What’s happening at AAMAS 2019

Our insights into this year’s conference

back to our blogs

What’s happening at AAMAS 2019

Our insights into this year’s conference

AAMAS is an influential conference for the machine learning community, where the latest developments in multiagent systems are showcased with a strong emphasis on bringing together the theory and the practical applications in this area. This is a theme that aligns well with PROWLER.IO’s own mission to unlock the power of theory to tackle real-world problems.

The study of multi-agent systems is crucial to optimise the behaviour of many real-world systems. From the complexities of the logistics sector to getting a cab ride, such systems are all around us and involve a number of decision makers.

As a result, there are plenty of challenges in the multi-agent systems space, which involve various cross-discipline approaches in everything from game theory to distributed optimisation and probabilistic modelling.

The different events scheduled at AAMAS are investigating some of these problems. For example, I am particularly looking forward to attending the Practical Applications of Game Theory session, which has a specific emphasis on theoretical contributions that provide practical solutions for fundamental problems. I am also interested in the Industrial Applications track, which will showcase recent developments in multi-agent systems when applied to problems within industry.

What is PROWLER.IO doing at AAMAS 2019?

The team is presenting four papers, including Coordinating the Crowd: Inducing Desirable Equilibria in Non-Cooperative Systems. This paper reveals our incentive design method, which is used to maximise the overall performance in, for example, the social welfare or total profit of any system composed of self-interested agents. Using our method, we can calibrate and maximise the overall performance of crowd-based systems, which could drive optimisation across a range of industries and for millions of agents. You can read more about our work in this area here.

The Using Stochastic Games for Learning to Control and Stop Optimally in Worst-Case Scenarios paper will be presented at the ALA workshop. The paper introduces a novel framework for tackling fault-tolerance within reinforcement learning by formulating the problem as a stochastic game between a controller and an adversary that chooses when to terminate the system process. It also tackles the problem of when to stop a process (optimal stopping) under worst-case scenarios. This serves to tackle problems such as when to buy or sell in unknown systems in market scenarios.

This paper complements the work outlined in another one of our papers Fault-Tolerant Reinforcement Learning in Continuous Time, which uses a path integral control method to tackle the fault-tolerance problem in continuous-time settings. This paper will be presented at the Games, Agents and Incentives Workshop (GAIW) 2019 workshop, which Sofia Ceppi, a senior machine learning researcher at PROWLER.IO, will also chair. 

During the Optimisation in Multi-Agent Systems (OptMAS) 2019 workshop, I’ll present another one of my papers, Efficient Reinforcement Dynamic Mechanism Design. This paper investigates the problem of how to optimally set a system of incentives (transfers and payments) among agents that have unknown preferences to induce a desired outcome. The method also introduces a reinforcement learning method to achieve these optimal outcomes.

We are looking forward to attending this year’s AAMAS conference. If you’d like to find out more about our research at PROWLER.IO, please click here.

Join us to make AI that will change the world

join our team