Volume 6 (2016): Edizione 3 (July 2016)

We have two motivations. Firstly, semantic gap is a tough problem puzzling almost all sub-fields of Artificial Intelligence. We think semantic gap is the conflict between the abstractness of high-level symbolic definition and the details, diversities of low-level stimulus. Secondly, in object recognition, a pre-defined prototype of object is crucial and indispensable for bi-directional perception processing. On the one hand this prototype was learned from perceptional experience, and on the other hand it should be able to guide future downward processing. Human can do this very well, so physiological mechanism is simulated here. We utilize a mechanism of classical and non-classical receptive field (nCRF) to design a hierarchical model and form a multi-layer prototype of an object. This also is a realistic definition of concept, and a representation of denoting semantic. We regard this model as the most fundamental infrastructure that can ground semantics. Here a AND-OR tree is constructed to record prototypes of a concept, in which either raw data at low-level or symbol at high-level is feasible, and explicit production rules are also available. For the sake of pixel processing, knowledge should be represented in a data form; for the sake of scene reasoning, knowledge should be represented in a symbolic form. The physiological mechanism happens to be the bridge that can join them together seamlessly. This provides a possibility for finding a solution to semantic gap problem, and prevents discontinuity in low-order structures.

Traditionally, the volatility of daily returns in financial markets is modeled autoregressively using a time-series of lagged information. These autoregressive models exploit stylised empirical properties of volatility such as strong persistence, mean reversion and asymmetric dependence on lagged returns. While these methods can produce good forecasts, the approach is in essence atheoretical as it provides no insight into the nature of the causal factors and how they affect volatility. Many plausible explanatory variables relating market conditions and volatility have been identified in various studies but despite the volume of research, we lack a clear theoretical framework that links these factors together. This setting of a theory-weak environment suggests a useful role for powerful model induction methodologies such as Genetic Programming (GP). This study forecasts one-day ahead realised volatility (RV) using a GP methodology that incorporates information on market conditions including trading volume, number of transactions, bid-ask spread, average trading duration (waiting time between trades) and implied volatility. The forecasting performance from the evolved GP models is found to be significantly better than those numbers of benchmark forecasting models drawn from the finance literature, namely, the heterogeneous autoregressive (HAR) model, the generalized autoregressive conditional heteroscedasticity (GARCH) model, and a stepwise linear regression model (SR). Given the practical importance of improved forecasting performance for realised volatility this result is of significance for practitioners in financial markets.

A novel approach for instance selection in classification problems is presented. This adaptive instance selection is designed to simultaneously decrease the amount of computation resources required and increase the classification quality achieved. The approach generates new training samples during the evolutionary process and changes the training set for the algorithm. The instance selection is guided by means of changing probabilities, so that the algorithm concentrates on problematic examples which are difficult to classify. The hybrid fuzzy classification algorithm with a self-configuration procedure is used as a problem solver. The classification quality is tested upon 9 problem data sets from the KEEL repository. A special balancing strategy is used in the instance selection approach to improve the classification quality on imbalanced datasets. The results prove the usefulness of the proposed approach as compared with other classification methods.

The control of nonholonomic canonical form using an invariant manifold is investigated to apply to a mobile robot steered by two independent driving wheels. A quasi-continuous exponential stabilizing controller is designed by using another input pattern. Additionally, the control gain designing method is proposed for this controller. Modified error system of nonholonomic double integrator model is used as nonholonomic canonical form. Generally, the gain cannot be calculated due to the non-linear transform of system. Owing to complicated relation of several parameters, the controller behavior is inconstant by gain pattern. We propose a method of designing gain which uses desired settling time. An approximate equation to obtain designed gains is derived based on the evaluation function. The design method to determine gains of the assumed actual system is simulated. The effectiveness of the proposed method is confirmed by these simulations.

Detecting genetic association models between single nucleotide polymorphisms (SNPs) in various disease-related genes can help to understand susceptibility to disease. Statistical tools have been widely used to detect significant genetic association models, according to their related statistical values, including odds ratio (OR), chi-square test (χ²), p-value, etc. However, the high number of computations entailed in such operations may limit the capacity of such statistical tools to detect high-order genetic associations. In this study, we propose lsGA algorithm, a genetic algorithm based on local search method, to detect significant genetic association models amongst large numbers of SNP combinations. We used two disease models to simulate the large data sets considering the minor allele frequency (MAF), number of SNPs, and number of samples. The three-order epistasis models were evaluated by chi-square test (χ²) to evaluate the significance (P-value < 0.05). Analysis results showed that lsGA provided higher chi-square test values than that of GA. Simple linear regression indicated that lsGA provides a significant advantage over GA, providing the highest β values and significant p-value.