My research agenda is dual. I am interested in both research meant to understand the learning barriers in the computing disciplines, and research on improving the adaptive abilities of computing systems by leveraging evolutionary principles.
I have spent about 10+ years on each of these research agendas & I am now balancing my efforts to pursue both in parallel. It is an interesting journey around the theme of learning, whether it occurs in natural or artificial systems.
You may be able to find most of my papers using the following;
Since around 2002, I have increased my researh focus on the still relatively young discipline of Computing Education Research - CER.
This field enabled me to approach teaching from the research perspective; basing interventions in educational theories or frameworks while applying formal evaluations methods to guide improvements.
This work led me to become director of the USF Computing Education Research & Adult Learning group which I founded around 2007. The group's objectives are to contribute to the body of knowledge on how novices acquire various computing skills, while also sharing & evaluating innovative pedagogies, course material, or software tools meant to help students overcome learning barriers which are specific to our discipline.
For details on the various NSF-funded projects this group is working on, please refer to the CEReAL website;
I started my research on Evolutionary Algorithms when I met my Ph.D. advisor, Philippe Collard, during my sophomore year. To this day, I am still fascinated by the study of these simple yet powerful bio-inspired meta-heuristics
Since I have not yet prepared a site for my EA research, I will try to provide a quick overview below;
Originally, Evolutionary Algorithms had a tendency to converge toward a global - sometimes even local - optimum in a fitness landscape. This limited considerably their adaptive qualities in so far that, if the environment happened to be modified after their convergence, most of their evolutionary dynamics was incapacitated thus preventing them from further adapting
Nowadays, research on improving EAs so they may continuously adapt to such environments is very active. I did my Ph.D. on this topics and it guided most of the contributions listed below. I am particularly interested in how the need to preserve diversity in populations makes good Time-Dependent EAs also potentially better at dealing with notoriously hard fitness landscape, noisy fitness evaluations, multi-objective optimization problems or even co-evolution. The latter has actually since then be shown to be directly linked to multi-objective optimization.
During my postdoc, I applied EAs to a Pittsburgh Style Learning Classifier System to evaluate their ability to handle Time-Dependent Learning environments - TDL. Results were encouraging and revealed that TDL features both the extrinsic dynamic of TDO problems with the inherently stochastic nature of classifiers evaluation based on a partial sample of all possible inputs. This made TDL problems even more challenging, & therefore interesting, as benchmarks to measure EAs adaptive qualities
This is actually where my pre-Ph.D. work started. My advisor was working then on a so-called Dual Genetic Algorithm which employed a gene expression mechanism to allow the underlying EA to also evolve its own perception of the fitness landscape.
We then extended the meta-genes model to introduce a Folding Genetic Algorithm able to evolve both the genes encoding the solution alongside the meta-genes influencing the representation of the fitness landscape.
As I started my Ph.D. work, the both DGA & FGA turned out to be particularly suited in Time-Dependent Optimization problems.
The success of the above algorithms motivated the exploration of EAs which are not only able to react efficiently to the environment being modified, but are also able to acquire experience when the variations are repeated over time
Inspiration came for the then brand new Artificial Immune Systems which led me to propose two immune-inspired evolutionary algorithms able to improve their performance when dealing with repetitive time-dependent problems;
Results on TDO problems motivated trying YASAIS on TDL problems which led to the Pittsburgh Immune Classifier Systems - PICS - & what I termed cognitive immunization. To my knowledge, this was the first instance of cognitive immunization to a varying environment. If you know of previous work, please drop me a line.
Rather than post a comprehensive list of publications, I decided to pick 3 papers for each of my areas of research to illustrate the kind of work I have done.