The objective of this research was to investigate which function is most appropriately describing the learning curve produced by C4.5 algorithm. The J48 implementation of the C4.5 algorithm was applied to datasets (n=121) from publicly available repositories (e.g. UCI) in step wise k-fold cross-validation. Four different functions (power, linear, logarithmic, exponential) were fit to the measured error rates. The decision trees error rate can be successfully modeled by an exponential function. Average mean squared error across all datasets was 0,052954 for exponential function, and was significantly different at P=0,001 from power and at P=0,000 from linear function. Our findings are consistent with tests performed in the area of human cognitive performance, e.g. with works by Heathcote et al, who were observing that the exponential function is best describing an individual learner; the same thus holds also for the artificial learner.
COBISS.SI-ID: 18250006
A novel user authentication and key agreement scheme has been developed for heterogeneous ad hoc wireless sensor networks. The proposed scheme enables a remote user to securely negotiate a session key with a general sensor node, using a lightweight key agreement protocol. The proposed scheme ensures mutual authentication between the user, sensor node and the gateway node (GWN), although the GWN is never contacted by the user. The proposed scheme has been adapted to the resource-constrained architecture of the WSN, thus it uses only simple hash and XOR computations. Our proposed scheme tackles these risks and the challenges posed by the IOT, by ensuring high security and performance features.
COBISS.SI-ID: 17757974
This Milestone Report firstly addresses the position regarding the areas of computers, computational intelligence and communications within IFAC. Subsequently, it addresses the role of computational intelligence in control. It focuses on four topics within the Computational Intelligence domain: neural network control, fuzzy control, reinforcement learning and brain machine interfaces. Within these topics the challenges and the relevant theoretical contributions are highlighted, as well as the expected future directions being pointed out.
COBISS.SI-ID: 18142742
In XML Schema development, the quality of XML Schemas is a crucial issue for further steps during the life cycle of an application, closely correlated with the structure of XML Schemas and different building blocks. Current research focuses on measuring the complexity of XML Schemas and mainly do not consider other quality aspects. This paper proposes a novel quality measuring approach, based on existing software engineering metrics, additionally defining the quality aspects of XML Schemas using the following steps: (1) definition of six schema quality aspects, (2) adoption of 25 directly measurable XML Schema variables, (3) proposition of six composite metrics, applying 25 measured variables and (4) composite metrics validation. An experiment was conducted using 250 standard XML Schemas collected from available e-business information systems. The results illustrate the influence of XML Schema's characteristics on its quality and evaluate the applicability of metrics during the measurement process, a useful tool for software developers while building or adopting XML Schemas.
COBISS.SI-ID: 18203926
Statistically fully valid sample is not needed for classification process; sampling from a database and building the classification models based on partial data is adequate. How much data to use depends on the algorithm, internal relations in data and the context. Generalization error function of learning algorithm in small l regime is presented and adaptive incremental k-fold cross-validation is developed. An important part of the adaptive incremental cross-validation are the conditions for detecting the point where the learning curve starts behaving well - the point where to stop the learning process.
COBISS.SI-ID: 18160918