Speaker
Description
ML-based event reconstruction is widely used for various reconstruction tasks for colliders, such as particle flow and jet flavor identification. Transformer-based algorithms are especially powerful for tasks where relation between objects/particles are essentially important. We have implemented the jet flavor tagging tool based on Particle Transformer for ILD full simulation. Recent works include optimization of training procedure with variable hyperparameters, investigating dependency of samples and parameters, pretraining with fast simulation etc. In addition, transformer-based jet clustering combined with flavor ID is being developed in the similar context, with consideration of extention to a foundation model. We will present these activities in this talk.