We are developing fast-simulation tools based on machine learning — rather than primarily using classical Monte Carlo — to simulate particle transport in the detectors of the LHC experiments. Such tools could play a significant role in helping the research community to cope with the LHC’s increasing computing demands.
The tools we are developing should be able to simulate a large variety of particles, energies, and detectors — all in a fraction of the time needed for classical simulation of particle transport. Our final objective is to integrate our tool in the existing code. This work is being carried out in collaboration with SURFsara and Cineca, as well as with Intel.
Over half of the WLCG’s computing workload is the result of a single activity, namely detector simulation. A single code, called Geant4, is used for this. Optimising this code has the potential to significantly reduce computing requirements, thus unlocking resources for other tasks.
Fast-simulation techniques have been developed in the past. However, the latest developments in machine learning (particularly in relation to deep neural networks) make it possible to develop fast-simulation techniques that are both more flexible and more accurate than existing ones.
Training time has turned out to be a major bottleneck for the meta-optimisation of our generative adversarial network. This includes not only the network weight, but also its architecture and the convergence parameters. Much of our work in 2018 concentrated on addressing this. We implemented distributed versions of our training code both on GPUs and CPUs, and we tested their performance and scalability in different environments (HPC clusters and cloud). The results are very encouraging: we observed almost linear speedup as the number of processors increased, with very limited or no degradation in results.
The other main area of work in 2018 related to the extension of the fast simulation tool to incorporate a larger set of kinematic conditions. We successfully extended the parameters related to incoming particles, integrating the angle of impact in the conditioning parameters. The tool is now mature enough to start planning its test integration with a classical Monte-Carlo code, such as Geant4.
We plan to continue improving the accuracy of the simulation, with particular attention to the tails of particle showers and single-cell energy distribution. We will also continue to investigate HPC training and explore various framework for hyper-parameter training. Finally, we will extend the simulation tool to different detectors, and collaborate on its integration into the existing simulation frameworks.
- F. Carminati, Quantum Computing for High Energy Physics Applications (21 February). Presented at the PhD Course on Quantum Computing at University of Pavia, Pavia, 2019. http://cern.ch/go/Z8Tx
- F. Carminati, S. Vallecorsa, G. Khattak, 3D convolutional GAN for fast simulation (5 March). Presented at IXPUG Spring Conference, Bologna, 2018. http://cern.ch/go/9TqS
- F. Carminati, G. Khattak, S. Vallecorsa, Three-dimensional energy parametrized adversarial networks for electromagnetic shower simulation (7 October). Presented at 2018 IEEE International Conference on Image Processing, Athens, 2018. http://cern.ch/go/lVr8
- F. Carminati, V. Codreanu, G. Khattak, H. Pabst, D. Podareanu, V. Saletore, S. Vallecorsa, Fast Simulation with Generative Adversarial Networks (12 November). Presented at The International Conference for High Performance Computing, Networking, Storage, and Analysis, Dallas, 2018. http://cern.ch/go/Z6Wg
- F. Carminati, G. Khattak, D. Moise, S. Vallecorsa, Data-parallel Training of Generative Adversarial Networks on HPC Systems for HEP Simulations (18 December). Presented at 25th IEEE International Conference on High Performance Computing, Data, and Analytics, HiPC, Bengaluru, 2018.