EPISODE · Jun 16, 2026 · 8 MIN
AI计算广告论文播报｜6月15日：出价均值陷阱、LLM改写持续训练与工业决策范式收敛
from 周六9点半
本期聚焦一个核心判断：商业化决策系统正从"模型够聪明就行"走向"训练、奖励、上线、回滚每一环都要可控"。三篇来自出价、搜索改写和电商定价的工业论文，骨架竟然是同一套——离线学习 + 在线约束 + 持续重训 + 可回滚。本期重点* DRIVE: Distributional and Retrieval-Augmented Bidding with Value Evaluation——直接面向广告自动出价，用GMM多峰动作头+历史轨迹检索+IQL价值打分的三段式推断结构，专治DT类模型的"Average Action陷阱"（高价和低价都合理，模型却输出无用的中间值）。AuctionNet上把PDiT收益提升约19%。* CoRe: A Continuously Reward-Finetuned LLM Query Rewriter for Multi-Stage Context-Awar...去小宇宙查看完整单集简介在小宇宙查看该单集文稿
Embed this episode
Ready to play
AI计算广告论文播报｜6月15日：出价均值陷阱、LLM改写持续训练与工业决策范式收敛
0:00
8:01
1×
No transcript for this episode yet
Similar Episodes
No similar episodes found.
Similar Podcasts
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 16, 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!