AI计算广告论文播报 06-09:排序scaling的下一步是换连接方式 episode artwork

EPISODE · Jun 10, 2026 · 8 MIN

AI计算广告论文播报 06-09:排序scaling的下一步是换连接方式

from 周六9点半

本期聚焦工业排序与召回的三大新动向:更深的残差、更长的序列、更统一的生成式架构。我们用两篇硬货告诉你——单纯堆层数堆长度已经不是最划算的路径,残差怎么连、ID怎么编、token怎么打包,才决定你的scaling曲线还能不能继续往上走。本期重点* 解耦稳定性与适应性的可扩展CTR残差结构(DeRes: Decoupling Residual Stability and Adaptivity for Scalable CTR Prediction):把残差拆成identity传送带和跨层注意力检索员两条并行通道,再用SiLU替Softmax让多兴趣可同时激活、无关历史可被负权重抑制;scaling指数是OneTrans的1.66倍,8层就能追平对手16层。* 超越Item ID的短视频长序列建模(Beyond Item IDs: Scaling Short-Form-Video Reco...去小宇宙查看完整单集简介在小宇宙查看该单集文稿

Episode metadata supplied by the publisher feed · Published Jun 10, 2026

Embed this episode

Ready to play

AI计算广告论文播报 06-09:排序scaling的下一步是换连接方式

0:00 8:05

No transcript for this episode yet

We transcribe on demand. Request one and we'll notify you when it's ready — usually under 10 minutes.

No similar episodes found.

No similar podcasts found.

Frequently Asked Questions

How long is this episode of 周六9点半?

This episode is 8 minutes long.

When was this 周六9点半 episode published?

This episode was published on June 10, 2026.

Can I download this 周六9点半 episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.
URL copied to clipboard!