=====================================
Background:
I currently work for Boeing where I completed and implemented several very successful optimization projects on metal and composite structure for the revolutionary 787 airplane program.
Magna Cum Laude graduate from the University of Michigan B.S.E. Mechanical Engineering, minor in Mathematics.
Currently pursuing M.S. in Aeronautical Engineering and M.S. in Applied Mathematics with thesis in composite optimization at University of Washington.
=====================================
Draft Presentation:
I plan to give an exciting presentation on scalability of optimization algorithms, specifically genetic algorithms, from basic concepts to the latest developments and their applications. I will begin with an introduction showcasing the importance of optimization. I will cover some basic examples where the obvious solution to an optimization problem is very non-intuitive (e.g. minimum surface area between two circles: catenoid not cylinder). I will show how the catenoid occurs in nature (bubbles), and how we can perhaps learn something from nature to optimize solutions (foreshadowing genetic algorithm approach to optimization). Finally I will introduce how easily simple practical challenges become very large (up to a billion variables) problems for optimization, and thus there exists a crucial need for scalability in optimization.
After the introduction, I will cover various basic optimization examples and their broad applications, including biology, computer science, engineering, and much more. This will lead into a discussion of the scalability of various basic optimization algorithms (Newton's method, hill climbing, gradient descent, genetic algorithms, etc.). Then I will introduce the importance of parallelism for scalability in algorithms and demonstrate the efficiency of solving small sections of a larger problem simultaneously. As a case study, we will go through the development and thought process in scaling a simple optimization algorithm.
Then we will move into the basics of nature's optimizer- genetic algorithms, and I will lead the audience into an understanding of the biological metaphor, theory, and terminology with simple binary examples. Using these building blocks I will discuss how the scalability of genetic algorithms is much more effective than simple 'hill-climber' algorithms for large variable problems. I will reference results of previous research to demonstrate that genetic algorithms will optimize problems quickly, reliably and accurately, regardless of noise for these large variable problems.
The final section of the talk will focus in detail on the most recent and exciting developments in genetic optimization algorithms. First I will discuss theoretically why genetic algorithms, in particular, are so scalable. This will involve some basics of Bayesian theory, and in more detail, how to analyze the structure and parameters (conditional probabilities) of Bayesian networks. Briefly I will also go into a discussion of coupling and interactions between variables and how these can complicate approaches to implementing a genetic algorithm, and how to address these complications. Then we will move on to actual examples of their proven scalability, including the very recently (February 2007) solved billion variable problem (blind, noisy, one-max) using a genetic algorithm. Finally I plan to introduce some creative new ideas and opportunities for further work and development of scalability of optimization algorithms.
In conclusion I will show the implications of these new developments in proven scalability of genetic algorithms through summarizing broad practical applications to various optimization problems. Then I will open the floor to any questions and further discussion the audience may wish to pursue.
=====================================
References:
Books
Goldberg D. E. (2002). The design of innovation: Lessons from and for competent genetic algorithms
Goldberg D. E. (1989). Genetic algorithms in search, optimization, and machine learning.
Mitchell M. (1998). An introduction to genetic algorithms (complex adaptive systems)
Articles
Goldberg D. E., Sastry K., & Llora X. (2006). Toward routine billion-variable optimization using genetic algorithms.
Papers
Benson S. J., McInnes L. C., & More J. J. (2001). A case study in the performance and scalability of optimization algorithms.
Fitzpatrick S., & Meertens L. (2002). Scalable, anytime constraint optimization through iterated, peer-to-peer interaction in sparsely-connected networks.
Ocenasek J., & Pelikan M. (2004). Parallel mixed Bayesian optimization algorithm: A scale-up analysis.
Pelikan M., Sastry K., & Goldberg D. E. (2002). Evolutionary Algorithms + Graphical Models = Scalable Black-Box Optimization.
Pelikan M., Sastry K., & Goldberg D. E. (2005). Multiobjective hBOA, Clustering, and Scalability.
Sastry K., Goldberg D. E., & Llora X. (2007). Toward billion bit optimization via efficient genetic algorithms.