How continuous learning differs from active learning?

These are two different concepts.

Continuous Learning

• refers only to the fact that model can learn continuously from incoming stream of data even after deployment.
• continous learning can refer to any type of machine learning in which model continues to learn after deployment (supervised, semisupervised, unsupervised, reinforcement learning).
• Example of continuous learning: purchase data stream for a recommendation system. All the data is labeled (bought item/didn't buy item) and so there are no unlabeled samples to choose from.

Active Learning

• model can select unlabeled samples which it is most unsure about.
• is a form of semi-supervised learning.
• uses an oracle (e.g. human annotator) for labeling selected data samples.
• can but doesn't have to be implemented in a continuous learning fashion.
• Example of active learning: you have a dataset of labeled and unlabeled samples. You measure prediction certainty and select unlabeled samples with the lowest prediction certainty and pass them on to human annotators. The model then learns from these additional samples and money is saved because only "difficult" samples were selected and manually annotated. This model may or may not be already deployed.