White Paper

Osterman Research: Improve Machine Learning Model Training with Collaborative Secure Learning 

Collaborative secure learning enables training data to be shared across
organizations in order to train machine learning models and improve predictive quality while preserving the privacy of each separate training data set. The approach relies on sharing fully encrypted training data sets, where the encryption remains in force throughout the entire sharing lifecycle—even during use, computation, and model building. Consequently, no participant can decipher identifiable characteristics in the training data sets that would compromise privacy mandates, offer an unfair competitive advantage, or fall afoul of regulatory requirements.

In this white paper, we explore model training through collaborative secure learning and provide three examples of recent combinations in three industries that highlight the benefits of collaborative secure learning.

Key takeaways:

• The data is available, but barriers stop many collaborative endeavors
• Relying on legal agreements is insufficient
• Collaborative secure learning offers a new approach
• Evidence is mounting across industries that collaborative secure learning
  offers substantial benefits