NanoAI welcomes Vasilis Sioros who will present his work, titled “Generating realistic nanorough surfaces via a Generative Adversarial Network”.
As always, an open discussion will follow.
When? May 6th, 15:00 Greece
Description: This work examines the possibility of providing multi-physics simulations with a computationally inexpensive way of integrating new nanorough surfaces, similar to the ones being measured.Modeling nanorough surfaces requires: (1) identifying the structural feature space so that the generation of new nanorough surfaces is possible and (2) the nanorough surface reconstruction process to be property-preserving, meaning that newly constructed nanorough surfaces should showcase structural properties similar to the ones being modeled. In this work, we will be designing a data-driven approach to constructing nanostructures with a pre-determined set of (explicit or implicit) structural properties.In order to achieve this, we will be fitting a supervised learning model, capable of learning the stochastic nature of the morphology of nanorough surfaces, with no a priori knowledge of their underlying characteristics, to a data set of appropriately generated synthetic nanorough surfaces.We will be examining how a Generative Adversarial Network (GAN) based framework can be trained to generate realistic nanorough surfaces.Additionally, we will also evaluate if and by what margin n-gram-graph-based similarity metrics can be utilized to improve the existing GAN architecture further.Finally, we will be exploring some limitations of this framework regarding different degrees of correlation, stochasticity, and smoothness of the input nanorough surfaces.
Vasilis Sioros is a software engineer and an ML and NLP enthusiast. He is also a recent Computer Science graduate of the National Kapodistrian University of Athens.
Subscribe to our mailing list: https://lists.iit.demokritos.gr/mailman/listinfo/nanoai
You can find more info on this upcoming talk as well as on our previous talks at http://nano-ai.iit.demokritos.gr/