Hi my name is Konstantin Nikolaou and I recently started as a doctoral student here, at the ICP. After finishing my master’s thesis at the ICP on a similar topic, I will now continue my research on neural network theory.
Until today, it has not been fully understood why neural networks are such powerful tools and how to properly interpret them. I will investigate this by combining neural network theory with an approach from statistical mechanics: I will apply the concept of Collective variables to neural networks.
The collective variables I investigate are scalar quantities, which allow you to look into a neural network. We can use them to study inner processes during learning to understand more about how they fundamentally work. With neural networks as a great example, I will later extend my work to a broader range of learning systems.