Multiple Influences Maximization Under Dynamic Link Strength in Multi-Agent Systems: The Competitive and Cooperative...

Area of Science:

• Artificial Intelligence
• Network Science
• Computer Science

Background:

• Multi-agent systems (MASs) present complex challenges in understanding influence propagation.
• Dynamic link strengths significantly impact diffusion processes, requiring advanced modeling.
• Optimizing influence spread is crucial for various applications in MASs.

Purpose of the Study:

• To propose a novel model for dynamic link strength in MASs to simulate multiple influences diffusion.
• To formulate the multiple influences maximization under dynamic link strength (MIMDLS) problem considering competitive and cooperative scenarios.
• To develop a distributed deep reinforcement learning (DRL) framework for efficient seed selection in MIMDLS.

Main Methods:

• A novel dynamic link strength model for MASs to simulate multiple influences diffusion.
• Two diffusion models: competitive multiple influences independent cascade (Cp-MIIC) and cooperative multiple influences linear threshold (Cr-MILT).
• A distributed DRL framework with asynchronous training and updating for seed selection, including Q-value estimation and constraint management.

Main Results:

• Validated the effectiveness and efficiency of the proposed models and algorithms for multiple influence diffusion.
• Demonstrated superior performance of the developed distributed DRL algorithm compared to state-of-the-art methods.
• Showcased the capability of the framework in handling both competitive and cooperative influence diffusion scenarios.

Conclusions:

• The proposed dynamic link strength model and DRL framework effectively address the MIMDLS problem in MASs.
• The Cp-MIIC and Cr-MILT models provide robust mechanisms for simulating competitive and cooperative influence diffusion.
• The distributed DRL approach offers an efficient and scalable solution for optimizing influence maximization in dynamic network environments.