JETNET: ML-based predictor for Event Classifier of JETs using 2015 CMS Open Data

An essential part of new physics searches at the Large Hadron Collider (LHC) at CERN involves event classification, or distinguishing potential signal events from those coming from background processes. To address this challenge, a rich research and development program is ongoing, proposing new tools, techniques, and approaches. A machine learning (ML) technique is one of the best options to demonstrate how novel workflows can scale to analysis needs at the high luminosity (HL)-LHC. Machine learning algorithms are capable of not only training on high-level features, but of performing feature extraction. In these experiments, we developed JETNET and analyzed for its performance over 2015 CMS Open Data from IRIS-HEP repository. Also, we used 5 samples, all part of the 2015 CMS Open Data release. Specifically, we apply machine learning algorithms to the Analysis Grand Challenge (AGC) project. We trained different neural network (NN) architecture (i.e. number of hidden neurons, activation function, learning rate, optimizer, and loss function) to understand if it is possible to achieve better accuracy. It is found that JETNET: ML-based per-I/O latency predictor capable of achieving >90% inference accuracy and 42μs inference latency for each event. By doing this research, we demonstrate the power of this approach in the context of a physics search and offer solutions to some of the inherent challenges, such as decorrelating the classifier from the search observable.