The real chink in the real-time ML lifecycle is….?
It is you and me
Undoubtedly, Real-Time Machine Learning (RTML) models are extremely valuable.
For real-time machine learning to succeed though, the enabling platforms need to be credible, scalable, quick, and cost-effective.
If we carefully analysed the RTML lifecycle, they would be:
- data ingestion
- data filtration through the pipelines
- feature extraction and analysis
- identifying the right model
- making the model available
- monitoring and deploying the model
- publishing the model output
- retraining the model
These are just data-related activities. There is an equivalent amount of complex activities in the infrastructure layer too.
The above lifecycle indicates that for RTML to stay true to its intent of being real-time, credible, quick, and cost-effective, the lifecycle should have no onerous human tasks — which may be the case in many enterprises.