mohlerm / eth-cil-exam-summary Goto Github PK

Not sure if we discussed this before, but maybe we could add the coherence condition (Lecture 12, p.34).

There is a small error in the Log-Likelihood objective. The second term should be:

instead of:

As he mentioned in the lecture, the X of the second sum contains elements multiple times.

So we're coming close to the exam :)

Is there anything you think that should be added or modified?
As discussed with @passuf, I think it would make sense to try to finish the exam summary until 16:00 (UTC+2).
@kilianrisse @Robin-des-Bois

Coding Step and Dictionary update step in more detail; how it can be computed if space permits it.

This is missing completely since we don't really have space for it.
However, I think it is also pretty irrelevant since the lecture does not go into detail and only presents the general idea.

What do you think?

In lecture 4 we also discussed clustering stability, including:

• Distance between clusters
• Calculation of stability
• Normalization of stability

We could consider adding those formulas to the summary, but I am not sure if this is relevant or if we have enough space. I can make a pull request if there is nothing more important we can add.

There is not a lot about it; things one could add:

.difference haar/fourier (in general that different bases needed for different signals)
.advantage/disadvantage PCA vs fixed basis
.overcomplete dicts and their advantages, disadvantages (mainly inversion)
.noisy observations?

What do we need?

Not sure if the condition on the summary is equivalent, but in the slides the k-means stopping condition is given by:

Whereas in the summary it states:

If we have space, we could add the formulas for empirical risk and L2 regularization.