Last updated on 2018.04.16
Location(s): PenOW, Welten Institute, Open University of Netherlands (OUNL)
Period: Postdoc research (2015 - 2019)
My role within the project
I am in charge of research and development of the Adaptation and Assessment (TwoA; Nyamsuren, van der Vegt, & Westera, 2017) component that uses a real-time probabilistic algorithm to assess and match the player's skill to the game difficulty. Both skill and difficulty are quantified as intuitively understandable numerical ratings. TwoA is agnostic to the game logic and does not need a mediator to translate its output into a game logic or game events. The minimal reliance on domain expertise enables an interested stakeholder, such as game designer or instructor, to discover effective adaptations without the need of specialized domain expertise. TwoA implements an improved and expanded version of the Computerized Adaptive Practice algorithm (Klinkenberg, Straatemeier, & Van der Maas, 2011). TwoA has a less item selection bias and higher accuracy and efficiency in estimating ratings. TwoA also includes a calibration phase to increase rating convergence to true values. Finally, TwoA also offers a graph-based method for analyzing the problem space and possible learning paths (Nyamsuren, van der Maas, & Maurer, 2018). Explore more the TwoA component in the dedicated page.
I am also part of the team in charge of development of the RAGE Asset Software Architecture. We research and develop software architecture and design conditions that are needed for the easy integration and reuse of software assets providing pedagogical values in existing game platforms. Based on the component-based software engineering paradigm the RAGE architecture takes into account the portability of assets to different operating systems, different programming languages and different game engines. It avoids dependencies on external software frameworks and minimizes code that may hinder integration with game engine code. Furthermore, it relies on a limited set of standard software patterns and well-established coding practices.
Objectives of the overall RAGE project
European gaming studios, developers, and researchers will soon have access to advanced gaming technology resources and state-of-the-art knowledge to develop Applied Games easier, faster and more cost-effectively thanks to RAGE (Realising an Applied Gaming Eco-system), a Horizon 2020 research and innovation project on gamification technologies to be launched on February 1st.
The EU based industry for non-leisure games - Applied Games - is an emerging business with multiple uses in industry, education, health and the public administration sectors. As such, it is still fragmented and needs critical mass to compete globally. Nevertheless, its growth potential is widely recognized and even suggested to exceed the growth potential of the leisure games market.
To take advantage of the fruitful opportunities of this growing industry, the RAGE project will deliver a collection of self-contained gaming assets that support game studios at developing applied games more efficiently and making them better suited for their purpose. RAGE will make these assets available along with a large volume of high-quality knowledge resources through an online portal and social space that will connect research, gaming industries, intermediaries, education providers, policy makers and end-users.
RAGE will help to seize these opportunities and advance industrial leadership and innovation by delivering to Applied Games studios:
The gaming technology assets gathered along the project lifecycle will be tested and evaluated by gaming companies integrated in the RAGE consortium. These companies will be creating games that will be empirically validated in real world pilots in different application scenarios representing different markets and target groups for the Applied Games industry.
RAGE's results will generate direct impact on the competitive positioning of thousands of European SMEs in the Applied Games market. Impacts from RAGE will be visible in terms of fulfilling new client needs by quicker and more challenging methods of skills acquisition, enabling new business models based on the usage of the assets repository, contribute to secure direct skilled jobs and further job creation, and strengthening collaboration across the entire Applied Games value chain.
RAGE is a 48-months Technology and Know-How driven Research and Innovation project co-funded by EU Framework Programme for Research and Innovation, Horizon 2020. The project is co-ordinated by the Open University of The Netherlands and it includes the participation of 19 key partners from the game industry, the education sector and research centres from 10 European countries: Austria, Bulgaria, France, Germany, Italy, Portugal, Romania, Spain, United Kingdom and The Netherlands.
Location(s): University of Groningen, University of Amsterdam
Period: PhD study (2010 - 2014); Postdoc research (2014 - 2015)
The original goal of this research was to study techniques and methods for creating believable artificial opponents for computer games. However, our study based on SET revealed that ACT-R cognitive architecture is not sophisticated enough to model complex tasks. At the same time, the study clearly indicated toward two major components lacking in ACT-R. Following these results, the original goal was redefined as research into minimum components of a cognitive architecture necessary to develop plausible cognitive models for complex tasks such as computer games.
Firstly, any complex tasks that requires making a decision based on the real-time knowledge needs a reliable means of gathering information from the environment. Among five senses, visual system is the most important information source. ACT-R has a vision module as one of the core modules. Unfortunately, the module provides only a bare bone implementation of a human visual system. Guidance of a visual attention is a very complex process that has both top-down and bottom-up components. On the one hand, visual attention is guided by bottom-up inherent properties of the visual scene such as contrast-based saliency of its constituent parts. On the other hand, visual attention is also guided by top-down components such as immediate goal and a context defined by previous experience. The default vision module provides no support to bottom-up attentional guidance. It also lacks several other fundamental functionalities such as long- and short-term visual memories, definition of visual objects along several feature dimensions and imagery capability.
I have developed Pre-Attentive and Attentive Vision (PAAV) module that is an extension to ACT-R's default vision module. As the name suggests, the PAAV introduces a significant amount of pre-attentive functionality to ACT-R's visual system. PAAV is essentially an implementation of a collection of well studied theories of human vision ranging from visual memory to contrast-based saliency maps for guiding visual attention. As a module, PAAV is completely task-independent. It was designed to be able to handle not only tasks commonly used to test computational models of human vision, but also more general and complex tasks such as SET.
Secondly, any problem-solving task requires some degree of reasoning. It can be any form of reasoning: reasoning by analogy, reasoning based on rules or simple associations. Individual steps in the process of reasoning can be tied to specific task context, but our general ability to reason is a fundamental process independent of any specific task. For example, reasoning based on analogy is perceived to be fundamental to human cognition. Deontic reasoning is probably innate. Finally, infants' knowledge acquisition is largely dependent on innate concepts and principles.
In terms of ACT-R, there should be task-independent general set of production rules that provide schematic rules for reasoning based on a given context. For example, rules in SET dictate that if two candidate cards are green and blue then the third card should be red in order to form a valid set. This rule can be written as (Blue, Green) => (Red). This is a task-specific rule. However, ability to reason based on this rule should be task general. ACT-R should have a task-general knowledge of conjunction of concepts. It also should have knowledge that conjunction of certain concepts can imply another concept.
Human Reasoning Module developed by me introduces a basic set of declarative and procedural knowledge to ACT-R that allows it to reason based on task-specific instructions. HRM introduces an explicit notion of a concept. It also knows that individual concepts can be combined to form more complex statements. In turn, those statements can be combined into declarative rules that can be used to encode task-specific instructions. Concepts, statements and rules based on statements form the declarative part of HRM knowledge. HRM's procedural knowledge is represented by a set of task-general production rules that describe how declarative and other forms of knowledge can be used for reasoning. Ideally, if HRM is used and a proper set of instructions about the task is given in the declarative memory, the modeler will have to write only task-specific production rules responsible mostly for meta-control. One of the features that set apart HRM from the traditional view of human mental logic is that reasoning in HRM is not purely top-down. Facts and evidences necessary for reasoning can be extracted on the fly from the information sources (such as visual memory) other than declarative memory.