Information Technology



NED-IIS: An Intelligent Information System for Forest Ecosystem Management

AAAI Conferences

In the past decade, organizations have been moving mainframe-based systems toward open, distributed computing environments. The demand for interoperability has been driven by the accelerated construction of largescale distributed systems for operational use and by increasing use of the Internet (Manola 1995). Distributed computing offers many advantages, including location transparency to users, scalability, fault tolerance, load balancing and resource sharing. As such, much of the interoperability literature has been concerned with distributed computing; for example, the recent emergence of Java, the Object Management Group's CORBA (Common Object Request Broker Architecture) and Microsoft's DCOM (Distributed Component Object Model) are all for this purpose (Grimes 1997, Leppinen et al. 1997, OMG 1997). In addition, object orientation (OO) is probably the most widely used approach in software development and the basis for the CORBA and DCOM interoperability architectures.


Recent Applications of Artificial Neural Networks in Forest Resource Management: An Overview

AAAI Conferences

Making good decisions for adaptive forest management has become increasingly difficult. New artificial intelligence (AI) technology allows knowledge processing to be included in decision-support tool. The application of Artificial Neural Networks (ANN), known as Parallel Distributed Processing (PDP), to predict the behaviours of nonlinear systems has become an attractive alternative to traditional statistical methods. This paper aims to provide an up-to-date synthesis of the use of ANN in forest resource management. Current ANN applications include: (1) forest land mapping and classification, (2) forest growth and dynamics modeling (3) spatial data analysis and modeling (4) plant disease dynamics modeling, and (5) climate change research. The advantages and disadvantages of using ANNs are discussed. Although the ANN applications are at an early stage, they have demonstrated potential as a useful tool for forest resource management.


Intelligent Decision Support for Aerial Spray Deposition Management

AAAI Conferences

Determining the parameter value settings to use as input to the AGDISP Aerial Spray Simulation Model [Bila89] in order to produce a desired spray material deposition is considered an instance of a parametric design problem [Davi91]. Parametric design is a specialization or subtype of the more generic design problem. Typically, when working on a design problem, the solution representation is a set of instructions or components for achieving the design goals. This representation can also be called a configuration, especially if the elements comprising the configuration are predefined. For the parametric design problem we are dealing with, these elements correspond to the AGDISP simulation input parameters.


Towards Environmental Decision Support Systems

AAAI Conferences

This is a position paper willing to fix concepts in the interface between Environmental Issues and Artificial Intelligence. In particular we will focus our attention in a class of systems called Environmental Decision Support Systems (EDSS) that have been under develop in the last 10 years.


A hybrid expert system, GIS and simulation modeling for environmental and technological risk management

AAAI Conferences

Based on the results from the international research project HITERM funded under the European ESPRIT technology programme for high-performance computing and networking (HPCN) for decision support, this paper presents RTXPS, the integration of a real-time forward chaining expert system and a backward chaining system as the DSS framework using simulation models and GIS for environmental and technological risk assessment and management. Application examples describe chemical emergency management cases for fixed installations and mobile sources (transportation accidents), based on ongoing case studies in Italy, Switzerland and Portugal.


On data clustering with artificial ants

AAAI Conferences

We present in this paper a new ant based approach named AntClass for data clustering. This algorithm uses the stochastic principles of an ant colony in conjunction with the deterministic principles of the Kmeans algorithm. It first creates an initial partition using an improved ant-based approach, which does not require any information on the input data (such as the number of classes, or an initial partition). Then it uses the Kmeans to speed up the convergence of the stochastic approach. In a second phase, AntClass uses hierarchical clustering where ants may cluster together heaps of objects and not just objects.


Genetic Algorithms as Post Processors for Data Mining

AAAI Conferences

The steps of the process from data to final discovered knowledge artefact are illustrated in figure 1. Genetic algorithms have been applied to several of these steps. Meggs (1996) reports good results when using a genetic algorithm to select feature subsets from a database before classifiers are learned using C4.5. Zelezbujow and Stranien (1998) report using a GA to select features before attempting knowledge discovery on a legal data set. Genetic algorithms have been applied to the mining step of the process: that is GA's have been used to search for the concepts or domain theory that generated the data in the Data base or data warehouse. For example GA's have been used to search for first order predicates entailed by data (Angier et al 1995) and discover rules in data (Pei et al 1997). Flockhart and Radcliffe (1996) have exploited the natural parallelism of GA's to develop a massively parallel data mining system. The next step in the classic KDD process as illustrated in figure 1 would be to evaluate the discovered rule set using, for example, statistical techniques such as significance tests and attempt to integrate it with any preexisting knowledge available. However in our experience a postprocessing step where the classifier or knowledge that has been discovered is optimised to remove low utility structures can be useful. It is the use of GAs in this phase of the process that we address in this paper.


Feature Selection Mechanisms for Ensemble Creation: A Genetic Search Perspective C sar Guerra-Salcedo

AAAI Conferences

Ensembles of classifiers has shown to be very effective for case-based classification problems. Several methods for ensembles has been proposed (Dietterich 1998) with significant improvings over single-classifier techniques. However, the ensemble creation algorithms have been used with the entire set of features available. Tin Kam Ho (Ho 1998b), (Ho 1998a) published method for construction of classifiers ensembles based on random feature selection. Her method relies on the fact that combining multiple classifiers constructed using randomly selected features can achieve better performance in classification than using the complete set of features for the ensemble creation.


Mining Predictive Patterns in Sequences of Events

AAAI Conferences

Learning to predict rare events from sequences of events with categorical features is an important, real-world, problem. Unfortunately, most machine learning methods that learn classification "rules" are not suited to solving this type of problem because they assume an unordered set of examples and cannot identify patterns between "examples" (i.e., events). Statistical time-series prediction methods are also not suitable, since they assume numerical features. Genetic algorithms, however, which have often been used to find patterns in data, are well suited to finding predictive temporal and sequential patterns in the event sequence data. In order to solve the event prediction problem, we developed Timeweaver, a genetic-based machine learning system that, given a pre-specified "target" event, learns to identify patterns in the data that successfully predict the future occurrence of that event.