I am a Senior Lecturer with the School of Software (Faculty of Engineering and Information Technology) at the University of Technology Sydney (UTS) where I teach core game design and games programming course as part of the Bachelor of Science in Game Development. I am also the Co-Director of the Games Studio research lab at UTS. Previously I worked as a Research Fellow on an Australian Research Council Linkage Project with Village Roadshow Limited and RMIT University. The focus of my ongoing research is in Machine Learning and Computational Intelligence and how they can be applied to games and other interactive experiences. I am highly motivated by the pursuit of knowledge and have a passion for studying state of the art machine learning techniques, general statistics, game design theory, human-computer interaction innovations, and virtual / augmented reality technologies. An Australian citizen by birth, I grew up in New Zealand, the U.S, and Singapore before returning to Australia to complete my university studies.
Post-Doctoral Research Fellow (2014 – 2017)
- Employed under an Australian Research Council (ARC) Linkage Project between Village Roadshow Ltd. and the Royal Melbourne Institute of Technology (RMIT)
This project was established during the second year of my Ph.D. candidature when I instigated and fostered a working relationship between Village Roadshow and RMIT. The project investigates novel forms of human-computer interaction that can take place in the large, congested, and public theme park space. The current focus is to improve the overall guest experience by making the queues for each attraction more enjoyable and reducing the guests’ perceived waiting time, specifically through the use of social augmented reality games.
- Establishing research objectives and setting project goals.
- Reporting and presenting to Village Roadshow executives and theme-park directors.
- Organize and conducting regular team meetings with numerous RMIT parties.
- Ensuring compliance with the Australian National Statement on Ethical Conduct in Human Research.
- Writing for publication.
- Supervising and mentoring Ph.D. candidates on the project.
Teaching & Supervising (2011 – Present)
Supervisor (2014 – Present)
- Listed on RMIT University’s Supervisors Registry having completed required training modules.
- Associate supervisor to Master and Ph.D. students.
- Primary supervisor to undergraduate final year Capstone Projects and Summer Projects.
Sessional Lecturer (2014 – 2016)
- Games Studio 2 (course code: COSC2349)
- RMIT University, Melbourne, Australia
- First year (second semester) introduction to games programming for Bachelor of Information Technology (Games and Graphics Programming) students.
- Game Mechanics and Gameplay Programming (course code: COSC2476)
- RMIT University, Melbourne, Australia
- Advanced third year game design/development course focusing on how games can benefit from artificial intelligence, machine learning, and data analytics techniques.
- Central contact and authority for all student and staff queries, grading, and course administration.
- Creation of course structure, lecture material, and assessment guides with consideration for students with disabilities.
Head Tutor (2011 – 2013)
- Game Mechanics and Gameplay Programming (course code: COSC2476)
- RMIT University, Melbourne, Australia
- I assumed this role in the second year that this course was offered. As such, I created tutorial/laboratory material, assignment specifications, and assignment marking guides.
- The assignment involved implementing gesture recognition into games using a mouse, WiiMote, or Xbox Kinect with an emphasis on using machine learning techniques.
Tutor, Lab Assistant, & Exam Supervisor (2011 – 2012)
- Java for Programmers (course code: COSC1290)
- RMIT University, Melbourne, Australia
Marketing Representative (2011 – 2015)
- Information presenter, guest speaker, and trainer for RMIT University.
- Represented RMIT University at industry seminars, open days, work experience weeks, and introductory courses.
Website Design, Development, & Administration (2010 – 2014)
- Sole consultant and developer for website design, structure, and maintenance.
- Utilizing WordPress and Magnolia content management systems for ease of continued content publishing by non-technical end-users.
- Extensive customization of existing CSS layouts and HTML/PHP page templates, SEO, social media integration, and end-user training.
- Diverse cultural experience and integration having previously resided in New Zealand, USA, Singapore and Australia
- Keen interest in exploring and mastering new technologies and processes
- Well-presented and articulate with excellent interpersonal skills
- Outstanding project management across multiple simultaneous projects
- Strong proficiency in public speaking and Business-Client core relationship management
- Exceptional attention to detail: presentation, visual and written
- Experienced in leadership, mentor and supervisor roles
- Diverse problem-solving skills with an ability to deal with the unexpected
AREAS OF EXPERTISE
- Statistical / Machine Learning
- Evolutionary Computing
- Data Analytics
- Artificial Intelligence in Games
- Games Programming
- Game Mechanics Design
- Computational Intelligence
- Recommendation Systems
- Human-Computer Interaction
Ph.D. Computer Science (2010 – 2014)
RMIT University, Australia
- Research Title: Personalised Procedural Map Generation via Evolutionary Algorithms.
