![]() ![]() This is done by using pairs of signal and background events during training 12, which effectively increases the number of training instances quadratically by using all possible combinations of events. A crucial part in this analysis workflow is to develop specialized algorithms that are well-suited for the data at hand.Īs the number of measured data points corresponding to an event including a Higgs boson (a signal event) is extremely small compared to background events, learning to rank techniques can be applied to increase the statistical variation of a data set. More information on the statistical methods applied in high energy particle physics can be found in the paper by Cranmer 11. This procedure reduces biased selection of data and p-hacking and has the further advantage that on simulated data it is known which event belongs to which process. Only when every part of classification chain is settled the methods are applied to blinded experimentally acquired data. In such experiments it is common practice to only consider the simulated data to fix the model. In order to analyze the experimental data, the number of measured processes, classified as signal, is compared to the expected number from the simulations. After training these machine learning methods on simulated data, these models can then be applied to experimentally acquired data and simulated data. events corresponding to the process of interest (in our case the Higgs Boson production via VBF), and those which have similar signature (for example t t ¯ in our case) 9, 10. ![]() ![]() In order to maximize the certainty with which the processes are classified, machine learning methods are trained on simulated data to classify signal and background events, i.e. This leads to the problem that single measurements of different processes can be identical, such that they cannot be distinguished. The data measured in the detector follows distinct, overlapping distributions for different physical processes. In practice, one only works on these simulated data in order to find ways to classify the process of interest from the detector data. Typically, this is achieved by first simulating the physical process via Monte-Carlo generators 6– 8. Since the measured data in the experiment are not labeled and different processes can look very similar in the detector data, it is crucial to find criteria for their distinction on simulated data, for which the corresponding process is known. Since this production channel is overwhelmingly dominated by background events, finding efficient ways to separate them from signal events is crucial. One approach is the precise measurement of properties of the newly discovered Higgs boson, for example through the measurements of Higgs production events via vector boson fusion (VBF) 5. An ongoing effort is therefore to establish limitations of the SM by investigating all of its parts thoroughly. the existence of dark matter in our universe 4. However, a few experimental observations are unexplained by the SM, e.g. The SM is currently the best description for physics at subatomic scales, and it explains most particle physics experiments of the past century. Assuming the observed particle is the Higgs boson predicted by the Standard Model of particle physics (SM), it would complete the SM to a self-consistent theory. In the summer of 2012, a possible candidate for the Higgs boson was discovered by the ATLAS 1 and CMS 2 experiments at the Large Hadron Collider (LHC) at the European Organization for Nuclear Research (CERN) 3. Furthermore, we use so-called transfer learning techniques to improve overall performance on different data types. In this work we compare our pairwise neural network algorithm, which is a combination of a convolutional neural network and the DirectRanker, with convolutional neural networks, multilayer perceptrons or boosted decision trees, which are commonly used algorithms in multiple Higgs production channels. This makes it robust to unbalanced data set scenarios and can improve the overall performance compared to pointwise models like the state-of-the-art boosted decision tree approach. The fact that training is then performed on pairwise comparisons of signal and background events can effectively increase the amount of training data due to the quadratic number of possible combinations. In this paper, we propose to use algorithms for learning to rank, i.e., to rank events into a sorting order, first signal, then background, instead of algorithms for the classification into two classes, for this task. ![]() To this end, machine learning techniques have been recently applied to processes like the Higgs production via vector-boson fusion. In the extensive search for new physics, the precise measurement of the Higgs boson continues to play an important role. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |