Generative Adversarial Networks for Particle Physics Applications

Deep Learning Networks have proven to perform well for a wide range of problems, especially those requiring large labelled dataset to learn patterns. A new algorithm that has taken the Deep Learning research community by a storm is that of Generative Adversarial Networks (GANs) introduced by Ian Goodfellow in 2015. This algorithm has a lot of potential owing to the fact that it can generate data that is quite similar to the data given for learning in addition to faster generation of samples when compared to fully visible belief nets. It therefore makes GANs immensely useful in simulation of particle physics and astrophysical data.

Since ROOT is a data analysis tool extensively used for applications in particle physics and features a dedicated machine learning submodule, Toolkit for Multivariate Analysis (TMVA), it is essential to include a GAN implementation in the toolkit.

My project would focus on integrating an optimized GAN implementation in the TMVA DNN library with the help of already existing implementations of Deep Network Models. It would also involve enabling GPU Implementation of GANs using Nvidia’s CUDA library.