Deep neural networks (DNNs) trained solely from data are known to suffer from instabilities. In the presence of some physical model of the observed system it is therefore natural to attempt to increase robustness by incorporating the prior knowledge.
Motivated by real problems in computational chemistry, in which the huge complexity of the chemical space may prevent the complete exploration of the space itself, this project aims at developing statistical learning tools to support chemists (but not only) in their research.
Entertainment and e-commerce websites produce abundant user-item interaction clicking data. This can be used for making recommendations to the users through recommender systems. Good quality recommendations are of value to both providers and users of items, helping users to find the items they are interested in.
Two factors are important when using a Deep Neural Network (DNN). First: the architecture of the DNN: determining layers, parameters, activation functions and loss function. Second: an optimisation method, such as SGD, Adam and RMSProp, which will chose the parameters of DNN.
In this project, we will bring DRL for AFC to the real world by implementing the first experimental demonstration of this technique.
The goal of the project is to make headway in the analysis of DNNs in the application to numerical methods for nonlinear hyperbolic partial differential equations (PDEs).