Towards Capturing Uncertainty in FastLAS Predictions using Subjective Logic

Abstract Policy-based ensembles have been proposed as a method for combining model predictions in a coalition envi- ronment. Combination policies can be defined so that model predictions are weighted by the uncertainty of the prediction. This paper builds upon the recent #predict mode of a symbolic learner, called FastLAS, and defines a notion of uncertainty of FastLAS predictions using subjective logic. This notion enables learned FastLAS models to be used within a hybrid policy-based ensemble alongside other machine learning models.
  • Daniel Cunnington (IBM UK)
  • Yaniv Aspis (Imperial)
  • Mark Law (Imperial)
  • Alessandra Russo (Imperial)
  • Krysia Broda (Imperial)
  • Jorge Lobo (Imperial)
  • Elisa Bertino (Purdue)
  • Dinesh Verma (IBM US)
Date Sep-2020
Venue 4th Annual Fall Meeting of the DAIS ITA, 2020