V. Scott Gordon

Lamarckian Evolution, The Baldwin Effect and Function Optimization

We compare two forms of hybrid genetic search. The first uses Lamarckian evolution, while the second uses a related method where local search is employed to change the fitness of strings, but the acquired improvements do not change the genetic encoding of the individual. The latter search method exploits the Baldwin effect. By modeling a simple genetic algorithm we show that functions exist where simple genetic algorithms without learning as well as Lamarckian evolution converge to the same local optimum, while genetic search utilizing the Baldwin effect converges to the global optimum. We also show that a simple genetic algorithm exploiting the Baldwin effect can sometimes outperform forms of Lamarckian evolution that employ the same local search strategy.

A note on the performance of genetic algorithms on zero-one knapsack problems

For small zero-one knapsack problems, simple branch-and-bound and depth-first methods generate solutions much faster than our genetic algorithms. For large problems, branch-and-bound and depth-first methods outperform the genetic algorithms both for finding optimal solutions and for finding approximate solutions quickly. The simple methods perform much better than genetic algorithms on this class of problem in spite of the existence of a genetic encoding scheme which exploits useful local information. The results highlight the need for a better understanding of which problems are suitable for genetic algorithms and which problems are not.

Reported effects of rapid prototyping on industrial software quality

Empirical data are required to determine the effect of rapid prototyping on software quality. We examine 34 published and unpublished case studies of the use of rapid prototyping in ‘real-world’ software development. We identify common observations, unique events, and opinions. We develop guidelines to help software developers use rapid prototyping to maximize product quality and avoid common pitfalls.

Partitioning strategies for modular neural networks

We observe the effects of a variety of splitting strategies for partitioning the input domain in a self-splitting modular neural network applied to the two-spiral classification problem, and assisted by a special-purpose visualization tool. The observations motivate the development of an improved strategy, consisting of a series of binary splits along the boundaries of trained areas, and a particular weight initialization strategy. The work is leading to fewer networks and better generalization for this application, when backpropagation is used.

Visualization tool for a Self-Splitting modular Neural Network

We describe and implement a visualization tool for a Self-Splitting Neural Network (SSNN). The SSNN is a modular neural network that partitions the input domain during training through the identification of solved chunks and a divide-and-conquer strategy. The visualization tool shows a 2D projection of the input domain as partitioning proceeds, highlighting the boundaries of trained regions. Greyscale can be used to contrast the ranges of outputs so that generalization can be visually assessed. The tool is useful for illustrating how the SSNN works and for comparing different learning and splitting strategies.

Adaptive terrain-based memetic algorithms

The Terrain-Based Memetic Algorithm (TBMA) is a diffusion MA in which the local search (LS) behavior depends on the topological distribution of memetic material over a grid (terrain). In TBMA, the spreading of meme values -- such as LS step sizes -- emulates cultural differences which often arise in sparse populations. In this paper, adaptive capabilities of TBMAs are investigated by meme diffusion: individuals are allowed to move in the terrain and/or to affect their environment, by either following more effective memes or by transmitting successful meme values to nearby cells. In this regard, four TBMA versions are proposed and evaluated on three image vector quantizer design instances. The TBMAs are compared with K-Means and a Cellular MA. The results strongly indicate that utilizing dynamically adaptive meme evolution produces the best solutions using fewer fitness evaluations for this application.

TOPOLOGICAL AND MOTION STRATEGIES FOR CELLULAR MEMETIC TREE-BASED PORTFOLIO OPTIMIZATION

Portfolio Optimization (PO) is a resource allocation problem where real valued weights are assigned to multiple financial assets in order to maximize the return and minimize the risk. The Memetic Tree-based Algorithm (MTGA), employing a tree representation allied with local search (LS) has shown great performance compared to other weight balancing techniques. In this work, we hybridize MTGA with topological frameworks — Cellular Memetic Algorithms (CMA) — and study four implementations, varying whether the individuals move through the grid, and whether meta-parameters are spread along the axes for self-adaptation. The approaches are compared using a historical data simulation. A CMA which incorporates motion performs best, while parameter tuning was less successful. The results not only describe an improved method for PO, but also have broader implications for cellular models wherein the benefits of motion are shown to deserve further investigation.

National Standardization Strategy in Health Information Security Environment

Source technology for standard platform and common security infrastructure, ID/UHID management, and EMR-PHR security interlocking, combined with information technology within the hospital, are expected in the near future. In this paper, we propose a new strategy for future national standardization through international standardization trends and linkage in the field of health information security.