Classical and Quantum Machine Learning for Anomaly Detection at the CMS Experiment and Beyond

This thesis presents two novel methodologies for anomaly detection in particle collision events, utilising both classical and quantum machine learning techniques. The first approach employs an unsupervised, data-driven variational autoencoder (VAE), trained on collision data recorded by the CMS detector at the Large Hadron Collider during Run 2, with a center-of-mass energy of $\sqrt{s}=13\,\mathrm{TeV}$. Events containing at least two large-radius jets are analysed, and the VAE-derived anomaly score is subsequently decorrelated from the dijet invariant mass ($m_\mathrm{jj}$) through a novel machine learning-based approach. A systematic search for potential anomalies that deviate from the expected smoothly falling background distribution is performed in the $m_\mathrm{jj}$ range from $1.8$ to $6\,\mathrm{TeV}$. No significant excess above the SM expectation is observed, with the highest observed local significance being $2.3\sigma$ at $4.9\,\mathrm{TeV}$. This method is subsequently used to derive exclusion limits on a large number of exotic signal models derived from BSM physics.

The second approach introduces a novel quantum machine learning framework, named 1P1Q, for encoding jet kinematic features onto two-level quantum bits (qubits). ... mehrUsing this encoding strategy, a quantum autoencoder is trained to capture the underlying distribution of jets initiated by a light quark or gluon, and subsequently generate anomaly scores for exotic jets with a more complex jet substructure. The developed quantum data encoding is also validated in a jet classification task, for distinguishing jets arising from the decay of a top quark, from those initiated by a light quark or gluon. For both the anomaly detection and supervised classification tasks, the 1P1Q method is demonstrated to achieve signal identification and separation performance comparable to or surpassing current state-of-the-art classical machine learning algorithms.