Introduction

Source code for the paper entitled "Joint Mean-Field Game and Multiagent Asynchronous Advantage Actor-Critic for Edge Intelligence-based IoT Malware Propagation Defense" submitted to IEEE Transactions on Dependable and Secure Computing for possible publication.

Requirements

• Linux
• Python 3.6+
• gym
• TensorFlow 1.x

Reference：https://github.com/geek-ai/MAgent

Environment

Comparative metrics：Average rewards, Successful defense numbers, Total rewards

64 EIoT nodes vs 64 Malware attackers (Figs. 4, 7, 12)

Comparison Algorithms : A2C, MFGA2C, A3C, MFGA3C

• A2C :

  python train_battle.py --algo ac

• MFGA2C :

  python train_battle.py --algo mfac

• A3C :

  python ./MFA3C/a3c.py

• MFGA3C :

  python ./MFA3C/mfa3c.py

Learning rate (Figs. 10, 11, 15)

Modify the learning_rate pass parameter of the A3C_Network class in the mfa3c.py file

• MFGA3C :

  python ./MFA3C/mfa3c.py

81 EIoT nodes vs 81 Malware attackers (Figs. 5, 8, 13)

Comparison Algorithms : A2C, MFGA2C, A3C, MFGA3C

• A2C :

  python train_battle.py --algo ac --mapsize 45

• MFGA2C :

  python train_battle.py --algo mfac --mapsize 45

• A3C :

  python ./MFA3C/a3c.py --mapsize 45

• MFGA3C :

  python ./MFA3C/mfa3c.py --mapsize 45

No mini-map environment (Figs. 6, 9, 14)

1. cd to the magent environment configuration directory

   cd / User Home Directory /anaconda3/envs/ User env name /lib/python3.6/site-packages/magent/builtin/config/

2. Modify battle.py file

   vim battle.py

3. Change the boolean value of minimap_mode to False

Comparison Algorithms : A2C, MFGA2C, A3C, MFGA3C

• A2C :

  python train_battle.py --algo ac

• MFGA2C :

  python train_battle.py --algo mfac

• A3C :

  python ./MFA3C/a3c.py

• MFGA3C :

  python ./MFA3C/mfa3c.py