# What does our group symbol represent?

In this short blog post we will shed some light on the meaning of our group symbol. The symbol is an abstract representation of a process in which two agents converge to an equilibrium point via repeated interaction. Each interaction is represented by two connected dots which represent the agents. The interactions spiral toward the centre, representing a temporal process of learning an adaptation in agent behaviours.

A concrete example of such a process in the game of Rock-Paper-Scissors can be seen in the figure below. The two agents (players) use an algorithm called WoLF-PHC [1]. Each point in the simplex represents a probability distribution over the three available actions, and the drawn lines show the temporal process of the two agents adapting to one another. The process converges to the only Nash equilibrium in the game which is for both agents to act uniformly randomly.

Designing learning algorithms which can robustly and provably converge to an optimal solution (such as Nash equilibrium) is an active area of research in artificial intelligence, machine learning, and game theory. A key problem which must be overcome by multi-agent learning algorithms is the non-stationarity caused by agents which learn and adapt concurrently. In recent years, there has been a surge of new exciting resesarch which aims to combine advances in deep learning with multi-agent reinforcement learning [2].

## Further reading

- Multi-Agent Reinforcement Learning: Foundations and Modern Approaches
- Lecture: Multi-Agent Learning I
- Lecture: Multi-Agent Learning II
- Dealing with Non-Stationarity in Multi-Agent Deep Reinforcement Learning

## References

- M. Bowling, M. Veloso. Multiagent Learning Using a Variable Learning Rate. Artificial Intelligence, Vol. 136, pp. 215-250, 2002.
- G. Papoudakis, F. Christianos, A. Rahman, S.V. Albrecht. Dealing with Non-Stationarity in Multi-Agent Deep Reinforcement Learning. arXiv:1906.04737, 2019.