thompson的博客

Reinforcement Learning

强化学习的数学基础（2）

参考书籍:1.Reinforcement Learning State-of-the-Art, Marco Wiering2.Reinforcement learning: An introduction, RS SuttonSuppose that there is an interactive system as follows which consists of an environment and an agent.Its running process is described as fo

假设对随机变量ξ，存在两个概率分布PQ。如果ξ为离散随机变量，定义从P到QDKL​P∣∣Qi∑​PilnQiPi​如果ξ为连续随机变量，定义从P到QDKL​P∣∣Q∫−∞∞​pxlnqxpx​dx。

Adaptive Value Decomposition with Greedy Marginal Contribution Computation for Cooperative Multi-Agent Reinforcement Learning

coach agent 能够观测到global state, 并可以根据global state产生信息发送给agent, 信息是定长的, 文中称为strategy, 是一个vector, 记作。. agent采取策略的时候将会参考coach agent发送的strategy信息. coach agent并不是每个step都给agent发送信息, 而是每。的目标. 此目标函数第1项表示: 完成这项任务的agent中能力最大的agent的评分, 第2项表示: 1-被派去的agent的数量.

摘自 Ding, R., Xu, Y., Gao, F., & Shen, X.(. (2022). Trajectory Design and Access Control for Air–Ground Coordinated Communications System With Multiagent Deep Reinforcement Learning. IEEE Internet of Things Journal, 9, 5785-5798.摘自 Zhang, W., Wang, Q., Liu

每隔一段取一个均值，然后把均值曲线绘制出来，包含全部点的曲线淡化处理摘自 Z. Mou, Y. Zhang, F. Gao, H. Wang, T. Zhang and Z. Han, “Deep Reinforcement Learning based Three-Dimensional Area Coverage with UAV Swarm,” in IEEE Journal on Selected Areas in Communications, doi: 10.1109/JSAC.2021.3.

强化学习重点文献总结

完整克隆整个代码包记下地址，例如 C:\Users\ABC\Downloads\multiagent-particle-envs-master打开 Anaconda Promptactivate + 你创建的环境的名称输入pip install -e C:\Users\ABC\Downloads\multiagent-particle-envs-master安装完成参考博客......

Policy Gradient Algorithms连续动作空间我们可以把待求的π∗(s)\pi^{*}(s)π∗(s)看作是一个stochastic policy:π∗(a∣s)={1a=π∗(s)0a≠π∗(s)\pi^{*}(a|s)=\left\{\begin{array}{rcl}1 & & {a=\pi^{*}(s)}\\0 & & {a\neq \pi^{*}(s)}\end{array} \right.π∗(a∣s)={10​​a=π∗(s)