A vrain-inspired spiking neural network improves multi-agent cooperation and competition

Theory of Mind (ToM) refers to the remarkable ability humans possess to understand and infer the mental states of others, such as their beliefs, intentions, and desires. This high-level social cognitive ability has intrigued researchers for years.

Recent advancements have shed light on the neural mechanisms underlying ToM, providing valuable insights into how we study and explore social interactions in multi-agent systems and human-computer interactions based on ToM.

Excitingly, a groundbreaking research project led by Prof. Zeng Yi from the Institute of Automation of the Chinese Academy of Sciences has introduced a spiking neural network called MAToM-SNN. This network is inspired by the multi-agent theory of mind and aims to enhance multi-agent cooperation and competition.

The MAToM-SNN is composed of two vital modules: the Self-MAToM and Other-MAToM. The former infers the behaviors of others based on the agent’s own experiences, while the latter does so using historical observations of others.

Prof. Zeng Yi, the study’s corresponding author, highlighted that the predicted behaviors of others by MAToM-SNN offer rich state representations for the decision-making model, enabling the decision network to adaptively adjust its policies.

Through the implementation of MAToM-SNN, agents can effectively use their own experiences or observations of others to infer behaviors and, consequently, adjust their policies for better interactions with their peers. Notably, MAToM-SNN significantly enhances the performance of multi-agent systems in cooperative and competitive tasks.

Zhao Zhuoya, the first author of the study, emphasized that MAToM-SNN exhibits impressive generalization capabilities in multi-agent reinforcement learning tasks, utilizing spiking neural networks and recurrent neural networks.

Furthermore, the researchers discovered that the Self-MAToM module aids the rapid learning of Other-MAToM. This finding suggests that self-awareness is crucial for inferring the mental states of others, especially when information about them is limited or incomplete.

In conclusion, the MAToM-SNN represents a significant step forward in developing AI systems with Theory of Mind/Cognitive Empathy. By emulating the brain’s functioning, this model ensures that AI can comprehend and predict human behaviors more accurately and responsibly, making it a trustworthy and valuable tool for society.

Source: Chinese Academy of Sciences

Leave a Reply

Your email address will not be published. Required fields are marked *