MLSEC-07: Keep only relevant data

Preserve data across computing environments (such as development and staging) and only store use-case relevant data to reduce data exposure risks. Implement mechanisms to enforce a lifecycle management process across the data. Decide when to automatically remove stale data.

Implementation plan

• Establish a data lifecycle plan - Understand usage patterns and requirements for debugging and operational tasks. Establish a data lifecycle plan to reduce data sprawl over time.

• Design for privacy - Remove sensitive elements that are not needed for the ML workflow. Detect and redact personally identifiable information (PII), while maintaining data usability. Determine what features are required to solve the business problem and valuable for future iterations.  

Documents

• Reference Guide: Extract More Value from your Data

Blogs

• Building a data analytics practice across the data lifecycle

• Detecting and redacting PII using Amazon Comprehend

• Now available in Amazon Transcribe: Automatic Redaction of Personally Identifiable Information

• Machine learning models that act on encrypted data

• Redacting sensitive information with user-defined functions in Amazon Athena

Videos

• AWS re:Invent 2020: Privacy-preserving machine learning

• AWS re:Invent 2019: Best practices for Amazon S3

Examples

• Field Notes: Redacting Personal Data from Connected Cars Using Amazon Rekognition

• How to Create a Modern CPG Data Architecture with Data Mesh