[Review] Sustainable MLOps: Trends and Challenges

Posted on One min read

Paper

Sustainable MLOps: Trends and Challenges provides some high-level, initial exploration of the concepts of sustainability in the contect of MLOps and AI software.

Data Points

  • 75% of Data Scientists are not computer scientists by training
  • 88% of Data Science teams are short of Data Engineers by a factor of 2

Relevant Concepts

MLOps consists of:

  • data ingestion/transportation
  • data transformation
  • continuous ML training
  • continuous ML deployment
  • output production/presentation to end-user

While many new tools being created, increased complexity (number of pieces or abstractions) yields decreased sustainability where sustainability could be:

  • economic - i.e. creates economic value
  • technical - i.e. enables ability to cope with changes
  • environmental - i.e. avoids harm to environment it is in
  • social - i.e. provides fair exchange of information between parties