For the past four decades of quant research, building a model meant first selecting the model SDE, and then calibrating its parameters
Today we are on the verge of transition to a new paradigm where model selection and model calibration are replaced by a single step - model learning
Model learning is much more than advanced interpolation - I will show that ML is able to reason about the interest rate and credit spread data in a highly sophisticated way rather than merely interpolate it
Even with the field still in its infancy, ML can already outperform the traditional techniques in important practical applications