About

Martin Balla has completed his PhD in the Game AI group at the Queen Mary University of London as part of the Intelligent Games and Game Intelligence Doctoral Training (IGGI) programme. His research focuses on training multi-task Reinforcement Learning agents that can transfer their knowledge across different environments. His research investigates Goal-conditioned Reinforcement Learning, Successor Features, transfer learning, multi-task learning, multi-agent systems and benchmarking game-playing agents in general. Martin has published his work in the top Game AI conference (IEEE Conference on Games) and journal (IEEE Transactions on Games), as well as in the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment. Prior to starting his PhD, he studied Computer Science at the University of Essex.

Besides his PhD research, Martin has been involved in various research projects, such as:

• Benchmarking: Implemented new benchmarks to test game-playing algorithms (Pommerman, TAG, PyTAG).
• Algorithms: Designed and developed new algorithms for existing problems (MURHEA, Pommerman, TAG)
• Frameworks: Improved existing frameworks to lower the entry barriers and support the development of new algorithms (Malmo, Pommerman).
• Content generation: Created a large collection of diverse levels and game variants using quality-diversity search (i.e.: MAP-elites) for CaveSwing and Pandemic (also hosted a competition at IEEE CoG)
• Playtesting: Performed large-scale playtesting to analyse board games (Tabletop R&D).
• LLMs: Adapted open-source Large-Language Models to play games (PyTAG) using various advanced prompting techniques i.e.: Chain-of-thought, Few-shot prompting and Retrieval Augmented Generation (RAG).