We welcome Nathan Buskulic as a visiting researcher. Nathan finished his machine learning master at Sorbonne Université (France) with distinction and decided to continue his way towards academia. He spent one year of his master in Tu-Delft to experience an international research environment. His master thesis supervised by Dr Guillaume Auzias looked into how to apply and fine-tune multi-graph matching algorithms to neuroscientific data to find better markers of brain topology. He also did a research internship with Dr Carola Doerr to compare different evolutionary algorithms performances and empirically showed that some previously-thought optimal results were not.
Nathan’s research interests are in the field of Neural Architecture Search (NAS), Algorithm Selection, and Automated Machine Learning (AutoML). More specifically he is interested in finding ways to adapt NAS to work in an online setting so that the architecture of a neural network evolve as well as the weights of the network. The long-term goal is to be able to build a framework that bridges network weights together with its architecture to be able to make the two components evolve together during training time.
These questions are essentials as they would lower the technical knowledge needed to apply artificial intelligence and neural networks models in different contexts. Trying to find answers to these problems also help us deepen our theoretical understanding of some algorithms like neural networks that are not yet fully understood by scientists.