Vol. 26 · No. 2

This document proposes a terminology and a model for representation of user data in information systems in the form of ``persona'' objects. It provides the mechanisms for evaluation how the personae relate to real-world subjects or to each other. A mechanism how to evaluate some anonymity and identity properties is proposed. This paper also describes the linking of personae by the use of shared identifiers. The local, global, persistent and transient personal linking types are considered. The model is used to describe persona linking in the different practical technologies as the example of model applicability.

Machine learning and knowledge representation are two fields of artificial intelligence that lead with intelligent reasoning, each one differently. When Knowledge Representation (KR) focuses more on the epistemological face of knowledge to carry out a power-expression model with detriment of the computation efficiency. Machine learning pays more attention to the computation efficiency, often with loss of expressing power. In this paper we show that features of one may overwhelm drawbacks of the other. Taking the uncertainty artefact from machine learning and symbolic representation from KR, we propose in this paper a new memory modelling for knowledge based systems which is at the same time machine-learning structure and a knowledge representation model. In terms of machine learning, our structure allows an unlimited flexibility where no restraining architecture is imposed at the beginning (think about decision trees citeRastogiS98, garofalakis00efficient, Mantaras91, buntine93learning). Classification can be performed with incomplete vectors where the most likely corresponding class is assigned to the vector with missing attributes. Viewed as a knowledge model, a basic knowledge is easily specified graphically. Inference is defined by rules expressed in the same manner, where the existing sub-instances are used to generate new connections and entities. Inference on existing knowledge is described in two algorithms. Our approach is mainly based on the representation of the context concept. Our model brings together advantages of the symbolic knowledge representation, namely human to computer knowledge coding, and those of machine learning structures, namely ease of efficient coding of inference to perform the so-called intelligent tasks like pattern recognition, prediction and others. Its graphical representation allows visualization of both dynamic and static sides of the model (i.e. inference and knowledge).

Algorithms for many complex computations assume that all the relevant data is available on a single node of a computer network. In the emerging distributed and networked knowledge environments, databases relevant for computations may reside on a number of nodes connected by a communication network. These data resources cannot be moved to other network sites due to privacy, security, and size considerations. The desired global computation must be decomposed into local computations to match the distribution of data across the network. The capability to decompose computations must be general enough to handle different distributions of data and different participating nodes in each instance of the global computation. In this paper, we present a methodology wherein each distributed data source is represented by an agent. Each such agent has the capability to decompose global computations into local parts, for itself and for agents at other sites. The global computation is then performed by the agent either exchanging some minimal summaries with other agents or travelling to all the sites and performing local tasks that can be done at each local site. The objective is to perform global tasks with a minimum of communication or travel by participating agents across the network.

In this article we present the general architecture of a cellular machine, which makes it possible to reduce the size of induction graphs, and to optimize automatically the generation of symbolic rules. Our objective is to propose a tool for detecting and eliminating non relevant variables from the database. The goal, after acquisition by machine learning from a set of data, is to reduce the complexity of storage, thus to decrease the computing time. The objective of this work is to experiment a cellular machine for systems of inference containing rules. Our system relies upon the graphs generated by the SIPINA method. After an introduction aiming at positioning our contribution within the area of machine learning, we briefly present the SIPINA method for automatic retrieval of knowledge starting from data. We then describe our cellular system and the phase of knowledge post-processing, in particular the validation and the use of extracted knowledge. The presentation of our system is mostly done through an example taken from medical diagnosis.

In this paper, two new paradigms are proposed with fractal collage and reconstruction residuals to enhance FIC. In the first new paradigm, FIC is optimized using the reconstruction residuals. In the second paradigm, the selected collage residuals are used to correct the iterated function system (IFS) of FIC, and an effective technique for coding the selected collage residuals is applied based on DCT and embedded bit-plane coding. In the first paradigm, the reconstruction quality is improved without increasing the bit rate. Using the second paradigm, we can improve the reconstruction quality with a little bit (about 0.01 bpp) increase in bit rate. Experimental results show that the proposed paradigms achieve better performance than JPEG at lower bit rate and similar performance at higher bit rate.