MLOps on Unsupervised

Hello all! Just a curiosity question.

So far, in this specialization, i notice that MLops, TFX and all the materials are squarely focused on label quality, data drift and so on more from a Supervised learning perspective.

How about the same for deployment and management of anomaly detections based on unsupervised learning in production?

Any thoughts/discussions on that front ?

@spadmanabhan

well, as you have probably noticed, today most of the big results from ML are coming from Supervised Learning. That probably justifies the focus given in the course.

Having said that, I think that most of the material applies as well to data preparation and data monitoring for unsupervised learning. You have always to extract data, prepare, compute statistics, and monitor that in the time there is any not unnoticed data drift.

So, in my view, there are differences in the algorithms that you apply, but the concepts and tools used to structure the data pipelines apply to unsupervised learning as well.

1 Like

Thanks, @luigisaetta ! Makes sense to me. thanks!