- Supervisors: Dr. Fabio Zambetta and Associate Professor Xiaodong Li.
- Completed under the RMIT Ph.D. Scholarship Award.
Bachelor of Computer Science (Honours) (2009)
RMIT University, Australia
- Research Title: Virtual Asset Protection using View-Based Similarity Matching.
- Supervisors: Professor Jiankun Hu and Dr. Fabio Zambetta.
- Graduated with First Class Honours (GPA 3.9/4.0).
- Top Student in the Bachelor of Computer Science (Honours) degree.
- Awarded the Vice-Chancellor’s List Award of Top 2% of all RMIT University students.
Bachelor of Computer Science (2006 – 2008)
RMIT University, Australia
- Majoring in Games, Graphics and Digital Media.
- Graduated with Distinction.
- C# / Java
- HTML / CSS / PHP
- Matlab / R
- C / C++
- Unity Game Engine
- Unreal Engine 4
- Android SDK
- WEKA Machine Learning Framework
- Git / Mercurial Repositories
- Hadoop / Map-Reduce
- WordPress / Magnolia CMS
- SPSS / Minitab
- Golden Key International Honour Society, RMIT Chapter, member since 2006.
- Best Oral Presentation in the Science Stream of the 2011 Higher Degree by Research Student Conference, RMIT University.
- Selected Reviewer for peer-reviewed conferences on game artificial intelligence and human-computer interaction.
- Senior member and contributor of the RMIT Evolutionary Computing and Machine Learning Group, the Centre for Game Design Research (Artificial Intelligence and Game Design node), and the Exertion Games Lab.
Ph.D. Thesis: Personalized Procedural Map Generation in Games via Evolutionary Algorithms
In digital games, the map (sometimes referred to as the level) is the virtual environment in which a player navigates through and interacts with during gameplay. The map outlines the boundaries of play, aids in establishing rule systems, and supports the narrative. The map also directly influences the challenges that a player will experience and the pace of gameplay, a property that has previously been linked to a playar’s enjoyment of a game. Therefore, the entertainment value of the game is maximized by altering the challenge to fit a player’s expected ability.
In most industry leading games, creating maps is a lengthy manual process conducted by highly trained teams of designers. However, for many decades procedural content generation (PCG) techniques have been used to automate the map creation process to provide players with a larger range of experiences than would normally be possible. However, in recent years, PCG has been proposed as a means of tailoring game content to meet the preferences and skills of a specific player. These approaches fall into the recently established field of research known as Experience-driven PCG (EDPCG), in which knowledge about the player is included in a generate-and-test PCG framework.
This thesis contributes to the growing EDPCG field with a focus on personalizing maps. Here, maps are represented via two distinct concepts; geometry and content layout. The geometry of a map defines the boundaries of play and the location of static virtual objects. Meanwhile, the content layout describes the location and quantity of interactive game assets, such as enemies and pick-ups. Both of these components affect a player’s experience to varying degrees in different game genres. A third concept, player preference modeling, is added when adapting a map to a player’s preferences and is conducted by collecting, interpreting, and utilizing knowledge about the player.
In this thesis, these three subcomponents are studied as interlocking mechanisms of the personalized procedural map generation process and solutions are proposed for each. Firstly, geometry generation is conducted for both exterior and interior environments by connecting pre-made map segments to form a larger, more complete map. Player preferences are incorporated into this process via interactive evolutionary computing. The layout of content throughout the map is then determined by using features of the geometry as input to a Compositional Pattern-Producing Network, a population of which are evolved through the Neuroevolution of Augmenting Topologies. Finally, a player’s preferences for content layout are captured and utilized through the use of a player model based upon a recommender system (RS) framework.
