npond will work on maintaining the central FTAG training framework used for GNN: Salt. He will take part in code review of merge requests and extend the continuous integration pipeline with unit tests and integration tests, and write documentation for SALT, extending the current documentation and contribute in the design of training resources for newcomers involved in the algorithm development work for the boosted bb/cc tagger.
While maintaining the framework, the following tasks will be performed:
1 Contribute to the training of the GN1+ tagger, which is the target tagger for rel22. This is an open objective and can involve R&D, including studies of other auxiliary tasks, such as regression of performance observables and studies of different architectures. He student should contribute to training and deployment of the GN1+ tagger.
2. Study the performance of taggers in mc21 (Run-3 MC), comparing the efficiency and rejection in different MC samples, with the aim of understanding the efficiency drop observed in mc21 compared to mc20. Heshould compare input features used by the algorithms. Once the issue is understood, he can contribute in the study of strategies to mitigate the performance loss. Once mc22 becomes available, Nikita will study the tagger performance in mc22.
3. Study the tagger performance in different generators for ttbar samples..
All the above tasks will be carried out using the central FTAG workflows with clear documentation. The work will be presented in the Algorithm sub-group meeting and Software meeting. The technical details will be documented in an internal note.
532770 Performance Studies - Flavour Tagging
556629 Performance Studies for Flavour Tagging