Stock selection has long been recognized as a challenging and important task in finance. Recent advances in machine learning and data mining are leading to significant opportunities to solve these problems more effectively. In this study, we enrich our previous work for stock selection using single-objective genetic algorithms (SOGA) by extending it to the multi- objective GA (MOGA). In our previous work, we devised a stock scoring mechanism to rank and select stocks to form a portfolio, and we employed the SOGA for optimization of model parameters and feature selection for input variables to the model. In this work, we show how our MOGA models outperform the benchmark and improve upon our previous SOGA- based methods. Based on the promising results, we expect this MOGA methodology to advance the current state of research in soft computing for the real-world stock selection applications.

The two things equity investors are most concerned with are: 1) stock picking - finding a good investment target, and 2) timing of purchase and sale. That said, these two things will be the main theme of this research. Although common technical indicators are being widely used by investors to determine time to enter market, these techniques have been used by too many, obtaining these information have been made relatively easy, thus making these references less valuable. In addition, past technical indicators did not take into consideration the relationship between stock prices and trading volume. Many researches have shown that using a single variable as the basis of investment is insufficient and unwise. In view of the above, this research will propose a new technical indicator that takes into consideration both stock price and trading volume at the same time, thus making it superior in terms of investment timing. This research involves seven scenarios of the price- volume relationship, converts the daily transaction data of individual stocks into daily scores, then takes the scores and transforms those into short-term, as well as long-term moving averages. The crossing of the two averages will be used to predict the trend of the stock prices in the future, thus indicating the timing of investments. For example, when short-term moving average line breaks above the long-term moving average, it indicates a buy; when short-term moving average falls below the long-term moving average, it indicates a sell. Based on the empirical test results, the performance of the strategy discussed in this research was comparable with historical records, but involves less transactions while being easy to use. We can conclude from this that the new technique can provide investors a more valuable market reference.

Automatic music composition has been a challenging, interesting, and yet still much to be explored task primarily because it is hard to distinguish which song is good or bad which significantly impedes the automated composition process. Despite this difficulty, automated music composition would benefit many groups of people who ought to use a piece of their own music, as composed for them by an AI system with compositional intelligence, without someone else’s copyright for some purpose such as a music piece for a commercial, or a song played in the background of a presentation. Our composition system composes eight-bar tracks, based on western music theory and listener evaluation. We present here the use of memetic algorithm, comparing to using the conventional genetic algorithm. The same representation and evaluation for both techniques are used because of the similarity of these two algorithms. The main difference of memetic algorithm with genetic algorithm is the local search process. Both algorithms are implemented separately to spot the difference between the results then we evaluated the algorithms. When the outcomes are compared, we found that the use of memetic algorithm performs better in terms of quality of musical piece and convergence speed.

The computerized adaptive testing is to provide items that are consistent with the current ability of testee, and to decide the difficulty for the next selected item according to the correctness of the testee’s answer. It achieves the goals of adaptive learning through the mechanisms of dynamic adjustment of item difficulty to accelerate the test process or to shorten the number of items in a test. A prerequisite of computerized adaptive testing is to estimate the difficulties of items correctly. In this study, we describe the parameters of items by an adjusted approach. It considers each knowledge block as an independent dimension and gives a value for each dimension of the difficulty. Combined with the particle swarm optimization algorithm, a dynamic item selection strategy is proposed to develop an adaptive testing system. Therefore, it adopts the multiple assessment methods for the abilities by giving a value for each dimension of the ability. By way of the the dynamic item selection in computerized adaptive testing, all the selected items will be highly correlated and more consistent with the current actual abilities of the testees.

Particle Swarm Optimization (PSO) is a population-based optimization method simulating the social behavior of populations. Self-Organizing Map (SOM) is a neural network technique with good clustering performance. This research proposes the hybrid PSO and SOM for predicting Blog success level. Basically, PSO algorithm is adopted to find the optimal weights of the input variables of SOM, to reinforce the data in a single cluster (or certain clusters) with the same outcome level. A research model of Blog Success was proposed and 210 valid samples were collected from Internet users with Blog using or building experience. Compared with other prediction methods, the PSO-SOM approach of yielded better results than those of C5.0, Classification and Regression Trees (CART), Support Vector Machine (SVM) for 10-fold cross-validation.

In recent years, real-time strategy (RTS) games have gained attention in the AI research community for their multitude of challenging and relevant real-time decision problems that have to be solved in order to win against human experts or to collaborate effectively with other players in team games. However, a big challenge for creating human-level game AI is the different traits of races of opponents and their locations of enemy units are partially observable. To overcome this limitation, we explore evolutionary-based approach for estimating the location of enemy units that have been encountered. In this paper, we propose an efficient framework to predict the winning ratio between the different races used in the real-time strategy game. We represent state estimation as an optimization problem, and automatically learn parameters for the evolutionary-based model by learning a corpus of expert StarCraft replays. The evolutionary- based model tracks opponent units and provides conditions for activating tactical behaviors in our StarCraft bot. Our results show that incorporating a learned evolutionary-based model improves the performance of EISBot by 60% over baseline approaches.

The functional QOL model combines the existing approaches by linking the subjective representations of objectively measurable resources to their functional value for pursuing individually meaningful activities and goals. In this paper, we discuss the application of fQOL to improve and evaluate the effectiveness of interventions, in particular the coimagination method.

