Machine Learning, statistics, (AI?)
Physics is an experimental science. Hadron spectroscopy is not different from that driven historically mostly by experimental observations. Plenty of modern experiments around the globe (at CERN, GSI, ELSA, Jefferson laboratory, BES, ...) provide us these days with an abundance of new high precision data, driven by multiple systemic and technological factors.
The question is then, what can one do with these data? Can one for example build a global hadronic model describing that data, filling in possible gaps in the measurement, remove possible misidentified results and ultimately allow us to extract universal parameters of excited states of matter? These are some of the bird's view questions we are aiming to answer in the JBW collaboration.
Data driven approaches run a danger of over-fitting or under-fitting the provided data. Another issue is the treatment of outliers and model misspecifications or other biases. Such problems are well known beyond physics and plenty of modern statistical or machine learning tools exists to mitigate them. One example of Neural Network based approach is shown in the right figure from a recent talk: "Denoising and sharpening the hadron spectrum through machine learning" at the recent dedicated workshop[link] in Thessaloniki.