A Spiking Neural Network Based Autonomous Reinforcement Learning Model and Its Application in Decision Making

郑重声明:原文参见标题,如有侵权,请联系作者,将会撤销发布!

International Conference on Brain Inspired Cognitive Systems, 2016

Abstract

  在本文中,我们提出了一种用于决策的自主脉冲神经网络模型。该模型是具有自动环境感知功能的基底神经节电路的扩展,它根据图像输入自动构建环境状态。本文的工作有以下贡献:(1) 在我们的模型中,开发了简化的Hodgkin-Huxley计算模型以实现接近LIF模型的计算效率,用于获取和测试认识。(2) 提出了一种基于脉冲的运动感知机制,无需大量训练即可从原始像素中提取学习过程的关键元素。我们将我们的模型应用到了“appybird”游戏中,经过几十次训练,它运行良好。该模型在训练开始时获得与人类相似的学习性能。此外,我们的模型模拟了在Hodgkin-Huxley模型中阻断某些钠或钾离子通道时的认知缺陷,这是对离子水平认知的探索。

Keywords: Spiking neural network, Hodgkin-Huxley, Basal Ganglia, motion perception

1 Introduction

2 Previous Work

2.1 The Basal Ganglia Model

2.2 A Spike Coding Model of the Basal Ganglia

2.3 The Hodgkin-Huxley Model

3 Methods

3.1 The Simplified Computing Hodgkin-Huxley Model

3.2 Spike Based Motion Perception

3.3 Spiking Neural Network Based Autonomous Reinforcement Learning Model

4 Experiments and Applications

4.1 The simplified Hodgkin-Huxley Model

4.2 Autonomous Reinforcement Learning Model in the Game

A. The Motion Perception

B. Playing the Game

C. Ionic Property Test of the H-H Model

5 Conclusion

原文地址:https://www.cnblogs.com/lucifer1997/p/15352942.html