At first, I really can’t understand, why Advanced labeling is optional in week 4. At second I can’t understand the main difference between described in video Active Learning and Semi-supervised learning.
Please share a link of the part of the course you would like clarifications
Could you clarify on whether you haven’t understood the difference between active learning and semi-supervised learning, or the differences between the methods in each video?
I’m not a part of the curiculum design but can imagine that in most cases, the ML team would limit the dataset to existing labeled data or just outsource the labeling. Therefore, the advanced techniques introduced in week 4 could either be too ahead of its time or not needed in conventional settings.
I don’t fully understand what are differences between choices that you gave me. But I think I doesn’t understand both.
Let’s say you want to teach somebody to distinguish between two types of birds. Ideally, you would show them examples where the two species look the most alike (the difficult examples), so the learner can quickly understand the fundamental differences. This process is similar to active learning, where humans do the labeling, but the most significant data points are prioritized first.
On the other hand, let’s imagine that you have a million photos of two species of birds, but only 50 of them are labeled. You decide to group the photos based on similarities (this process can be automated), and assign those groups with whichever label is the most prevalent among them. This is a form of semi-supervised learning, where humans don’t necessarily come up with 100% of the label.
How do these analogies do for you? It’ll help us to answer further if we can pin-point whatever you’re trying to understand.
so, in first example we have an all data labeled and we have a choice what examples are most demonstrative. And what do we do next?
on second example about the semi-supervised learning we have very dedicated labels. And we learn NN on this 50 examples or create custom function that will create us two approximate groups. And from this two groups we will try to find most demonstrative examples as in active learning. Correct?