Privacy Preserving Machine Learning Challenges and Solution Approach for Training Data in ERP Systems

Abstract

Listen

Translate article

The digital transformation is ubiquitous and pushing the case further for ERP companies to incorporate more machine learning algorithms in order to drive intelligent real-time decision-making capabilities. ERP systems have started incorporating machine learning use cases powered by the huge enterprise data, cloud and compute capabilities. However, privacy of data remains a challenge. Data privacy is at the core of a machine learning model that is trained on sensitive information. Not just for profit businesses, but even academic endeavors in the field of medicine cannot progress if they can’t access sensitive medical information in a privacy preserved format. Ramifications of applying a ML model without even fully understanding what is happening inside its hidden layers can be disastrous and the resulting perils can lead to legal consequences. Therefore, Privacy preserving AI techniques started evolving in last few years. The privacy preserving AI field is still growing and there is an understanding gap in organizations and individuals, which makes privacy breach or compromise a pervasive business challenge. This paper focuses on what are key challenges for ERP companies as far as Training machine learning models on their enterprise data is concerned. And how can these challenges be overcome by applying data anonymization and differential privacy techniques.