Date of Award

2019

Publication Type

Doctoral Thesis

Degree Name

Ph.D.

Department

Computer Science

First Advisor

Robin Gras

Keywords

Autoencoders, Combinatorial Optimization, Deep Learning, Estimation of Distribution Algorithms, Individual Based Modeling, Variational Autoencoders

Rights

info:eu-repo/semantics/openAccess

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Abstract

Individual based modeling provides a bottom up approach wherein interactions give rise to high-level phenomena in patterns equivalent to those found in nature. This method generates an immense amount of data through artificial simulation and can be made tractable by machine learning where multidimensional data is optimized and transformed. Using individual based modeling platform known as EcoSim, we modeled the abilities of elitist sexual selection and communication of fear. Data received from these experiments was reduced in dimension through use of a novel algorithm proposed by us: Variational Autoencoder based Estimation of Distribution Algorithms with Population Queue and Adaptive Variance Scaling (VAE-EDA-Q AVS). We constructed a novel Estimation of Distribution Algorithm (EDA) by extending generative models known as variational autoencoders (VAE). VAE-EDA-Q, proposed by us, smooths the data generation process using an iteratively updated queue (Q) of populations. Adaptive Variance Scaling (AVS) dynamically updates the variance at which models are sampled based on fitness. The combination of VAE-EDA-Q with AVS demonstrates high computational efficiency and requires few fitness evaluations. We extended VAE-EDA-Q AVS to act as a feature reducing wrapper method in conjunction with C4.5 Decision trees to reduce the dimensionality of data. The relationship between sexual selection, random selection, and speciation is a contested topic. Supporting evidence suggests sexual selection to drive speciation. Opposing evidence contends either a negative or absence of correlation to exist. We utilized EcoSim to model elitist and random mate selection. Our results demonstrated a significantly lower speciation rate, a significantly lower extinction rate, and a significantly higher turnover rate for sexual selection groups. Species diversification was found to display no significant difference. The relationship between communication and foraging behavior similarly features opposing hypotheses in claim of both increases and decreases of foraging behavior in response to alarm communication. Through modeling with EcoSim, we found alarm communication to decrease foraging activity in most cases, yet gradually increase foraging activity in some other cases. Furthermore, we found both outcomes resulting from alarm communication to increase fitness as compared to non-communication.

Share

COinS