Search Results

< Back Dr Michelangelo Conserva Queen Mary University of London iGGi Alum Michelangelo Conserva is a second year PhD researcher studying principled exploration strategies in reinforcement learning. He is particularly interested in randomized exploration and, more generally, Bayesian methods for reinforcement learning. He holds a BSc in Statistics, Economics and Finance from Sapienza, University of Rome and an MSc in Computational Statistics and Machine learning from University College of London. A description of Michelangelo's research: As a PhD student at Queen Mary University of London, Michelangelo aims to leverage Bayesian models to develop principled algorithms for reinforcement learning in the context of function approximations. The main challenge lies in finding a balance between computational costs and optimality. Evaluating such balance requires careful evaluation, which is currently lacking in reinforcement learning. Email m.conserva@qmul.ac.uk Website LinkedIn Mastodon BlueSky GitHub Other Link Supervisors: Prof. Simon Lucas Dr Paulo Rauber Featured Publication(s): Exploration with Foundation Models: Capabilities, Limitations, and Hybrid Approaches ForestCast: Forecasting Deforestation Risk at Scale with Deep Learning Foundation Models as World Models: A Foundational Study in Text-Based GridWorlds Heterogeneous graph neural networks for species distribution modeling Mapping Farmed Landscapes from Remote Sensing On the Limits of Tabular Hardness Metrics for Deep RL: A Study with the Pharos Benchmark What are you looking at? Team fight prediction through player camera Posterior Sampling for Deep Reinforcement Learning Hardness in Markov Decision Processes: Theory and Practice Recurrent Neural-Linear Posterior Sampling for Nonstationary Contextual Bandits The Graph Cut Kernel for Ranked Data Themes Game AI - Previous Next

< Back Alex Fletcher Queen Mary University of London iGGi Alum Alex Fletcher is a freelance audio engineer and junior game developer working on understanding the perceived flow and player experiences in mobile rhythm games and how a dynamic difficulty adjustment system would improve these experiences. The function of EEG and other biosensors as an additional measurement of player experience is of particular interest as further research in its use as an adaptive system. Other areas of research interest include game-based learning and games with a purpose. Please note: Updating of profile text in progress Email Website LinkedIn Mastodon BlueSky GitHub Other Link Themes Applied Games Game Audio Player Research - Previous Next

< Back Prof. Peter Cowling Queen Mary University of London iGGi Director Supervisor Peter Cowling has led teams that have won £45 million for research into games and digital creativity. After decades of experience in novel models and algorithms for AI decision-making, his research is now targeted on finding and promoting promising research directions in AI, games and digital creative technology, to benefit people and wider society. Playful ideas, curiosity and games have a central role! As Principal Investigator, he led the teams which won the grants for IGGI (2014 and 2019) and Digital Creativity Labs (2015). He is a member of the Programme Advisory Board which informs strategy in the Digital Economy area of UK research council funding. He has sat on several research council grant funding prioritisation panels, chairing two. He has presented ideas for the use of games as a tool to influence and understand the human condition at a number of venues, including TEDx and 10 Downing Street. He has published over 100 papers, winning 2 best paper awards at AIIDE. His research technology has over 5 million installs in commercial games – he was invited to talk at GDC about that. He would be interested to supervise students whose research uses games as a tool to gather opinion or promote understanding: to identify research directions and harness the future potential of games, creativity and AI to benefit people and society. He is particularly interested in how games and other curious, creative things can help us to understand a world of complex interacting agents, each living a world created by their own thought (!). Research themes: Research visions for games and AI Game design/development to influence, inform and understand people and society Game AI Email peter.cowling@qmul.ac.uk Website LinkedIn Mastodon BlueSky GitHub Other Link Themes Applied Games Design & Development Game AI - Previous Next

