KDD

When Offline Evaluation Misleads: A Diagnostic Protocol for Reward and Policy Selection in Delayed-Feedback Contextual Bandits

当离线评估产生误导时：延迟反馈情境多臂老虎机中奖励与策略选择的诊断协议

Sang Su Lee, Vineeth Loganathan, Shishir Dash, Vijay Raghavan

机构 * Thumbtack, Inc.（Thumbtack公司） 纠错

AI总结 针对延迟反馈情境多臂老虎机的奖励与策略选择，提出有序诊断协议，验证其有效性并揭示奖励排序易误导、个性化溢价易高估的结论。

Comments Accepted at the 5th Workshop on End-to-End Customer Journey Optimization (KDD 2026)

Certifying What Helps Customer-Return Timing: A Screen-and-Confirm Test for Conditioning Signals, and Why Decay Is Nearly Enough

验证什么有助于客户退货时机：一种针对条件信号的筛选与确认测试，以及为何衰减几乎足够

Sang Su Lee, Vineeth Loganathan, Shishir Dash, Vijay Raghavan

机构 * Thumbtack, Inc.（Thumbtack公司） 纠错

AI总结 本研究提出筛选与确认协议及无模型上限工具，验证了连续时间衰减几乎可满足客户退货时机预测，而额外添加的条件信号多为冗余或有害。

Comments Accepted at the 5th Workshop on End-to-End Customer Journey Optimization (KDD 2026)

Defending against Model Extraction for GNNs with Model Reprogramming

通过模型重编程防御图神经网络的模型提取攻击

Yan Wen, Zhenyi Wang, Heng Huang

机构 * University of Maryland, College Park（马里兰大学帕克分校） ； University of Central Florida（中佛罗里达大学） 纠错

AI总结 针对图神经网络的模型提取攻击，提出主动防御框架GraphRP，通过结构感知门控机制构建动态结构防火墙，可显著降低攻击有效性且保留良性效用。

Comments Accepted by KDD 2026