Jerzy W. Bala

Visual routine for eye detection using hybrid genetic architectures

We address the problem of crafting visual routines for detection tasks. Emphasis is placed on both competition and learning to help with specific visual tasks involved in localization and identification. Crafting of visual routines presents difficult optimization problems and leads to evolutionary computation using a hybrid genetic architecture consisting of natural selection, learning, and their beneficial interactions. Base features representations and visual routines for detection represented as decision trees are evolved. The visual routine considered is that of eye detection. The experimental results reported herein prove the feasibility of our approach in terms of feature selection (data compression) and the corresponding eye detection (pattern recognition).

Shape analysis using genetic algorithms

Abstract This paper introduces a novel methodology for shape discrimination by combining pattern recognition techniques such as morphological processing with concepts from artificial intelligence and machine learning such as genetic algorithms (GAs). High-performance shape discrimination operators, defined as variable structuring elements and sequenced as program forms, are derived using GAs. The population of operators, iteratively evaluated according to an performance index corresponding to shape discrimination ability, evolves into an optimal set of operators using the evolutionary principles of genetic search. Experimental results are included to illustrate the feasibility of our novel methodology for developing robust shape analysis methods.

A Decision Tree Algorithm for Distributed Data Mining: Towards Network Intrusion Detection

This paper presents preliminary works on an agent-based approach for distributed learning of decision trees. The distributed decision tree approach is applied to intrusion detection domain, the interest of which is recently increasing. In the approach, a network profile is built by applying a distributed data analysis method for the collection of data from distributed hosts. The method integrates inductive generalization and agent-based computing, so that classification rules are learned via tree induction from distributed data to be used as intrusion profiles. Agents, in a collaborative fashion, generate partial trees and communicate the temporary results among them in the form of indices to the data records. Experimental results are presented for military network domain data used for the network intrusion detection in KDD cup 1999. Several experimental results show that the performance of distributed version of decision tree is much better than that of non-distributed version with data collected manually from distributed hosts.

Agent based distributed data mining

This paper presents an agent-based distributed data mining approach dealing with heterogeneous databases located at different sites. It introduces a modified decision tree algorithm on an agent based framework, which produces an accurate global model without transferring data between agents. The novel approach is evaluated over a test bed of texture feature data of 184 aerial photograph images. The experimental results show that the distributed version with more agents outperforms the version with fewer agents when the rule generation from the large database is not complicated.

Shape analysis using hybrid learning

Abstract This paper is concerned with hybrid learning and it describes how to combine evolution and symbolic learning for shape analysis. The methodology introduced in this paper integrates genetic algorithms (GAs) characteristic of evolutionary learning with empirical inductive generalization characteristic of symbolic learning. GAs evolve operators that discriminate among image classes comprising different shapes, where the operators are defined as variable morphological structuring elements that can be sequenced as program forms. The optimal operators evolved by GAs are used to derive discriminant feature vectors, which are then used by empirical inductive learning to generate rule-based class description in disjunctive normal form (DNF). The GA constitutes the data-driven, performance-oriented part of the shape analysis system, while the empirical inductive generalization is the model-driven part of the system. The rule-based descriptions are finally optimized by removing small disjuncts in order to enhance the robustness of the shape analysis system. Experimental results are presented to illustrate the feasibility of our novel methodoloy for discriminating among classes of different shaped objects and for learning the concepts of convexity and concavity.

Combining structural and statistical features in a machine learning technique for texture classification

This paper presents a method for applying inductive learning techniques to texture description. Local texture features described as eight attributes have been extracted for each pixel from small windows (5x5, 7x7 or 9x9) centered around the pixel. The extra ninth attribute is computed from larger global area (25*25) as a co-occurrence matrix parameter. All nine attributes from an event, which is essentially a point in a 9-dimensional attribute space. Sets of such events are computed for different texture classes, and the inductive learning AQ algorithm is used to generate a given class description. Such learned descriptions are evaluated against different texture samples. Results of experiments performed on eight textural images are presented.

Performance evaluation of an agent based distributed data mining system

This paper presents a distributed approach to build decision trees in a lock step manner with each node proposing an attribute on which to split A central mediator chooses the attribute, among the candidates, with the highest information gain The chosen split is then effectively communicated to the other agents to partition their data The distributed decision tree approach is performed on the agent based architecture dealing with distributed databases This paper mainly focuses on the evaluation of the system performance in distributed data mining Even though there are several trials suggesting algorithms of distributed data mining, few efforts have made on the definition of the system performance It is very important to define the performance for the further development of distributed data mining.

Application of a distributed data mining approach to network intrusion detection

In very many situations the collection of data from distributed hosts for its subsequent use to generate an intrusion detection profile may not be technically feasible (e.g., due to data size or network security transfer protocols). This situation is especially evident for data intensive intrusion profile generation (e.g., inducing profiles via data mining techniques). An alternative solution is to build a network profile by applying distributed data analysis methods (e.g., agent based computing). Such an approach is described in this paper. Global profiles are built using a Distributed Data Mining approach that integrates inductive generalization and Agent based computing. In this approach, classification rules are learned via tree induction from distributed data to be used as intrusion profiles. Agents, in a collaborative fashion, generate partial trees and communicate the temporary results among them in the form of indices to the data records. The process is terminated when a final tree is induced. This communication mechanism does not involve any data transfers, and in addition, a compression approach is used to reduce the communication bandwidth of data index transfers.

Shape analysis using morphological processing and genetic algorithms

A novel way of combining morphological processing and genetic algorithms (GAs) to generate high-performance shape discrimination operators is presented. GAs can evolve operators that discriminate among classes comprising different shapes. The operators are defined as variable structuring elements and can be sequenced as program forms. The population of such operators, evaluated according to an index of performance corresponding to shape discrimination ability, evolves into an optimal set of operators using genetic search. Experimental results are presented to illustrate the feasibility of the approach for shape discrimination.<<ETX>>

Using Genetic Algorithms to Improve the Performance of Classification Rules Produced by Symolic Inductive Methods

In this paper we present a novel way of combining symbolic inductive methods and genetic algorithms (GAs) applied to produce high-performance classification rules. The presented method consists of two phases. In the first one the algorithm induvtively learns a set of classification rules from noisy input examples. In the second phase the worst performing rule is optimized by GAs techniques. Experimental results are presented for twelve classes of noisy data obtained from textured images.