< Back Andrew Martin Queen Mary University of London iGGi Alum Applications in game development for programming language theory and AI Modern game development is highly iterative. Iteration is usually discussed in terms of a team completing design iterations, but can also be considered at the level of an individual developer attempting to complete a task, or experimenting with some ideas. At this level, the feedback loop provided by the tool becomes critical. Programming environments in particular often have a very poor feedback loop. Programming feedback can be thought of in terms of how quickly and seamlessly the user is able to observe the results of their work. This process is usually plagued with manual tasks and long pauses. It is common that a user will need to recompile, relaunch their program, and then manually recreate whatever state is required to observe the behaviour that they are working on. Frameworks like Elm, React and Vuejs are establishing a new norm of automatic hot-reloading with state preservation. These systems represent a branch of programming language research that is strongly focused on developer experience. In order to improve upon this work for game development, we must overcome the unique challenges that game development entails. Although the systems mentioned are all quite recent, there is a rich vein of research to draw on, which can be traced through dataflow programming, Smalltalk, Erlang, functional-reactive programming, Lisp and more. Predictive completions are considered by many to be a natural next-step in the evolution of live programming environments. An AI programming assistant would propose program fragments as completions or alternatives. The agent may seek to anticipate the user’s intent, or to provide creative suggestions. There is much relevant research in the fields of program synthesis, inductive logic programming, machine learning and genetic programming. One significant problem is how to smoothly and safely integrate a system like this into the user’s workflow. Many of the properties useful for safely enabling live programming features, such as isolation of side-effects, will also permit an AI agent to safely generate and execute code. Andy graduated from Imperial College London with an MEng in Computing in 2011. Following this he worked on game engine tools and technology at a startup called Fen Research, and then as a senior developer at a software consulting firm called LShift. In 2016 he spent six months working as a Research Associate in the Computational Creativity group at Goldsmiths, before starting his PhD. Please note: Updating of profile text in progress Email Website LinkedIn Mastodon BlueSky GitHub Other Link Themes Game AI - Previous Next

< Back Prof. Greg Slabaugh Queen Mary University of London Supervisor Gregory G. Slabaugh is Professor of Computer Vision and AI and Director of the Digital Environment Research Institute (DERI) at Queen Mary University of London. He is also a Turing Fellow at the Alan Turing Institute. His research work spans computer vision and computer graphics including geometric modelling and image/video-based understanding. He is interested in deep learning approaches including generative techniques like normalizing flow an generative adversarial networks. He previously worked in the games industry as a 3D graphics programmer and his PhD thesis focussed on how to model 3D objects from a collection of images. He is interested in how to create engaging content and interaction from images as well as procedural methods to reduce the effort of 3D modelling. Email g.slabaugh@qmul.ac.uk Website LinkedIn Mastodon BlueSky GitHub Other Link Themes Applied Games Creative Computing Immersive Technology - Previous Next

< Back Dr Ivan Bravi Queen Mary University of London iGGi Alum Ivan Bravi has obtained his B.Sc and M.Sc in Engineering of Computer Systems at the Politecnico di Milano, Italy. From January to July 2016 he was Visiting Scholar at the NYU’s Game Innovation Lab in New York, under the supervision of Prof. Julian Togelius. Since October 2017 he's an IGGI PhD student at Queen Mary University of London under the supervision of Simon Lucas. Ivan has published several workshop and conference papers in different venues such as IJCAI, Evostar, CIG, FDG, AAAI and CoG. Automatic playtesting of games can significantly streamline the process of designing, developing and releasing a game. It is also a possible application of Artificial General Intelligence (AGI): having a set of flexible algorithms that can play games regardless of their type decouples the two problems (playtesting and developing AGI algorithms) advancing both independently. When it comes to developing new AGI algorithms for game-playing a crucial characteristic is the ability of expressing different behaviours. Most of the research has focused on peak performance game-playing agents, this research project instead focuses on producing agents that are able to show different playing styles (behaviours) with no explicit domain information embedded in the algorithm. Behavioural expressivity arises from the parameterisable components of an algorithm. In classical Statistical Forward Planning (SFP) it is very straightforward to adjust these, e.g. how far ahead it's planning. A very important component of SFP algorithms is the heuristic function used to evaluate the quality of game states. Being able to define heuristics in a game-agnostic manner is a key element in maintaining the algorithms generally. Email i.bravi@qmul.ac.uk Website LinkedIn Mastodon BlueSky GitHub Other Link Supervisor(s): Dr Diego Pérez-Liébana Prof. Simon Lucas Featured Publication(s): Program Committee and Subreviewers Evaluating and Enhancing Gameplay Behavioural Expressivity of Planning-Playing Artificial Intelligence for Automatic Playtesting Self-adaptive MCTS for General Video Game Playing Rinascimento: Playing Splendor-Like Games With Event-Value Functions Rinascimento: searching the behaviour space of Splendor Rinascimento: using event-value functions for playing Splendor Learning local forward models on unforgiving games Rinascimento: Optimising statistical forward planning agents for playing splendor A local approach to forward model learning: Results on the game of life game Game AI hyperparameter tuning in rinascimento Efficient evolutionary methods for game agent optimisation: Model-based is best Shallow decision-making analysis in general video game playing Evolving UCT alternatives for general video game playing Evolving game-specific UCB alternatives for general video game playing Themes Game AI Player Research - Previous Next

