#curriculum learning

8 stories · 1 in 7 days · 1 in 30 days

reinforcement learning 2 AI safety 1 GRPO 1 LLM 1 LTL 1 PAC 1 RSIAgent 1 TBTCL 1

Stories, products and papers tagged "curriculum learning", newest first.

Nov 11

Papers Builders11:43

Chinese contentTBTCL 框架用 Temporal Behavior Trees 统一强化学习的规格、奖励与课程设计

用行为树一个东西同时搞定任务描述、稠密奖励、课程和策略诊断，不用再在 LTL 和奖励函数之间来回翻译。做机器人 RL 或形式化方法的可以看看论文里的课程 DAG 提取算法，6 个规格的实验覆盖了离散和连续环境。

#TBTCL #Temporal Behavior Trees #reinforcement learning #LTL

arXiv cs.LG@Vladimir Krsmanovic, Bernd FinkbeinerSource

Sep 25

Papers BuildersSelected72°23:05

RSIAgent curriculum training lifts agent scores to 74.54%

The wide-then-deep practice order, with a verifier gating what gets saved, is the lever — the 18-point gap suggests task sequencing matters more than adding training data. Builders deploying agents in novel apps can replicate the curriculum-actor-verifier loop without collecting new datasets. It compares directly against the costly collect-and-retrain route that most teams default to today.

#RSIAgent #agents #curriculum learning #AI safety

rohanpaul_ai@rohanpaul_aiSource

Sep 1

PapersSelected73°10:44

PAC Curriculum for LLM Reinforcement Learning

PAC improves LLM training efficiency by 15-30% over existing methods, reducing computational costs for researchers in ASEAN and China.

#PAC #LLM #reinforcement learning #curriculum learning

arXiv cs.LG@Yuanqiang Yu et al. (11)Source

Aug 11

Chinese content预测安全滤波器增强的课程学习车辆动力学控制

这论文用安全滤波器和课程学习搞车辆控制，在CarSim上验证了，比传统方法省调参还更安全。

#CarSim #curriculum learning

arXiv cs.AI@Baocong Zhang et al. (3)Source

Jul 28

Chinese contentO^2-CritiCuRL：离线-在线课程强化学习提升多模态推理

这篇论文提出了一种新方法，能解决多模态模型推理时中间步骤乱但结果对的问题，效果SOTA还更高效，做推理增强的可以看看。

#O^2-CritiCuRL #multimodal reasoning #reinforcement learning #curriculum learning

arXiv cs.AI@Wendi Deng et al. (14)Source

Jun 19

Chinese content模型驱动方法开发强化学习环境家族

这篇论文提出用模型驱动和遗传算法自动生成RL环境变体，省去手写大量相似环境的麻烦，野火场景验证过，做课程学习的可以看看。

#reinforcement learning #curriculum learning #genetic algorithms

arXiv cs.LG@Xiaoran Liu, Istvan DavidSource

Jun 10

Papers72°11:30

Chinese contentRoboNaldo：运动引导课程强化学习实现人形机器人精准射门

做足式机器人运动控制或强化学习的团队，RoboNaldo 的课程学习思路能直接借鉴——用单条参考动作引导复杂技能学习，解决了从仿真到真实部署的精度和速度难题，值得点开看方法细节。

#humanoid robots #reinforcement learning #curriculum learning #motion control

arXiv cs.AI@Yichao Zhong et al. (12)Source

May 13

PapersSelected19:12

Chinese contentFG-ExPO：自适应KL与高斯课程采样提升GRPO数学推理

做LLM数学推理RL训练的团队，GRPO的KL系数和采样策略可以照搬这个改进，AIME 2025上13个点的提升值得一试。

#GRPO #reinforcement learning #math reasoning #curriculum learning

arXiv: DeepSeek@Mingxiong Lin et al. (9)Source