The geometry generation process is firstly implemented into an evolutionary terrain tool that was created to aid novice game developers build the terrain of a map. All the solutions are then combined into the action-shooter game PCG: Angry Bots and evaluated through a large-scale public user experiment. The solutions are shown to perform well and can be generalized to other game genres. However, the experiment also gives rise to new questions and so this thesis concludes with a look at potential future work.LinkPDF
Honours Thesis: Virtual Asset Protection using View-Based Similarity Matching
The copyright protection of virtual assets in massively multiplayer online games that utilise user generated content is an under-appreciated research field. Current processes are manual, requiring users and business who have designed a virtual asset to lodge complaints or initiate legal action to protect their copyright. We believe that this system should be automated in order to promote population growth and creative business ideas in the virtual world. In this paper, we specifically examine the asset type of 3D models. To achieve the protection of these assets, we propose that when a new model is introduced to a system, it is checked for similarity against all other existing models in the system, reporting a copyright infringement if the similarity is too high. We approach the issue of judging similarity between 3D models by first examining the field of content driven data retrieval systems. We use a View Based method, gathering 2D projections so that we can work with pixel data rather than 3D vertex coordinate data. This allows us to investigate the use of advanced techniques from the fields of biometrics and computer vision. Primarily, we use the Correlation Filtering and Hausdorff Distance algorithms to detect similarity between two 2D projections. These algorithms perform poorly in our early experiments so we introduce a clustering structure which improves both efficiency and effectiveness. Our final system produces promising results and provides a genuine acceptance rate (GAR) of 88%. However, this is an initial step in a large body of research. Our system is not at a point where it can be automated and efficiently executed in a game environment and we finalise this paper by discussing future work towards progressing this research.LinkPDF
Peer-Reviewed Journals and Conferences:
M. Tamassia, W.L. Raffe, R. Sifa, A. Drachen, F. Zambetta, & M. Hitchens, Predicting player churn in destiny: a Hidden Markov Models approach to predicting player departure in a major online game, IEEE Conference on Computational Intelligence and Games (CIG), 2016
Destiny is, to date, the most expensive digital game ever released with a total operating budget of over half a billion US dollars. It stands as one of the main examples of AAA titles, the term used for the largest and most heavily marketed game productions in the games industry. Destiny is a blend of a shooter game and massively multi-player online game, and has attracted dozens of millions of players. As a persistent game title, predicting retention and churn in Destiny is crucial to the running operations of the game, but prediction has not been attempted for this type of game in the past. In this paper, we present a discussion of the challenge of predicting churn in Destiny, evaluate the area under curve (ROC) of behavioral features, and use Hidden Markov Models to develop a churn prediction model for the game.
M. Tamassia, F. Zambetta, W.L. Raffe, F.F. Mueller, X. Li, Dynamic choice of state abstraction in Q-learning. European Conference on Artificial Intelligence (ECAI), pp. 46-54, 2016.
Q-learning associates states and actions of a Markov Decision Process to expected future reward through online learning. In practice, however, when the state space is large and experience is still limited, the algorithm will not find a match between current state and experience unless some details describing states are ignored. On the other hand, reducing state information affects long term performance because decisions will need to be made on less informative inputs. We propose a variation of Q-learning that gradually enriches state descriptions, after enough experience is accumulated. This is coupled with an ad-hoc exploration strategy that aims at collecting key information that allows the algorithm to enrich state descriptions earlier. Experimental results obtained by applying our algorithm to the arcade game Pac-Man show that our approach significantly outperforms Q-learning during the learning process while not penalizing long-term performance.
S. Demediuk, W.L. Raffe, X. Li, An Adaptive Training Framework for Increasing Player Proficiency in Games and Simulations. ACM Symposium on Computer-Human Interaction in Play, pp. 125-131, 2016.
To improve a player’s proficiency at a particular video game, the player must be presented with an appropriate level of challenge. This level of challenge must remain relative to the player as their proficiency changes. The current fixed difficulty settings (e.g. easy, medium or hard) provide a limited range of difficulty for the player. This work aims to address this problem through developing an adaptive training framework that utilities existing work in Dynamic Difficulty Adjustment to construct an adaptive AI opponent. The framework also provides a way to measure the player’s proficiency, by analysing the level of challenge the adaptive AI opponent provides for the player. This work tests part of the proposed adaptive training framework through a pilot study that uses a real-time fighting game.
W.L. Raffe, F. Zambetta, X. Li, & K.O. Stanley, An integrated approach to personalized procedural map generation using evolutionary algorithms. IEEE Transactions on Computational Intelligence and AI in Games, vol. 7, pp. 139–155, 2015.
