Douglas R. McGregor

Designing application-specific neural networks using the structured genetic algorithm

Presents a different type of genetic algorithm called the structured genetic algorithm (SGA) for the design of application-specific neural networks. The novelty of this new genetic approach is that it can determine the network structures and their weights solely by an evolutionary process. This is made possible for the SGA primarily due to its redundant genetic material and a gene activation mechanism which in combination provide a multi-layered structure to the chromosome. The authors focus on the use of this learning algorithm for automatic generation of a complete application specific neural network. With this approach, no a priori assumptions about topology are needed and the only information required is the input and output characteristics of the task. The empirical studies show that the SGA can efficiently determine the network size and topology along with the optimal set of connection weights appropriate for desired tasks, without using backpropagation or any other learning algorithm.<<ETX>>

A MORE BIOLOGICALLY MOTIVATED GENETIC ALGORITHM: THE MODEL AND SOME RESULTS

For more than two decades, genetic algorithms (GAs) have been studied by researchers from different fields. Over the years, many modifications have been suggested to alleviate the difficulties encountered by GAs in solving different problems. Despite these modifications, with the increase in application traditional GAs remain inadequate for many practical purposes. This paper introduces a new genetic model called the structured genetic algorithm (sGA) to address some of the difficulties encountered by the simple genetic approaches in solving various types of problems. The novelty of this genetic model lies primarily in its redundant genetic material and a gene activation mechanism that utilizes a multilayered structure for the chromosome. This representation provides genetic variation and has many advantages in search and optimization. For example, it can retain multiple (alternative) solutions or parameter spaces in its representation. In effect, it also works as a long-term distributed memory within the...

High-Performance Operations Using a Compressed Database Architecture

Future database applications will require significant improvements in performance beyond the capabilities of conventional disk based systems. This paper describes a new approach to database systems architecture, which is intended to take advantage of solid-state memory in combination with data compression to provide substantial performance improvements. The compressed data representation is tailored to the data manipulation operations requirements. The architecture has been implemented and measurements of performance are compared to those obtained using other high-performance database systems. The results indicate from one to five orders of magnitude speed-up in retrieval, equivalent or slightly faster performance during insertion (and compression) of data, while achieving approximately one order of magnitude compression in data volume. The resultant architecture is thus capable of greater cost/effectiveness than conventional approaches.

Data compression in database systems

This paper addresses the question of how information-theoretically-derived compact representations can be applied in practice to improve storage and processing efficiency in DBMS. Compact data representation has the potential for savings in storage, access and processing costs throughout the systems architecture and may alter the balance of usage between disk and solid state storage. To realise the potential performance benefits, however, novel systems engineering must be adopted to ensure that compression/decompression overheads are limited. This paper describes a basic approach to storage and processing of relations in a highly compressed form. A vertical columnwise representation is adopted in which columns can dynamically vary incrementally in both length and width. To achieve good performance query processing is carried out directly on the compressed relational representation (using a compressed representation of the query), thus avoiding decompression overheads. Measurements of performance of the Hi-base prototype implementation are compared with those obtained from conventional DBMS.

Short term unit-commitment using genetic algorithms

The authors present a genetic approach for determining the priority order in the commitment of thermal units in power generation. The objective of the problem is to properly schedule the on/off states of all thermal units in a system to meet the load demand and the reverse requirement at each time interval, such that the overall system generation cost is a minimum, while satisfying various operational constraints. The authors examine the feasiblity of using genetic algorithms and report some simulation results in near-optimal commitment of thermal units in a power system.

Digital image registration using structured genetic algorithm

This paper describes a new genetic approach called the structured genetic algorithm (sGA) for automatic registration of digital images. The specialty of this genetic model lies primarily in its redundant genetic material and a gene activation mechanism which utilizes a multi-layered structure for the chromosome. The additional genetic material serves to retain multiple optional solution spaces in parameter optimization. The structured genetic model is applied here to minimize the registration measures in image transformations, as investigated by Fitzpatrick and Grefenstatte with the simple GA. The results demonstrate that sGA is a much faster and robust search method that is guaranteed to reach a global optimum by adaptively estimating the subspace from the maximum space during the evolutionary process. Preliminary experimental results are reported.

Genetically designing neuro-controllers for a dynamic system

In this paper, me describes the application of a structured genetic algorithm for integrating the process of design and training neural networks for a specific task. The important feature of this genetic approach is that it can determine the network structures and their weights solely by an evolutionary process. The paper presents some experimental results in the automatic design of neural network based controllers for balancing a typical dynamic (pole-cart) system using this genetic approach.

Elimination of Redundant Views in Multidimensional Aggregates

On-line analytical processing provides multidimensional data analysis, through extensive computation based on aggregation, along many dimensions and hierarchies. To accelerate query-response time, pre-computed results are often stored for later retrieval. This adds a prohibitive storage overhead when applied to the whole set of aggregates. In this paper we describe a novel approach which provides the means for the efficient selection, computation and storage of multidimensional aggregates. The approach identifies redundant aggregates, by inspection, thus allowing only distinct aggregates to be computed and stored. We propose extensions to relational theory and also present new algorithms for implementing the approach, providing a solution which is both scalable and low in complexity. The experiments were conducted using real and synthetic datasets and demonstrate that significant savings in computation time and storage space can be achieved when redundant aggregates are eliminated. Savings have also been shown to increase as dimensionality increases. Finally, the implications of this work affect the indexing and maintenance of views and the user interface.

Database Compression Using an Offline Dictionary Method

Off-line dictionary compression is becoming more attractive for applications where compressed data are searched directly in compressed form. While there has been large body of related work describing specific database compression algorithms, the Hibase [10] architecture is unique in processing queries in compressed data. However, this technique does not compress the representation of strings in the domain dictionaries. Primary keys, data with high cardinality and semi-structured data contribute very little or no compression. To achieve high performance irrespective of type of data, the string representation must be in compressed form. At the same time, the direct addressability of compressed data is maintained. Serial compression techniques cannot be used. In this paper, we present a prefix dictionary-based off-line method that can be incorporated with systems like Hibase where compressed data can be accessed directly without prior decompression. The complexity is O(n) in time and space.

Integrating virtual reality and telepresence to remotely monitor construction sites: A ViRTUE project

The paper presents VIRTUE, a funded research project, which aims to integrate the Virtual Reality (VR), Telepresence (TP) and mobile video telecommunications technologies. A mobile, real-time, 3D-hybrid VR/TP system is being built at Strathclyde University, Glasgow. A system prototype has been completed and is being tested. The system will permit the user to integrate telepresence images with computer generated virtual environments superimposed over the remote real worldview. This integrated system incorporates emerging mobile telecommunications technologies to give rapid and easy access to the real and virtual construction sites from arbitrary locations. This system allows remote surveillance of the construction site, and integration of real world images of the site with virtual reality representations, derived from planning models, for progress monitoring.