Double Descent and Overparameterization in Particle Physics Data
This is an incremental application of an existing ML concept to a new domain, potentially aiding particle physicists in model selection.
The paper demonstrates the double descent phenomenon and benefits of overparameterization for the first time in particle physics data, exploring conditions where it leads to improved generalization error.
Recently, the benefit of heavily overparameterized models has been observed in machine learning tasks: models with enough capacity to easily cross the \emph{interpolation threshold} improve in generalization error compared to the classical bias-variance tradeoff regime. We demonstrate this behavior for the first time in particle physics data and explore when and where `double descent' appears and under which circumstances overparameterization results in a performance gain.