< Back Dr Ozan Vardal University of York iGGi Alum Ozan studied undergraduate psychology at the University of Groningen, and holds a master's degree in Performance Psychology from the University of Edinburgh, where he wrote theses on the dynamics of psychological momentum in sport competition and the decision-making of expert applied psychologists respectively. He has long been fascinated with the psychological mechanisms underpinning complex skills, owing to his own background as a classically trained musician and his previous work as a performance psychology consultant with competitive athletes. His primary research interests involve the behavioural and neural factors surrounding human learning and skilled performance. A description of Ozan's research: Ozan views games as behaviourally rich environments for the study of complex skills and human learning. The competitive and immersive nature of games encourages millions of players to develop profound skill over hours, days, and even years of practice. Ozan’s work takes advantage of large data repositories generated by such players to study how different patterns of practice result in differences in learning outcomes. He also uses experimental methods in his work, and is currently using neuroimaging methods (MEG) and modelling techniques to identify how shifts between different behavioural and neural states affect performance as people play Tetris. By using games as a vehicle to study psychology, Ozan aims to develop scalable solutions to studying human learning. He hopes for a future where the science of learning is sufficiently advanced, such that (artificial) trainers can recommend optimised practice schedules for motivated learners, in any performance domain. Please note: Updating of profile text in progress Email ov525@york.ac.uk Website LinkedIn Mastodon BlueSky GitHub Other Link Featured Publication(s): Mind the gap: Distributed practice enhances performance in a MOBA game Themes Design & Development Esports Game Data - Previous Next

< Back Susanne Binder Queen Mary University of London iGGi Manager iGGi Admin iGGi Manager @ QMUL ; alongside David Hull (iGGi Manager @ UoY) , and supported by Shopna Begum , Helen Tilbrook and Oliver Roughton, she's mostly in charge of making things run at iGGi with particular focus on iGGi-QMUL-specific admin iGGi-QMUL-specific student concerns PR, website and social media industry liaison Email s.binder@qmul.ac.uk Website LinkedIn Mastodon BlueSky GitHub Other Link Themes - Previous Next

< Back Dr Sebastian Berns Queen Mary University of London iGGi Alum Sebastian is a designer and researcher working on use-inspired fundamental research in generative machine learning for creative and artistic applications. Sebastian holds a master’s degree in artificial intelligence and has a background in visual communications. He has worked several years as an independent graphic and type designer with a specialisation in web development. His design work has been awarded national and international design prizes. A description of Sebastian's research: "Generative machine learning methods are trained on raw data, modelling the primary patterns that constitute typical examples. They enable the production of high-quality artefacts in very complex domains and provide useful models for generative systems, in particular in the visual arts and video games. However, modelling a training data distribution perfectly is less valuable for applications in art production and video games. In particular, our analysis of the use of generative models in visual art practices motivates the need to increase the output diversity of generative models. In my dissertation, I focus on diversity in generative machine learning for visual arts and video games. Our findings benefit the application of generative models in generative systems, quality diversity search, art production and video games. Rather than a ‘ground truth’ that needs to be modelled perfectly, we argue that training datasets are merely a limited snapshot of a complex world with inherent biases. To be useful for applications in visual arts and video games, generative models require higher output diversity. Relatedly, higher generative diversity benefits efforts of equity, diversity and inclusion by reducing harmful biases in generative models." Email s.berns@qmul.ac.uk Website LinkedIn Mastodon BlueSky GitHub Other Link Featured Publication(s): Not All the Same: Understanding and Informing Similarity Estimation in Tile-Based Video Games Towards Mode Balancing of Generative Models via Diversity Weights Increasing the Diversity of Deep Generative Models Automating Generative Deep Learning for Artistic Purposes: Challenges and Opportunities Expressivity of Parameterized and Data-driven Representations in Quality Diversity Search First experiments in the automatic generation of pseudo-profound pseudo-bullshit image titles Generative Search Engines: Initial Experiments Adapting and Enhancing Evolutionary Art for Casual Creation. Creativity Theatre for Demonstrable Computational Creativity Bridging Generative Deep Learning and Computational Creativity NEST 2.18. 0 Active Divergence with Generative Deep Learning--A Survey and Taxonomy Themes Creative Computing - Previous Next

