Speaker
Description
Creating electrical circuit elements from one atom or molecule is one of the main issues in current molecular electronics research. Nowadays, investigation of conductance values of a single molecule can be realised at high mechanical stability by the mechanically controlled break junction (MCBJ) technique.
Among the high amount of conductance traces generated by break junction measurements, some curves reflecting the presence of molecules can be found. One of the fundamental tasks is to filter out the molecular curves from the measured data. To do that, an automatized algorithm is needed due to the high number of measured curves. Common methods for data classification and for quick data processing related to this problem are algorithms using complex criteria to filter curves showing molecular characteristics.
My target in this research project is to realize a neural network-based classification, which constructs its transmission based on earlier measurements of conductance curves. During my research, I aimed to create a simple and clear model to classify conductance curves directly in the IGOR-based experimental data handling software. In my work, I consider the choice of model parameters and the issues of generalisation.
Utilizing the simplicity of applied neural network I would also like to illustrate the information learned by the model. With analysis of the weights of the corresponding neural networks, I would also like to get insight into decision-making considerations of the network.
