International Journal of Knowledge-Based and Intelligent Engineering Systems

Self-explanation has been used successfully in teaching Mathematics and Physics to facilitate deep learning. We are interested in investigating whether self-explanation can be used in an open-ended, ill-structured domain. For this purpose, we enhanced KERMIT, an intelligent tutoring system that teaches conceptual database design. The resulting system, KERMIT-SE, supports self-explanation by engaging students in tutorial dialogues when their solutions are erroneous. The results of an evaluation study indicate that self-explanation leads to improved performance in both conceptual and procedural knowledge.

We present a great improvement while performing semi-supervised learning tasks from training data sets when only a small fraction of the data pairs is labeled. In particular, we propose a novel decision strategy based on normalized model outputs. We give the explanation why the normalization step helps. The paper compares performances of two popular semi-supervised approaches (Consistency Method and Harmonic Gaussian Model) on the unbalanced and balanced labeled data by using normalization of the models' outputs and without it. Experiments on text categorization problems suggest significant improvements in classification performances for models that use normalized outputs as a basis for final decision.

This paper first introduces a unified representation of both real-coded genes and binary-coded genes, in order to efficiently analyse the convergence of a genetic algorithm. In the literature, the syntagm "the entire population premature convergence" is used with the meaning of closing the evolution before reaching the optimal point. It can be emphasised only on a test function with known landscape. If the function landscape is unknown, one can only notice the population convergence. This paper aims to answer to the question: "how can we influence the control parameters of the genetic algorithm so that the exploration period of the parameter space be longer and the risk of the premature convergence be reduced?" The answer to the above question implies the selection of a crossover operator with good performance in the landscape exploration and the use of two indicators for the detection of the population convergence. In order to choose the appropriate control parameters of the genetic algorithm, the fitness function landscape must be taken into consideration.

DNA memories have the potential not only to store a vast amount of information with high density, but also may be able to process the stored information through laboratory protocols that match content and context. This might lead to knowledge mining applications on a massively parallel scale, and to a limited capability for intelligent processing of stored data to discover semantic information. In this paper, a design for such a DNA memory is presented.

Recently, the chaotic method is employed to forecast a short-term future using uncertain data. This method makes it possible to restructure the attractor of given time-series data in the multi-dimensional space through Takens' embedding theory.

However, some time-series data have less chaotic characteristic. In this paper, Time-series data are divided using Wavelet Transform. It will be shown that the divided orthogonal elements of time-series data are employed to forecast more precisely than original time-series data. The divided orthogonal time-series data are forecasted using Chaos method. Forecasted data are restored to the original data by inverse wavelet transform.

In the previous work, we have proposed a new approach to pattern recognition tasks, in which not only a classifier but also a feature space is trained incrementally. To implement this idea, Incremental Principal Component Analysis (IPCA) and Evolving Clustering Method (ECM) were effectively combined. However, the original IPCA only gives a way to determine the increase of a new feature based on a threshold value, whose value must be optimized for different datasets. In this paper, to alleviate the dependency on datasets, the accumulation ratio is introduced as its criterion, and an improved algorithm of IPCA is derived. To see if correct feature construction is carried out by this new IPCA algorithm, the classification performance is evaluated over some standard datasets when Evolving Clustering Method (ECM) is adopted as a prototype learning method for Nearest Neighbor classifier. Our simulation results show that the proposed IPCA works well without elaborating sensitive parameter optimization and its recognition accuracy outperforms that of the previous model.

In this paper, we propose a new collision avoidance algorithm which is inspired by insect vision. Insects can obtain necessary information for their behavior, such as position and velocity of obstacle, even when they are moving fast. Insects realize this by using micro photo sensors and simple processing. We propose a visual system based on structure of compound eye and simple processing. In the proposed system, ommatidia realized by photo diodes are arranged in semicircle type. When obstacle appears in visual field, three information are extracted, which are the number, the changes in the number and the changes of center position of the active ommatidia. Then the collision possibility is estimated by fuzzy inference using these information. If the collision possibility becomes large, avoidance behavior is executed. We show its performance and compare the results to previous approaches.

This paper proposes multi-objective behavior coordination based on fuzzy control and perceptual system for vision-based mobile robots used as a soccer robot. The perception can be assumed as a searching process of information that can navigate an agent toward a target state in real-time. In this paper, we propose a time-series of a searching method based on a steady-state genetic algorithm for the visual perception. The search of the target objects such as a ball and a goal is done in the multi-resolutional image. Next, we propose multi-objective behavior coordination with fuzzy controller for soccer robots. The relative position and size of the ball and goal are used as inputs to the fuzzy controller. Finally, we show several experimental results of vision-based soccer robots.