Recurrent Neural Networks and LSTMs on GPUs for Particle Physics Applications

Toolkit for Multivariate Analysis (TMVA) is a machine learning toolkit for the ROOT scientific software framework used in many particle physics data analysis and applications. The CNNs and DNNs has been proven in the variety of applications like classification, tracking of particles etc. The aim of the project is to expand the current library of TMVA DNN by implementing efficient Recurrent Neural Networks and LSTM Networks and get the production ready GPU version of convolutional deep learning library along with support for GPU training.

GPUs are much more effective in terms of high performance when compared with CPUs. In this project, CUDA (Compute Unified Device Architecture) technology will be used by NVIDIA for proper implementation of RNN and LSTM to support GPU architecture.