Student A
Prof Beerenwinkel eased into the exam by asking what a Bayesian network is and then proceeded to ask the conditional independence statements defined by a BN are.
He then asked to define what an HMM is and I explained the overall structure and what the parameters are.
He then asked how we would estimate the parameters and I tried to explain the Baum-Welch Algorithm. He then decided to go a bit deeper into the structure of the EM algorithm - what do we do in the E step and the M step?
He also asked if the EM algorithm gives us the ML estimate of the parameters. I said no, and then he asked well then what does it give us? I was a bit confused here .What he expected was that the EM algorithm estimates the local maximum of the likelihood function.
He then asked about variational inference and then I explained what we do in that setting . He also asked where we use variational inference …and I forgot this. But it was in the factorial HMM. He then asked me to explain the factorial HMM and then asked what makes computation difficult in FHMM ——we have more than one HMM so possible state transitions we have to explore is enormous(need to consider all combinations of factors).
He then moved on to NEM and asked first what do we use it for and then asked to explain the structure . He then asked how we might get the parameters for the NEM from the data. He also asked how we can infer the signalling network , and I was confused but he expected the different greedy search algorithms like the pairwise search, triplet search etc.
He then asked about usage of NEMs in the context of single cell data.
He lastly asked about how we estimating genetic diversity from virus populations . So I explained local haplotype inference/Dirchilet process mixture and also a bit about global haplotype inference . He asked about where we used the HMM in diversity in inference ,and I was a bit confused and he prompted me by saying we used it in the context of recombination. So then I explained the structure of the jumping HMM. And that was all , folks!
Overall the exam was fair. Prof Beerenwinkel was friendly and helpful as well - if I got stuck, he would prompt or change the topic if things were going nowhere. At the end, he would tell whether we passed or failed but not our grade. I also expected that he would give us a choice of topic to start off with (this is what happens in Evolutionary Dynamics) but I was not given any choice in topics.
As you can see, it is important to know the structure and parameters of the different kinds of models so I would recommend focusing on that.