Resampling Conserves Redundancy & Mediation (Approximately) Under the Jensen-Shannon Divergence
"KL散度" 相关文章
Sculpting Latent Spaces With MMD: Disentanglement With Programmable Priors
Sculpting Latent Spaces With MMD: Disentanglement With Programmable Priors
Deceptive Exploration in Multi-armed Bandits
Unlocking Reasoning Capabilities in LLMs via Reinforcement Learning Exploration
交叉熵：深度学习中最常用的损失函数
研究人员提出新型表示学习框架，填补深度学习模型缺乏因果刻画的空白
西交利物浦大学 | 针对大型语言模型的目标导向生成式提示注入攻击
SFT真不如RL？MIT团队抛出“RL的剃刀”，砍掉遗忘直通终身学习
SFT真不如RL？MIT团队抛出“RL的剃刀”，砍掉遗忘直通终身学习
SFT远不如RL？永不过时的剃刀原则打开「终身学习」大模型训练的大门
西交利物浦大学 | 针对大型语言模型的目标导向生成式提示注入攻击
Selective Generalization: Improving Capabilities While Maintaining Alignment
Off-Policy Reinforcement Learning RL with KL Divergence Yields Superior Reasoning in Large Language Models
$500 + $500 Bounty Problem: An (Approximately) Deterministic Maximal Redund Always Exists
【带你读】花书《深度学习》导读 第三章 概率与信息论 下
如何准确且可解释地评估大模型量化效果？
Beyond Accuracy: Evaluating LLM Compression with Distance Metrics
