Active learning can be used to efficiently estimate non-parametric choice models, overcoming identifiability challenges that arise with offline data.