Recently, Japan (also world-wide countries) has become aged society, and a wide variety welfare device and system have been developed. But evaluation of welfare system and device are limited only stability, intensity and partial operability. So, evaluation of usefulness is insufficient. Evaluation of usefulness is necessity to consider about interaction of human and welfare device. In this paper, we measure load of sitting and standing movement to use EMG (Electoromyogram) and 3D Motion Capture and set a goal to establish objective evaluation method. We think that establishing objective evaluation method is necessity to develop useful welfare device. We examined possibility of assessing load and fatigue from measuring brain activity to use NIRS (Near Infra-Red Spectroscopy). We think that measuring load and fatigue is very important for developing user-friendly welfare device. Idea of universal design is widespread in welfare device and system. Measuring require verification of all generations. But, we performed to measure younger subjects as a first step.

Final purpose in this research is to contribute to developing of assistive robot and apparatus. Recently, there is a pressing need to develop a new system which assists and acts for car driving and wheelchair for the elderly as the population grows older. In terms of developing a new system, it is thought that it is important to examine behaviors as well as spatial recognition. Therefore, experiments have been performed for an examination of human spatial perceptions, especially right and left recognition, during car driving using NIRS. In previous research, it has been documented that there were significant differences at dorsolateral prefrontal cortex at left hemisphere during virtual driving task and actual driving. In this paper, brain activity during car driving was measured and detailed analysis was performed by segmentalizing brain activity during car driving on the basis of subjects’ motion. So, we report the relationship between brain activity and movement concerned with perception during driving in this paper.

Dementia is the progressive decline in cognitive function due to damage or disease in the body beyond what might be expected from normal aging. Dementia persons cannot reasonably live their lives. In order to support dementia persons’ lives, various approaches are proposed. Bozeat and Hodges showed affordance might give a certain support to (semantic) dementia persons of understanding (meanings of) objects. In this paper, based on the concept of affordance, abduction, and chance discovery, a relationship between dementia care and affordance is discussed.

More activities using digital cameras have been introduced to elderly people. The author has conducted such activities with elderly people in a Geriatric Health Services Facility for four years. Among them are Ms. A and Mr. B who take relatively　a fewer number of photos of fellow residents and make somewhat unfriendly comments on the photos they took. However, in the wake of the diseases they experienced and the losses of their family members, changes in their feelings towards fellow residents were observed. This paper discusses the psychological changes they underwent by sorting out and analyze the photos they took over the past two years.

Despite the growth of Consumer Health Informatics (CHI) websites, searching for relevant health information remains a challenging task for most users, especially for the consumer of non-medical professionals. Most consumers are not familiar with medical terminology; therefore the query keywords often do not accurately reflect their health information needs. The diversity of consumer familiarity with health topics also leads to frustration since the information presented may fall outside the consumer's comprehension level. This paper aims to investigate the effects of consumer familiarity on their health information seeking behavior and to develop a model to predict consumer familiarity with health topics. We conducted a user study with four health search tasks. This study analyzes several characteristics of search behavior, such as query keywords, query reformulation pattern, and search strategy. The query keywords submitted by the participants are classified into general and medical terminology by matching them with Consumer Health Vocabularies and Medical Subject Headings. The identification of query reformulation pattern addresses both syntax and semantic changes. The preliminary user study findings show that participants performed different query reformulation patterns for the familiar and non-familiar tasks. For future research, the findings and model from this study will be used to build an adaptive CHI system.

We report a case study of a family who chose to feed their elderly parent with cognitive difficulties using a feeding tube. We describe how the family members adopted tube feeding and how their lifestyle changed.

Local search algorithms are widely used to tackle combinatorial problems in several domains, one of the main features of these algorithms lies in the fact that they can tackle very large problems. Moreover, an interesting possibility is the potential speedup obtained by executing multiple copies of the sequential algorithm. In this paper, we review the current state-of-the art and future trends of parallel local search algorithms for the SAT problem, one the most important NP-complete problems. Generally speaking these algorithms can be broadly classified into two categories: parallel portfolios, where several algorithms compete and cooperate to solve a given problem instance and multi-flip algorithms where several flips of the variables are performed at the same time.

Scoring rule is getting attention in agent driven incentive designing. Smart grid is a new promising application field of multi-agent based strategies. Decentralization in power market brings out a new paradigm of energy providing where generation companies (GENCOs) sell energy to electricity providers (EP) who act as a distributer selling electricity to the consumers. A new smart pricing scheme for the consumers which is essentially based on a hierarchical multi-layered scoring rule and cost optimization is presented. The model assumes the availability of dynamic day-ahead pricing. Consumer agents (CA) predict the device scheduling for next day and report to EP. EP computes the total energy requirement that satisfies consumers demand in accordance with that prediction. EP will report the prediction to GENCO using a probability distribution. GENCO computes the discount based on the prediction provided by EPs while satisfying their demands. The discount will be distributed over EPs based on a scoring rule for their truthful predictions. Finally EP will charge consumers and distribute that discount according to the score CAs achieve for their device scheduling prediction. Continuous Ranked Probability Score is applied as the scoring mechanism.