In this paper we propose the strategy of integrating multiple evolutionary processes for personalized procedural content generation (PCG). In this vein, we provide a concrete solution that personalizes game maps in a top-down action-shooter game to suit an individual player’s preferences. The need for personalized PCG is steadily growing as the player market diversifies, making it more difficult to design a game that will accommodate a broad range of preferences and skills. In the solution presented here, the geometry of the map and the density of content within that geometry are represented and generated in distinct evolutionary processes, with the player’s preferences being captured and utilized through a combination of interactive evolution and a player model formulated as a recommender system. All these components were implemented into a test bed game and experimented on through an unsupervised public experiment. The solution is examined against a plausible random baseline that is comparable to random map generators that have been implemented by independent game developers. Results indicate that the system as a whole is receiving better ratings, that the geometry and content evolutionary processes are exploring more of the solution space, and that the mean prediction accuracy of the player preference models is equivalent to that of existing recommender system literature. Furthermore, we discuss how each of the individual solutions can be used with other game genres and content types.
W.L. Raffe, M. Tamassia, F. Zambetta, X. Li, S.J. Pell, & F.F. Mueller, Player-computer interaction features for designing digital play experiences across six degrees of water contact. ACM Symposium on Computer-Human Interaction in Play, pp. 295-305, 2015.
Physical games involving the use of water or that are played in a water environment can be found in many cultures throughout history. However, these experiences have yet to see much benefit from advancements in digital technology. With advances in interactive technology that is waterproof, we see a great potential for digital water play. This paper provides a guide for commencing projects that aim to design and develop digital water-play experiences. A series of interaction features are provided as a result of reflecting on prior work as well as our own practice in designing playful experiences for water environments. These features are examined in terms of the effect that water has on them in relation to a taxonomy of six degrees of water contact, ranging from the player being in the vicinity of water to them being completely underwater. The intent of this paper is to prompt forward thinking in the prototype design phase of digital water-play experiences, allowing designers to learn and gain inspiration from similar past projects before development begins.
J. Ivanovo, W.L. Raffe, F. Zambetta, & X. Li, Combining Monte Carlo tree search and apprenticeship learning for capture the flag. IEEE Conference on Computational Intelligence and Games (CIG), pp. 154-161, 2015.
In this paper we introduce a novel approach to agent control in competitive video games which combines Monte Carlo Tree Search (MCTS) and Apprenticeship Learning (AL). More specifically, an opponent model created through AL is used during the expansion phase of the Upper Confidence Bounds for Trees (UCT) variant of MCTS. We show how this approach can be applied to a game of Capture the Flag (CTF), an environment which is both non-deterministic and partially observable. The performance gain of a controller utilizing an opponent model learned via AL when compared to a controller using just UCT is shown both with win/loss ratios and True Skill rankings. Additionally, we build on previous findings by providing evidence of a bias towards a particular style of play in the AI Sandbox CTF environment. We believe that the approach highlighted here can be extended to a wider range of games other than just CTF.
W.L. Raffe, M. Tamassia, F. Zambetta, X. Li, & F.F. Mueller, Enhancing theme park experiences through adaptive cyber-physical play. IEEE Conference on Computational Intelligence and Games (CIG), pp. 503-510, 2015.
In this vision paper we explore the potential for enhancing theme parks through the introduction of adaptive cyber-physical attractions. That is, some physical attraction that is controlled by a digital system, which takes participants’ actions as input and, in turn, alters the participants’ experiences. This paper is thus divided into three main parts; (1) a look at the types of attractions that a typical theme park may offer and, from this, the identification of a gap in an agency versus structure spectrum that recent research and industry developments are starting to fill; (2) a discussion of the advantages that cyber-physical play has in filling this gap and a few examples of envisioned future attractions; and (3) how such cyber-physical play can uniquely allow for adaptive attractions, whereby the physical attraction is personalized to suit the capabilities or preferences of the current attraction participants, as well as some foreseeable design considerations and challenges in doing so. Through the combination of these three parts, we hope to promote further research into augmenting theme parks with adaptive cyber-physical play attractions.
M. Tamassia, F. Zambetta, W.L. Raffe, & X. Li, Learning options for an MDP from demonstrations. Springer Australasian Conference on Artificial Life and Computational Intelligence, pp. 226-242, 2015
The options framework provides a foundation to use hierarchical actions in reinforcement learning. An agent using options, along with primitive actions, at any point in time can decide to perform a macro-action made out of many primitive actions rather than a primitive action. Such macro-actions can be hand-crafted or learned. There has been previous work on learning them by exploring the environment. Here we take a different perspective and present an approach to learn options from a set of experts demonstrations. Empirical results are also presented in a similar setting to the one used in other works in this area.
