<![CDATA[Shea的博客]]>

<![CDATA[Shea的博客]]>https://blog.youkuaiyun.com/HE2096798708zh-cnhttps://blog.youkuaiyun.com/<![CDATA[大模型偏好对齐横向对比：PPO、DPO、GRPO 怎么选？面试核心考点汇总]]>https://blog.youkuaiyun.com/HE2096798708/article/details/166353327https://blog.youkuaiyun.com/HE2096798708/article/details/166353327HE2096798708Tue, 22 Sep 2026 11:59:39 +0800<![CDATA[GRPO 组相对策略优化：面向数学/代码推理的偏好对齐，DPO 的 group 版本]]>https://blog.youkuaiyun.com/HE2096798708/article/details/166350596https://blog.youkuaiyun.com/HE2096798708/article/details/166350596HE2096798708Tue, 22 Sep 2026 11:36:23 +0800<![CDATA[DPO 直接偏好优化（二）：数据集、训练、调参、局限与变种]]>https://blog.youkuaiyun.com/HE2096798708/article/details/166347130https://blog.youkuaiyun.com/HE2096798708/article/details/166347130HE2096798708Tue, 22 Sep 2026 11:06:11 +0800<![CDATA[DPO 直接偏好优化（一）：为什么不需要奖励模型？从 RL 目标推导 DPO 损失]]>https://blog.youkuaiyun.com/HE2096798708/article/details/166335973https://blog.youkuaiyun.com/HE2096798708/article/details/166335973HE2096798708Tue, 22 Sep 2026 10:22:51 +0800<![CDATA[RLHF 三阶段详解：从奖励模型到 PPO 近端策略优化]]>https://blog.youkuaiyun.com/HE2096798708/article/details/165292552https://blog.youkuaiyun.com/HE2096798708/article/details/165292552HE2096798708Mon, 14 Sep 2026 11:37:23 +0800<![CDATA[大模型后训练（Post-training）入门：从 SFT 到偏好对齐]]>https://blog.youkuaiyun.com/HE2096798708/article/details/165290328https://blog.youkuaiyun.com/HE2096798708/article/details/165290328HE2096798708Mon, 14 Sep 2026 10:28:23 +0800<![CDATA[从神经元到深度网络：神经网络为什么能起作用]]>https://blog.youkuaiyun.com/HE2096798708/article/details/164854127https://blog.youkuaiyun.com/HE2096798708/article/details/164854127HE2096798708Thu, 10 Sep 2026 17:07:57 +0800<![CDATA[【NeurIPS 2024】LLM-ESR：用大语言模型破解序列推荐的长尾难题]]>https://blog.youkuaiyun.com/HE2096798708/article/details/146300303https://blog.youkuaiyun.com/HE2096798708/article/details/146300303HE2096798708Sun, 16 Mar 2025 19:59:46 +0800<![CDATA[【文献阅读】SPRec：用自我博弈打破大语言模型推荐的“同质化”困境]]>https://blog.youkuaiyun.com/HE2096798708/article/details/146213159https://blog.youkuaiyun.com/HE2096798708/article/details/146213159HE2096798708Wed, 12 Mar 2025 20:39:43 +0800<![CDATA[【文献阅读】DeepRAG：大语言模型的检索增强推理新范式]]>https://blog.youkuaiyun.com/HE2096798708/article/details/146212279https://blog.youkuaiyun.com/HE2096798708/article/details/146212279HE2096798708Wed, 12 Mar 2025 19:23:27 +0800<![CDATA[ LLM增强强化学习：开启智能决策的新篇章]]>https://blog.youkuaiyun.com/HE2096798708/article/details/145763099https://blog.youkuaiyun.com/HE2096798708/article/details/145763099HE2096798708Thu, 20 Feb 2025 21:01:38 +0800<![CDATA[【文献阅读分享】Mitigating Matthew Effect in Conversational Recommendation]]>https://blog.youkuaiyun.com/HE2096798708/article/details/145726617https://blog.youkuaiyun.com/HE2096798708/article/details/145726617HE2096798708Wed, 19 Feb 2025 12:58:57 +0800<![CDATA[【文献精读】AAAI24:FacetCRS：打破对话推荐系统中的“信息茧房”]]>https://blog.youkuaiyun.com/HE2096798708/article/details/145713479https://blog.youkuaiyun.com/HE2096798708/article/details/145713479HE2096798708Tue, 18 Feb 2025 20:01:52 +0800<![CDATA[文献阅读分享FedNews：联邦学习下的新闻推荐模型研究]]>https://blog.youkuaiyun.com/HE2096798708/article/details/145607855https://blog.youkuaiyun.com/HE2096798708/article/details/145607855HE2096798708Thu, 13 Feb 2025 11:29:35 +0800<![CDATA[文献阅读分享：LLM4Rerank——基于LLM的推荐系统自动重排序框架]]>https://blog.youkuaiyun.com/HE2096798708/article/details/145585435https://blog.youkuaiyun.com/HE2096798708/article/details/145585435HE2096798708Wed, 12 Feb 2025 10:13:30 +0800<![CDATA[KERL文献阅读分享：知识图谱与预训练语言模型赋能会话推荐系统]]>https://blog.youkuaiyun.com/HE2096798708/article/details/145557607https://blog.youkuaiyun.com/HE2096798708/article/details/145557607HE2096798708Mon, 10 Feb 2025 20:50:59 +0800<![CDATA[【文献阅读分享】PAP-REC：个性化自动提示生成框架✨]]>https://blog.youkuaiyun.com/HE2096798708/article/details/145556724https://blog.youkuaiyun.com/HE2096798708/article/details/145556724HE2096798708Mon, 10 Feb 2025 20:25:34 +0800<![CDATA[文献阅读分享《新闻推荐中的审议式多样性：操作化与实验用户研究》]]>https://blog.youkuaiyun.com/HE2096798708/article/details/145482310https://blog.youkuaiyun.com/HE2096798708/article/details/145482310HE2096798708Thu, 06 Feb 2025 21:31:27 +0800<![CDATA[文献阅读分享：MixRec——异构图协同过滤的创新之作]]>https://blog.youkuaiyun.com/HE2096798708/article/details/145452755https://blog.youkuaiyun.com/HE2096798708/article/details/145452755HE2096798708Wed, 05 Feb 2025 12:07:06 +0800<![CDATA[文献阅读分享：《负面情绪更吸睛？利用大型语言模型重构新闻推荐系统中的情感框架》]]>https://blog.youkuaiyun.com/HE2096798708/article/details/145401134https://blog.youkuaiyun.com/HE2096798708/article/details/145401134HE2096798708Thu, 30 Jan 2025 22:03:48 +0800