Benchmark of Deep Learning models

In the field of high-energy physics, software performance is critical due to the high CPU and I/O costs of processing and analyzing billions of events. Running benchmarks is therefore essential as it provides consistent performance metrics that can be tracked over time. Aim of this project was to develop benchmark code to evaluate the CPU and GPU performance for inference of some typical deep learning models used by the LHC experiments. Benchmark tests for Convolutional neural network (CNN) and Recurrent neural network (RNN) for both TMVA and Keras were created and are to be included in ROOTBench. Keras outperforms TMVA in terms of CPU performance for inference in both CNN and RNN. In terms of GPU performance, TMVA surpasses Keras in both CNN and RNN benchmarks.