpenAlex
  • S2被引:0
  • OpenAlex被引:0
  • 影响力被引:0
  • TLDR:The proposed User Behavioral Densing Law is proposed, providing practical guidance for tokenization configuration selection in large-scale user representation learning and ALGN, an adaptive variable-length tokenization method that improves capacity allocation, is developed.
  • OpenAlex ID:W7204214481
  • OpenAlex DOI:10.48550/arxiv.2608.23392
  • DOI:10.48550/arxiv.2608.23392
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://arxiv.org/pdf/2608.23392
  • OpenAlex更新:2026-08-31
  • 待LLM分类:否
  • 标题中文:迈向十亿级容量用户表示学习的稠密定律
  • TLDR中文:本文提出 User Behavioral Densing Law,为大规模用户表示学习中的 tokenization 配置选择提供实用指导,并开发了 ALGN——一种自适应变长 tokenization 方法,可改善容量分配。
  • 成熟度:research
  • 场景:representation-learning、scaling-law
  • 来源文件:
  • /inbox/tom/_candidates/2026-08-25-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]