Improving the search for new physics and the identification of electrons using machine learning at the ATLAS experiment
Bibliographic record
Abstract
The study of high-energy collisions by the ATLAS experiment at the Large Hadron Collider (LHC) is essential to test the validity of the Standard Model of particle physics (SM), the current theoretical framework that describes the fundamental particles and how they interact, as well as to constrain its possible extensions. In light of its third data-taking period and the next generation of accelerators, the ATLAS experiment faces challenges associated with the high-dimensionality of the recorded signals and the large amount of data still left unexplored. In this context, the use of deep learning techniques has great potential to enhance the performance of the classification of the physics objects that originate from these signals, as well as to provide new tools to perform fast statistical inference from the data. This thesis presents applications of deep learning techniques to improve the electron identification algorithm performance, as well as a new strategy to search for resonances in invariant mass distributions with the ATLAS experiment. Firstly, new measurements of the efficiency of the current electron identification algorithm are presented, using data recorded at the beginning of Run 3, as well as the reprocessing of the Run 2 data with the new version of the ATLAS software. A small reduction on the discrepancies between the values obtained from MC simulated events and the data is observed, a consequence of the improvements made to the ATLAS software at the start of Run 3. Next, the development of a new electron identification algorithm is presented, where low-level detector information is processed in the form of images via a convolutional neural network. A study of the importance of its input features shows the relevance of all the current inputs considered. Furthermore, the rejection of the larger background class is decreased when a model trained with examples from simulated events is used to reject those obtained from experimental data. We show this rejection power is recovered if these data examples are incorporated into the training. Lastly, a novel strategy to search for resonances in invariant mass histograms is presented. It uses a neural network to predict the local statistical significance of resonances from its bin entries. The implementation of this method using realistic simulation data shows good results, with the prediction of the maximum significance within a histogram having no bias and a small variance. Work towards an implementation of this method within the ATLAS experiment is also presented, including the production of invariant mass histograms using ATLAS simulation data.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.003 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".