Interpreting the features for domain experts Using WIDE learning

Snehasis Banerjee, Tanushyam Chattopadhyay, Ayan Mukherjee

Jun 21, 2017 (modified: Jun 21, 2017) ICML 2017 WHI Submission readers: everyone
  • Abstract: In this paper, authors have proposed an interpretable feature recommendation method for solving sensor signal analytics problem in machine maintenance domain. The basic Wide learning based architecture for feature recommendation is out of the scope of discussion in this paper and authors have emphasized on the interpretation of the recommended features and how this human in loop interpretation system can can be used as a prescriptive system. The proposed system was deployed in solving a regression problem for one internal data set of machine maintenance record, as well as a prescriptive system on the popular bearing data-set from NASA prognostic repository. The proposed system is also used to analyze the causality of a machine maintenance problem.
  • TL;DR: shows how features can be interpretated based on wide learning approach and domain experts make sense of them in prognostics
  • Keywords: feature interpretation, wide learning, prognostics