[MUSIC] Now that we've gone through even more
material on clustering, let's spend a little bit of time describing what
we didn't cover in this course. So in this course, we tried to focus on the methods that
we believe are the most practical and most widely used tools out there for
performing retrieval and clustering. But there were still some other
important topics that we didn't cover in this course. So for example, in retrieval, there are lots of other distance
metrics that we could have described. We listed some of these in a module, but we didn't actually describe
these in any detail. Another thing we didn't cover was
something called distance metrics learning where these are procedures that
allow you to actually learn metrics that are useful for
the task at hand. Then for clustering, we didn't talk about
things like nonparametric clustering. Actually our hierarchical
clustering methods can be used for nonparametric clustering, but
we didn't describe this very explicitly. So nonparametric clustering, these are
methods where the complexity of the models or the description of the clustering can
grow as you get more and more data points. Another method is called spectral
clustering that can be robust to different cluster shapes like the Swiss roll type
images we showed in the second module, when we're motivating clustering and talking about some of the limitations of
the methods we were going to describe, where we have to actually specify
the shape of the clusters. But unfortunately spectral
clustering methods don't tend to have very good scalability
properties to large data sets. And then there are a set of ideas that are
related to things that we talked about but not specifically for
retrieval or clustering. So, as an example, mixtures of
Gaussians are commonly used for something called density estimation. So we showed this picture of this
histogram over the blue intensity of all the images in our data set and we talked about the distribution
on those intensities. So we can think of that as a density
over intensities, and try and estimate the form of that
density explicitly, rather than thinking about mixtures of Gaussians
as a means of clustering data points. So they're hand in hand,
the same tool, but different tasks. And then within the context
of density estimation, you can start talking about
things like anomaly detection, where this is a question of you
get some new data point, and does it look significantly different than
the data points that you've seen so far. But with what you've learned in this
class you have the tools to go out and learn these other methods. [MUSIC]