W.L. Raffe, Personalized procedural map generation in games via evolutionary algorithms. ACM SIGEVOlution Newsletter, vol 7.1, pp. 27-28, 2014
In digital games, the map (sometimes referred to as the level) is the virtual environment that outlines the boundaries of play, aids in establishing rule systems, and supports the narrative. It also directly influences the challenges that a player will experience and the pace of gameplay, a property that has previously been linked to a player’s enjoyment of a game . In most industry leading games, creating maps is a lengthy manual process conducted by highly trained teams of designers. However, for many decades procedural content generation (PCG) techniques have posed as an alternative to provide players with a larger range of experiences than would normally be possible. In recent years, PCG has even been proposed as a means of tailoring game content to meet the preferences and skills of a specific player, in what has been termed Experience-driven PCG (EDPCG) .
W.L. Raffe, F. Zambetta, & X. Li, Neuroevolution of content layout in the PCG: Angry Bots video game. IEEE Congress on Evolutionary Computation (CEC), pp. 673-680, 2013
This paper demonstrates an approach to arranging content within maps of an action-shooter game. Content here refers to any virtual entity that a player will interact with during game-play, including enemies and pick-ups. The content layout for a map is indirectly represented by a Compositional Pattern-Producing Networks (CPPN), which are evolved through the Neuroevolution of Augmenting Topologies (NEAT) algorithm. This representation is utilized within a complete procedural map generation system in the game PCG: Angry Bots. In this game, after a player has experienced a map, a recommender system is used to capture their feedback and construct a player model to evaluate future generations of CPPNs. The result is a content layout scheme that is optimized to the preferences and skill of an individual player. We provide a series of case studies that demonstrate the system as it is being used by various types of players.
W.L. Raffe, F. Zambetta, & X. Li, A survey of procedural terrain generation techniques using evolutionary algorithms. IEEE Congress on Evolutionary Computation (CEC), pp. 1-8, 2012
This paper provides a review of existing approaches to using evolutionary algorithms (EA) during procedural terrain generation (PTG) processes in video games. A reliable PTG algorithm would allow game maps to be created partially or completely autonomously, reducing the development cost of a game and providing players with more content. Specifically, the use of EA raises possibilities of more control over the terrain generation process, as well as the ability to tailor maps for individual users. In this paper we outline the prominent algorithms that use EA in terrain generation, describing their individual advantages and disadvantages. This is followed by a comparison of the core features of these approaches and an analysis of their appropriateness for generating game terrain. This survey concludes with open challenges for future research.
W.L. Raffe, F. Zambetta, & X. Li, Evolving patch-based terrains for use in video games. Proceedings of the 13th annual Conference on Genetic and Evolutionary Computation (GECCO), pp. 363-370, 2011
Procedurally generating content for video games is gaining interest as an approach to mitigate rising development costs and meet users’ expectations for a broader range of experiences. This paper explores the use of evolutionary algorithms to aid in the content generation process, especially the creation of three-dimensional terrain. We outline a prototype for the generation of in-game terrain by compiling smaller height-map patches that have been extracted from sample maps. Evolutionary algorithms are applied to this generation process by using crossover and mutation to evolve the layout of the patches. This paper demonstrates the benefits of an interactive two-level parent selection mechanism as well as how to seamlessly stitch patches of terrain together. This unique patch-based terrain model enhances control over the evolution process, allowing for terrain to be refined more intuitively to meet the user’s expectations.
W.L. Raffe, J. Hu, F. Zambetta, & K. Xi, A dual-layer clustering scheme for real-time identification of plagiarized massive multiplayer games (MMG) assets. The 5th IEEE Conference on Industrial Electronics and Applications (ICIEA), pp. 307-312, 2010
Theft of virtual assets in massive multiplayer games (MMG) is a significant issue. Conventional image based pattern and object recognition techniques are becoming more effective identifying copied objects but few results are available for effectively identifying plagiarized objects that might have been modified from the original objects especially in the real-time environment where a large sample of objects are present. In this paper we present a dual-layer clustering algorithm for efficient identification of plagiarized MMG objects in an environment with real-time conditions, modified objects and large samples of objects are present. The proposed scheme utilizes a concept of effective pixel banding for the first pass clustering and then uses Hausdorff Distance mechanism for further clustering. The experimental results demonstrate that our method drastically reduces execution time while achieving good performance of identification rate, with a genuine acceptance rate of 88%.
Website developed by William L. Raffe. Copyright 2015.