Papers by Kenneth Holladay
Proceedings of the 11th Annual conference on Genetic and evolutionary computation - GECCO '09, 2009
Solving complex, real-world problems with genetic programming (GP) can require extensive computin... more Solving complex, real-world problems with genetic programming (GP) can require extensive computing resources. However, the highly parallel nature of GP facilitates using a large number of resources simultaneously, which can significantly reduce the elapsed wall clock time per GP run. This paper explores the performance characteristics of an MPI version of the Genetic Programming Environment for FIFTH (GPE5) on a high performance computing cluster. The implementation is based on the island model with each node running the GP algorithm asynchronously. In particular, we examine the effect of several configurable properties of the system including the ratio of migration to crossover, the migration cycle of programs between nodes, and the number of processors used. The problems employed in the study were selected from the fields of symbolic regression, finite algebra, and digital signal processing.
IEEE Military Communications Conference, 2003. MILCOM 2003., 2003
Rapid adaptability is a critical requirement for modern communication intelligence systems. This ... more Rapid adaptability is a critical requirement for modern communication intelligence systems. This paper explores the concept of combining model based design with a runtime framework to realize an easily reprogrammable signal analyzer for embedded COMINT systems. Signal processing experts design algorithms using modeling tools tailored to the domain. The tools produce output that can be loaded directly to the embedded system. Platform independence is achieved by using XML and CORBA as principal elements of the design. *
Proceedings of the 13th annual conference companion on Genetic and evolutionary computation - GECCO '11, 2011
Modeling to predict flame spread and fire growth is an active area of research in Fire Safety Eng... more Modeling to predict flame spread and fire growth is an active area of research in Fire Safety Engineering. A significant limitation to current approaches has been the lack of thermophysical material properties necessary for the simplified pyrolysis models embedded within the models. Researchers have worked to derive physical properties such as density, specific heat capacity, and thermal conductivity from data obtained using bench-scale fire tests such as Thermo-Gravimetric Analysis (TGA). While Genetic Algorithms (GA) have been successfully used to solve for constants in empirical models, it has been shown that the resulting parameters are not valid individually as material properties, especially for complex materials such as wood. This paper describes an alternate approach using Genetic Programming (GP) to automatically derive a mass loss model directly from TGA data.
A priori analysis and classification of digital communication signals is important both in the co... more A priori analysis and classification of digital communication signals is important both in the communication industry for developing new receiver technology and in the intelligence community for deciphering intercepted signals. In both domains, algorithm research focuses on extracting signal characteristics, such as modulation type and symbol rate, even when the received signal is distorted by noise and fading in the transmission channel. The lack of common standards for test files, data recording, and presentation of results make it difficult to compare the performance of algorithms from disparate sources. This work proposes an XML-based framework and an associated XML Schema as a solution to this standards problem. The XML Schema defines base types for communication signal constructs. These base types are then used to define two principal XML document formats: a signal library and an algorithm test record.
This paper demonstrates that FIFTH™, a new vector-based genetic programming (GP) language, can au... more This paper demonstrates that FIFTH™, a new vector-based genetic programming (GP) language, can automatically derive very effective signal processing algorithms directly from signal data. Using symbol rate estimation as an example, we compare the performance of a standard algorithm against an evolved algorithm. The evolved algorithm uses a novel approach in developing a symbol transition feature vector and achieves an impressive 97.7% overall accuracy in the defined problem domain, far exceeding the performance of the standard algorithm. These results suggest that vector based GP approaches could be useful in developing more expressive features for a large class of signal processing and classification problems.
Genetic Programming, 2007
FIFTH ™ , a new stack-based genetic programming language, efficiently expresses solutions to a la... more FIFTH ™ , a new stack-based genetic programming language, efficiently expresses solutions to a large class of feature recognition problems. This problem class includes mining time-series data, classification of multivariate data, image segmentation, and digital signal processing (DSP). FIFTH is based on FORTH principles. Key features of FIFTH are a single data stack for all data types and support for vectors and matrices as single stack elements. We demonstrate that the language characteristics allow simple and elegant representation of signal processing algorithms while maintaining the rules necessary to automatically evolve stack correct and control flow correct programs. FIFTH supports all essential program architecture constructs such as automatically defined functions, loops, branches, and variable storage. An XML configuration file provides easy selection from a rich set of operators, including domain specific functions such as the Fourier transform (FFT). The fully-distributed FIFTH environment (GPE5) uses CORBA for its underlying process communication.
Performance analysis of signal processing algorithms should yield insight into expected performan... more Performance analysis of signal processing algorithms should yield insight into expected performance as a function of all varying factors including algorithm parameters, signal characteristics, and transmission channel propagation effects. This paper presents a test framework for characterizing signal processing algorithm behavior. The framework provides functions to automate the evaluation process including creating large numbers of test files, running these files through the algorithm under test, collecting data, and analyzing the results. Automating this cycle allows rapid testing and evaluation of algorithm enhancements, as well as identifying the significant factors that affect performance. We demonstrate this technique by comparing and characterizing two different published symbol rate estimation algorithms.
For automated surveillance applications, estimating the symbol rate of an unknown digital communi... more For automated surveillance applications, estimating the symbol rate of an unknown digital communication signal is an important step in the analysis process. Several papers have investigated using the wavelet transform in symbol rate estimation algorithms. Due to its complexity, closed form analysis of performance is often limited, and simulations may not include practical factors such as carrier frequency offset or symbol pulse shaping. This paper uses an automated statistically based test framework to investigate the performance of the wavelet transform against PSK signals with parameters that span a realistic portion of the High Frequency (HF) signal space. The analysis identifies signal and algorithm parameters that affect performance. We also demonstrate that accurate metrics for estimating the probability of failure/success under realistic operating conditions are available for the db6 wavelet.
Uploads
Papers by Kenneth Holladay