< Back Connor Watts Queen Mary University of London iGGi PG Researcher I am a machine learning research engineer and software developer with commercial experience deploying and maintaining models for start-ups and larger organizations. I have experience researching and developing novel algorithms, as well as designing custom environments for application in domains such as combinatorial optimization, finance and games. Email c.watts@qmul.ac.uk Website LinkedIn Mastodon BlueSky GitHub Other Link Supervisor: Dr Paulo Rauber Themes Game AI - Previous Next

< Back Dr Yizhao Jin Queen Mary University of London iGGi Alum Currently a student at Queen Mary University of London (QMUL), I have delved deep into the realms of artificial intelligence and game design. With a passion for understanding the complexities behind real-time strategy (RTS) games and their dynamic, unpredictable nature, I have committed myself to contribute novel insights to this domain. Research: My primary research area is Hierarchical Reinforcement Learning (HRL) for Real-Time Strategy (RTS) games. RTS games, known for their intricate mechanics and vast decision spaces, present a formidable challenge for traditional AI approaches. By employing HRL, I aim to develop agents that can not only understand the multi-layered tactics and strategies of these games but also learn to adapt to ever-changing game scenarios efficiently. The main objectives of my research are: Better Generalization: To create agents that can seamlessly transition between different RTS games or various maps within the same game without extensive retraining. This involves understanding common strategic threads across multiple game domains. Efficient Training: RTS games are inherently time-consuming due to their vast decision spaces and prolonged gameplay. My research seeks ways to optimize the training process, ensuring that AI agents can learn faster and with fewer computational resources. Email acw596@qmul.ac.uk Website LinkedIn Mastodon BlueSky GitHub Other Link Supervisors: Prof. Greg Slabaugh Prof. Simon Lucas Themes Game AI Previous Next

< Back - Meet me @ Develop:Brighton 2026 - George Long Queen Mary University of London iGGi PG Researcher Available for placement George is an iGGi PhD student interested in AI assisted game design, specifically the process of game balancing. After graduating with a BSc in Computer Science at the University of Essex, he joined iGGi in 2021. His research focuses on how AI can be used to balance games, in particular games with asymmetric elements. He has most recently been working on using agentic Large Language Models to theorycraft optimal strategies in games. A description of George's research: "My research focuses on balancing games through identifying overpowered or unfair rules, which can then be accounted for: My previous work identified how we can estimate unit point costs in miniature wargames. These wargames are played by forming armies comprised of units, the sum of each unit's point costs must be within the total army budget. Therefore it is crucial that these point costs accurately represent how useful the unit is in combat. We used Linear Regression and Linear Programming to estimate these point costs and improve the balance of an example game. Currently, I am investigating how we can use Large Language Models to theorycraft optimal strategies in games.I used an agentic approach where an agent comes up with an initial theory, tests in out in games, and uses the results of the game to create an iterative feedback loop to improve the strategy." Email g.e.m.long@qmul.ac.uk Website LinkedIn Mastodon BlueSky GitHub Other Link Supervisor(s): Dr Diego Pérez-Liébana Featured Publication(s): PyTAG: Challenges and Opportunities for Reinforcement Learning in Tabletop Games Themes Design & Development Game AI Game Data - Previous Next