Jong-Won Yoon

Optimal strategy selection of non-player character on real time strategy game using a speciated evolutionary algorithm

In the real-time strategy game, success of AI depends on consecutive and effective decision making on actions by NPCs in the game. In this regard, there have been many researchers to find the optimized choice. This paper confirms the improvement of NPC performance in a real-time strategy game by using the speciated evolutionary algorithm for such decision making on actions, which has been largely applied to the classification problems. Creation and selection of members to use for this ensemble method is manifested through speciation and the performance is verified through ‘conqueror’, a real-time strategy game platform developed by our previous work.

An intelligent synthetic character for smartphone with Bayesian networks and behavior selection networks

As cell phones have become more common, personalized intelligent services in smartphones have become more highly desired. The mobile intelligent synthetic character is an example of one of these desired services. It is hard to apply an intelligent synthetic character to the smartphone environment because of its dynamism and complexity. This paper proposes a method for generating behaviors of a smart synthetic character that infers user contexts with Bayesian networks. In order to generate more realistic behaviors, the OCC model is utilized to create the characters emotion. Behaviors are produced through large-scale modular behavior networks with inferred contexts. A working progress is the mobile log collected with a Samsung SPH-M4650 smartphone that is used to verify the naturalness and flexibility of the generated behaviors.

Adaptive mixture-of-experts models for data glove interface with multiple users

Hand gestures have great potential to act as a computer interface in the entertainment environment. However, there are two major problems when implementing the hand gesture-based interface for multiple users, the complexity problem and the personalization problem. In order to solve these problems and implement multi-user data glove interface successfully, we propose an adaptive mixture-of-experts model for data-glove based hand gesture recognition models which can solve both the problems. The proposed model consists of the mixture-of-experts used to recognize the gestures of an individual user, and a teacher network trained with the gesture data from multiple users. The mixture-of-experts model is trained with an expectation-maximization (EM) algorithm and an on-line learning rule. The model parameters are adjusted based on the feedback received from the real-time recognition of the teacher network. The model is applied to a musical performance game with the data glove (5DT Inc.) as a practical example. Comparison experiments using several representative classifiers showed both outstanding performance and adaptability of the proposed method. Usability assessment completed by the users while playing the musical performance game revealed the usefulness of the data glove interface system with the proposed method.

Enhanced user immersive experience with a virtual reality based FPS game interface

The immersive game interfaces which reflect the users actions to the game in real-time can make the user be immersed in the game. In this paper, we applied the virtual reality (VR) techniques to a first-person shooting (FPS) game, Unreal Tournament 2004, for the enhanced user immersion with making the range of sensible information be widen by using the head-mounted display (HMD) and the 5.1 channel headphone, and also making the interaction between the user and the game system be natural with the head tracker and the data gloves. Additionally, the intelligent game agents are applied to improve the quality of the game contents. The proposed immersive interface was substituted for the traditional interface system represented by a monitor, a keyboard and a mouse. To evaluate the proposed system, we performed the usability test with both the physiological and the psychological measures. Through the experiment, we verified that the proposed system provided better immersive experience to users than the traditional system.

An efficient genetic algorithm with fuzzy c-means clustering for traveling salesman problem

Genetic algorithms (GA) are one of effective approaches to solve the traveling salesman problem (TSP). When applying GA to the TSP, it is necessary to use a large number of individuals in order to increase the chance of finding optimal solutions. However, this incurs high evaluation costs which make it difficult to obtain fitness values of all the individuals. To overcome this limitation we propose an efficient genetic algorithm based on fuzzy clustering which reduces evaluation costs with minimizing loss of performance. It works by evaluating only one representative individual for each cluster of a given population, and estimating the fitness values of the others from the representatives indirectly. A fuzzy c-means algorithm is used for grouping the individuals and the fitness of each individual is estimated according to membership values. The experiments were conducted with randomly generated cities, and the performance of the method was evaluated by comparing to other GAs. The results showed the usefulness of the proposed method on the TSP.

Enhancing hand gesture recognition using fuzzy clustering-based mixture-of-experts model

Hand gestures have been widely applied to interface as the way of interaction between human and computers. Since a human hand can express various shapes of gestures, previous models for recognizing them cannot distinguish them accurately since they use only single model for recognition. For efficient hand gesture recognition with its enhanced performance, we propose the fuzzy c-means clustering based mixture-of-experts (FME). The proposed method uses multiple local experts obtained via fuzzy c-means clustering and decisions from them are combined with the gating network. To evaluate the performance of the proposed method, we conduct experiments including comparisons with alternative models for hand gesture recognition. As the result of experiments, the proposed model shows improved gesture recognition performance, especially performance on similar hand gesture recognition.

Gesture based dialogue management using behavior network for flexibility of human robot interaction

The usage of robots becomes more sophisticated, direct communication by means of human language is required to increase the efficiency of their performance. However, the dialogue systems that reply to the user with a set of predefined answers tend to be static. In this paper, we propose a gesture based dialogue system using behavior network for flexibility of human robot interaction. Gestures take an important part of interactions. By using gestures in dialogues, it could support a flexible and realistic interaction with humans. We confirm the usability of gestures through several scenarios and SUS subject test.

Global/local hybrid learning of mixture-of-experts from labeled and unlabeled data

The mixture-of-experts (ME) models can be useful to solve complicated classification problems in real world. However, in order to train the ME model with not only labeled data but also unlabeled data which are easier to come, a new learning algorithm that considers characteristics of the ME model is required. We proposed global-local co-training (GLCT), the hybrid training method of the ME model training method for supervised learning (SL) and the co-training, which trains the ME model in semi-supervised learning (SSL) manner. GLCT uses a global model and a local model together since using the local model only shows low accuracy due to lack of labeled training data. The models enlarge the labeled data set from the unlabeled one and are trained from it by supplementing each other. To evaluate the method, we performed experiments using benchmark data sets from UCI machine learning repository. As the result, GLCT confirmed the feasibility of itself. Moreover, a comparison experiments to show the excellences of GLCT showed better performance than the other alternative method.

Fitness approximation for genetic algorithm using combination of approximation model and fuzzy clustering technique

A genetic algorithm can be applied to various search or optimization problems. However, there exists a problem that it takes too much cost to evaluate a large number of individuals. To deal with the problem, the fitness approximation method which reduces the cost of the evaluation with the similar performance to the general GA is needed. We proposed the fitness approximation using a combination of the approximation model and the fuzzy clustering technique. There exist two advantages of the proposed method. First, it reduces the cost of the fitness evaluation. Second, it shows the similar performance to the general GA. To verify the performance of the method, we designed the experiments using several benchmark functions and compared other fitness approximation methods.