Last week several students in my laboratory completed their Master and PhD defenses. Congratulations to all of them! In this post I will summarize their graduation works:
Master Students
- Luiz FSE Santos: "Modeling STDP-Driven Neuronal Activity and Connectivity Dynamics in Recurrent Spiking Neural Networks"
Luiz' thesis proposed a new model for spiking neural network based on his observations of the behavior of neurons in vitro. This new model makes the simulated networks behave more like the natural ones, while still being able to be trained and used for computational task. The hope is that this line of research can eventually lead us to find computational models of neural networks that are as efficient as the biological ones (instead of the extremely inneficient neural network models we have today).
- Gwenole Moisson: "Evolution of self-organizing patterns in Lenia using CPPNs".
Lenia is a real-valued generalization of Conway's Game of Life, a cellular automata that has captured the imagination of many a computer scientist over the year. Gwenole's thesis ask the question of whether it is possible to show that Lenia, like the Game of Life, is Turing Complete. This question is made difficult by the continuous nature of Lenia, and the lack of existing theory about it. Gwenole approaches it from an experimental point of view, using the evolution of CPPNs as a tool to generate Lenia patterns that could be used as logic gates.
- Yuki Kaneshika: "Analysis of Event Design on Genre Development in Music Distribution Ecosystems using Multi-Agent Simulation".
Kaneshika's Thesis is interested in using Multi Agent Systemn to explore how the music scene evolves and reacts to the organization of large scale events. His work asks the questions of what incentives and policies can promote (or hinder) the development of new kinds of music, as opposed to situations where everyone just copies the same types of music over and over again. His model includes music creators, music listeners and event promoters. Events are ways that connect creators and listeners. These events may cause creators and listeners to cluster into a small number of "winning" musical genres, or introduce listeners and creators to more fringe types of music.
PhD Students
- Jair Pereira: "Similarity Analysis of Optimization Metaheuristics".
Jair's PhD thesis studies the analysis of the behavior of Metaheuristic algorithms through the lenses of their optimization behavior. Up until very recently, this kind of analysis was done by directly comparing the structure of the algorithms. In this context, Jair's thesis proposes a different way to see the problem: Metaheuristic Algorithms are characterized by a list of performance metrics obtained from observing the optimization of a set of selected benchmark problems. This defines a "Performance Profile", which can be used in comparative analysis across algorithms and problems. The key idea of the thesis is to offer a complementary method to the theoretical comparison of optimization algorithms, and to show that the Performance Profile has explanatory power.