Tag
该论文提出了一种人类反馈引导的多智能体逆强化学习框架（HFG-MAIRL），通过局部马尔可夫博弈和Bradley–Terry偏好反馈对交互式驾驶行为进行奖励建模，在AIDE数据集上超越了现有IRL基线。
本文提出 FINDERS 框架，利用扩散语言模型的迭代去噪过程构建多粒度重建空间，通过去噪、分解、路由与压缩四个操作实现面向多场景的自适应用户表征，并在十个工业场景的实验与线上 A/B 测试中超越现有方法。
This paper introduces Knowledge-Geometry Decoupling (KGD), a method for pretrain-then-transfer in streaming recommendation systems. It separates pretrained behavioral knowledge from task-specific geometry, enabling continual model refresh without interference, and reports 4-12% improvements over baselines plus successful deployment at Shopee.
Introduces IRIS, a framework that learns dynamic user personas from implicit interaction streams without explicit feedback, outperforming static and memory-only baselines on decision prediction.
A free seminar on 22 July 2026 featuring Kim Stachenfeld from Google DeepMind discussing DataDIVER, a method using LLMs to discover interpretable symbolic models of human and animal behavior.
OdysSim presents a systematic investigation into behavioral foundation models for simulating human behavior, introducing the Soul taxonomy, a corpus of 21.4M interactions, and a training recipe that achieves state-of-the-art on 8 of 23 benchmark tasks while producing more human-like outputs.