Speaker
Description
Modern high energy nuclear and particle physics experiments produce large amounts of sparse data that must be reduced significantly in size before performing physics analysis. Recently, there has been an effort to evaluate Foundation Models as a potential tool for performing data analysis. However, applying this capability to experimental particle physics is challenging due to the sparse, spatially distributed nature of detector data, which differs dramatically from natural language used to train large language models. The FM4NPP project has developed a self supervised training method for time projection chamber data at the sPHENIX experiment. The FM4NPP model demonstrates neural scalability with up to 188 million parameters and outperforms other models for several physics reconstruction-motivated downstream tasks. Furthermore, the frozen foundation model weights improve the performance of these downstream tasks when compared to labeled training alone. This talk will discuss the FM4NPP model, its performance, and future research directions.