PhD Position on Causal Effects of Communication in MARL

In human society, communication is an effective mechanism for coordinating the behaviors of humans. In the field of deep multi-agent reinforcement learning (MARL), agents can also improve the overall learning performance and achieve their objectives by communication. MARL with communication research learns to solve multi-agent tasks (such as navigation, traffic, and video games), by communicating and sharing information.

Your job
In this project, we focus on interpretability of the communication in MARL algorithms. We aim to bring together causality and multi-agent reinforcement learning, using causal methods to understand the effects of communication on MARL agents’ (learning) behaviors. This could cover better interpreting the potential causal relations between communication strategies and the learning performance, or using causal representation learning for developing more effective and interpretable MARL with communication algorithms.

As a PhD candidate, you will primarily perform research within the scope of the project culminating in a successful dissertation, as well as writing academic articles and presenting your work on international AI and machine learning conferences.

This position offers you rich development opportunities. You will be part of the Hybrid Intelligence consortium, a network of excellence of universities and institutes in the Netherlands focused on the combination of human and machine intelligence. You will then have the chance to participate in international summer schools, conferences and workshops to broaden your research skills and network. In addition, you will have the opportunity to contribute to teaching and supervising AI-related thesis projects at both the Bachelor’s and Master’s levels (10-15% of your time).

This project is funded by Hybrid Intelligence gravity project. This particular project is a collaboration between the Intelligent Systems Lab at Utrecht University and the Amsterdam Machine Learning Lab at University of Amsterdam. You will be part of the Intelligent Systems group at Utrecht University, and will work under the supervision of Dr Shihan Wang (daily supervisor), Dr Sara Magliacane (co-supervisor) and Professor Mehdi Dastani (promotor).

Requirements:

• A MSc in Computer Science or Artificial Intelligence;
• a strong background in Machine Learning;
• demonstrable experience with/knowledge of Reinforcement Learning and/or Causal Inference;
• knowledge of Multi-agent Reinforcement Learning would be a plus, but not necessary;
• a strong interest in the topic and open to communicate with others in the group and the Hybrid Intelligence consortium, as collaboration and knowledge exchange are key to the success of this PhD position.

We aim to create hybrid intelligence for everyone, see also our Diversity Statement. To do this, we need an inclusive and diverse team of researchers. We especially encourage people from underrepresented groups to apply for this job.

Salary Benefits:

• a position for 18 months, with an extension to a total of four years upon successful assessment in the first 18 months;
• a gross monthly salary between €2,901 and €3,707 in the case of full-time employment (salary scale P under the Collective Labour Agreement for Dutch Universities (CAO NU));
• 8% holiday pay and 8.3% year-end bonus;
• a pension scheme, partially paid parental leave and flexible terms of employment based on the CAO NU.

In addition to the terms of employment laid down in the CAO NU, Utrecht University has a number of schemes and facilities of its own for employees. This includes schemes facilitating professional development, leave schemes and schemes for sports and cultural activities, as well as discounts on software and other IT products. We also offer access to additional employee benefits through our Terms of Employment Options Model. In this way, we encourage our employees to continue to invest in their growth. For more information, please visit Working at Utrecht University.

36 - 40 hours per week

Princetonplein 5