ML as a Service (MLaaS)

Machine learning pipeline

The machine learning pipeline consists of many repetitive steps, for example, data preprocessing, model training/retraining, and model inference. Serverless architecture can be utilized in many of these steps for different tasks. For example:

• For servicing pull and push data/objects from the backend or data buckets (for example, AWS Amazon S3).

• For building APIs to transform and clean data (the data processing stage of ML).

• For new training or retraining in scenarios where conditions for concept driftingIn machine learning, model prediction accuracy may change with time due to minor drifts in physical phenomenon or models. Consequently, it causes change(s) in the statistical properties of the target variable over time in unexpected ways. Retraining with new training data is one trivial solution to this problem. are met.

• For batch/ensemble predictions.