So in summary, spectral clustering builds an affinity matrix of the data points.

In addition, it's multi spectral.

Identical in shape, size, spectral content.

We now talk about spectral clustering.

Let me illustrate this once again with the data set that has a different spectral clustering than a conventional clustering.

Realtime audio processor based on frequency spectral manipulation

For instance, you see the spectral line of the gold is the only spectral line in the spectrum of the Sun.

This is called affinity based clustering or spectral clustering.

Spectral partitioning, L1 minimization. This comes from the optimization community.

Now the key element of spectral clustering is called the affinity martrix, which is a 9 by 9 matrix in this case, where each data point gets graphed realtive to each other data point.

The fundamental idea of spectral clustering is to cluster by affinity.

So spectral clustering gets annotation of affinity to make clustering happen.

So let me look at the simple example for spectral clustering that would also work for K means or EM, but they'll be useful to illustrate spectral clustering.

Data, data, data...

Bing! That's data. Data, data, data.

And it turns out once again that spectral sensitivity is very important here.

One is the phong reflection model, which gives us a spectral highlight, shiny spot.

And to understand the importance of spectral clustering, let me ask you a simple intuitive quiz.

Say sorted data is data.

little data, old data, no data. So back in 1995,

Spectral stars that are like those that make elliptical convolved with say Gaussian distribution of Doppler shifts.

You learned about K means, you learned about expectation maximization, about dimensionality reduction and even spectral clustering.

These hyper spectral images are false colored so you can it's really good for determining vegetative use.

Data

Data...

Data

Data

Data

Data

like DNA data or brain data.

Raw data now! Raw data now!

And later we went to study the spectral stars to see is there anything strange with these objects.

The background radiation is simply too static to be able to be seen under the normal spectral analysis.

looking at wind data and solar data.

We could talk about government data, enterprise data is really important, there's scientific data, there's personal data, there's weather data, there's data about events, there's data about talks, and there's news and there's all kinds of stuff.

Data Viewer

(binary data)

raw data

? target data?

Merge data

Data Mode

Data mode

Data size

Data encoding

Data Section