Predicting Model Failures in Production

Abstract: 

ML Models in production lose accuracy over time. A number of factors contribute to this change, like Demographic Mix, Consumer Behavior change, etc. Using these models' output results in incorrect decisions that could lead to catastrophic failures for the organization. This existing whitespace calls for a solution(s) to help Data Science teams predict the failure in advance. We at Tredence have developed a suite of libraries which are able to predict model accuracy drop & trigger alerts to proactively fix the model. All of these libraries are packaged in the E2E model management accelerator – ML Works.

Bio: 

Aravind heads the Data Science Org at Tredence, and his team works on the R&D & algorithm development for new Data Science solutions