Valid distribution-free inferential models for prediction



  1. Vladimir VovkApril 12th, 2020 at 05:17 pm

    Great work!  A few comments and speculations attached.  Volodya

  2. Ryan MartinApril 8th, 2020 at 02:21 pm

    @JonWilliams: Thanks for your comments!  An updated version of the manuscript has been posted that revised base in part on your helpful suggestions.


  3. Jonathan WilliamsFebruary 13th, 2020 at 08:24 pm

    Very exciting stuff!  Please see attached for some comments.



Add to the Conversation


A fundamental problem in statistics and machine learning is that of using observed data to predict future observations. This is particularly challenging for model-based approaches because often the goal is to carry out this prediction with no or minimal model assumptions. For example, the inferential model (IM) approach is attractive because it has certain validity guarantees, but requires specification of a parametric model. Here we show that a new perspective on a recently developed generalized IM approach can be applied to construct an IM for prediction that satisfies the desirable validity guarantees without specification of a model. One important special case of this approach corresponds to the powerful conformal prediction framework and, consequently, the desirable properties of conformal prediction follow immediately from the general IM validity theory. Several numerical examples are presented to illustrate the theory and highlight the method's performance and flexibility.


➤  Version 1 (2020-01-23)


Leonardo Cella and Ryan Martin (2020). Valid distribution-free inferential models for prediction. Researchers.One,

© 2018-2020 Researchers.One