<TITLE: Signal Processing
ACADEMIC DOMAIN: technology
DISCIPLINE: information technology
EVENT TYPE: post-graduate seminar discussion
FILE ID: USEMD140
NOTES: continuation of USEMP06A (not transcribed), seminar also includes presentation USEMP06B

RECORDING DURATION: 11 min 34 sec

RECORDING DATE: 13.4.2005

NUMBER OF PARTICIPANTS: 11

NUMBER OF SPEAKERS: 5

S1: NATIVE-SPEAKER STATUS: Finnish; ACADEMIC ROLE: research student; GENDER: female; AGE: 31-50

S2: NATIVE-SPEAKER STATUS: Romanian; ACADEMIC ROLE: senior staff; GENDER: male; AGE: 31-50

S3: NATIVE-SPEAKER STATUS: Romanian; ACADEMIC ROLE: junior staff; GENDER: male; AGE: 31-50

S4: NATIVE-SPEAKER STATUS: Finnish; ACADEMIC ROLE: research student; GENDER: male; AGE: 24-30

S5: NATIVE-SPEAKER STATUS: Romanian; ACADEMIC ROLE: research student; GENDER: male; AGE: 24-30>



<PRESENTATION USEMP06A by S1, NOT TRANSCRIBED>

<S1> so do you have any questions </S1>
<S3> er could you go backward to that example with the three sequences , containing only <S1> [this one] </S1> [C and G] er y- yes and er now er the next slide <P:10> er i think er here is it possible to to calculate even more for that probability of C-C-C of three C's in such a way to see that the probability doe- does not explicitly depend on D3 and and D4 but only on their sum </S3>
<S2> mhm [(which)] </S2>
<S3> [(yeah that)] (one) </S3>
<S2> yeah </S2>
<S1> (i think er i don't know) </S1>
<S3> so er this is not a very final er form possible form of the expression of the probability </S3>
<S1> so you mean this er can be [calculated further] </S1>
<S3> [because in] yes because in the previous slide you have shown that er you actually substituted <S1> this [one] </S1> [yeah] you ye- ye- substituted D4 plus D3 <S1> yeah [(xx)] </S1> [by] by this the [D3 for example] </S3>
<S1> [and this is and this is] the </S1>
<S3> yes exactly </S3>
<S1> and <S3> okay </S3> these too </S1>
<S3> okay but er uh actually you have shown the calculation for the example A for the figure A in the next slide <S1> yeah </S1> isn't that true , and er [the the third one yes] </S3>
<S1> [yes the first one is A but] it's then i have <S3> [yeah] </S3> [calculated] it becomes the same you have to do like this <S3> yes exactly </S3> if if they have because only D's are 16 the order [(xx)] </S1>
<S3> [yes but the point] is that for example for example er for er case A for figure A the probability depends on the sum of D4 plus D3 and not on D4 and D3 and this is a reason for which we can (calculate) the substitute in such a way to <S1> [yeah] </S1> [get the] yeah so t- this was just a very short comment </S3>
<S2> no i i could also maybe , it was so complicated this has partly maybe to do you have a <S1> yeah </S1> brief look at what er you really tried to do in that part so now er you have tools to compute the this program (xx) looks for for all possible configurations but the problem is that you have so many configurations you can't do it in a er reasonable time for all of them so you want er to sample from some (institution) that's why you have two approaches <S1> [yeah] </S1> [one was the] bootstrap the other one was how it is called er (DAT) <S1> [yeah] </S1> [sampling] (DAT) sampling </S2>
<S1> the first one was not (xx) but (xx) it is </S1>
<S2> okay so the first one which er er what what is the principle in that one , they say generate a new tree from the current tree and look if the probability of the new tree is better then you keep it <S1> [er] </S1> [if not] you may still keep it if it <S1> [(xx)] </S1> [er in a] small number of times not not so (awkward) <S1> yeah </S1> and by that you have the chance to improve all the time the likelihood so that finally you get the the the best likelihood and you you're not also getting you know (equal) local minimum that you get on a (xx) if you agree all the time and improve you are just going into that direction and you don't see what is on the sides so , that that is er interesting idea but that that was complicated a lot by these er changes of the er topology of the tree so they wanted to show here that this er full number of changes you can cover all possible topologies that's my impression your your argument was so complicated that that it's difficult to say too much er if you just look at er only this (xx) , so the topology was , showing er , what is the order of the of the leaves in the tree and also by these different heights the the sum of times that that's basically what you have in those two axes you have something related to the order of the leaves and then on the vertical axis you have time from the loop , so the minimum change is which they propose us just to alter this time from the loop <S1> yeah </S1> into (low) or to make the er the (xx) or feature to to branches , i i i don't see what is the the big discovery here because these changes are . yeah you see these are probably if you come to more than that , it's maybe the fact that you you don't have to to adjust to to very far away you know it's enough just for (for the time) adjusting to neighbours <S1> yes [it's important] </S1> [so i i want] to to to find what is the (xx) inside er this (powerpoint) because i might </S2>
<S1> i think they have er different topologies [i would say i can take er (xx)] </S1>
<S2> [yes so it it looks like you a get a problem of] </S2>
<S1> the topology maximise (so i think) </S1>
<S2> oh you can cover lots of them by by just <S1> [yeah] </S1> [these] two operations and then the final one bootstrap was er you didn't like it's not the best choice as they <S1> yeah </S1> as they analyse it it's not very good one <P:07> okay i don't have anything else to ask but maybe , do you have now questions , the reason is that you understood perfectly or it's , now i think you can hope to ask and to understand better , the answer is yes (in both cases) are there [questions] </S2>
<S4> [i don't have] (xx) </S4>
<S1> okay is there no more questions maybe (xx) </S1>
<S2> so <NAME S5> has the <S1> [<NAME S5>] </S1> [next] presentation he just escaped but er if you you take a few minutes er <S1> [yeah break yeah] </S1> [break i think he comes] [(xx)] </S2>
<S1> [so it's his] turn </S1>
<COFFEE BREAK, TRACK CHANGE>

<PRESENTATION USEMP06B by S5>

<P:18>
<S5> so it was it seems to be only half of a chapter but a lot of information and i concentrated (xx) </S5>
<S2> yeah about this final comparison i i wanted to (xx) ask er what is necessary to do before using one probabilistic model or one er not probabilistic model er in the , the lecture of last week we have seen that er for weighted parsimony we needed to know a weight matrix which was not given by in any books or anywhere properly you have to , find it by your own means er it was er algorithm using some weights for each of the of the labels added those ways and finally getting the best er tree as the one minimising that sum of weights but those weights were arbitrary ones they they they were not er given er quite the same as here you use lots of probabilities conditional probabilities of that and that and er you don't have to know that before and er , i would say still the big problem is how to get this prior knowledge before starting to do the matrix because thi- this contains as well only toy comparisons the examples are very small and i believe it's clear biological problem to have very confident that where the (xx) is it 200 probably that is a big problem with all the things we've discussed today that you don't have a ground to support this , in engineering we are used to have a ground for the erm say this is the original we just corrupted maybe that and that but not in the biology we don't know exactly the evolution we just have hypothesis and er that that is a big problem how to assess a matter against another when you don't know anything for sure everything in this evolution is (either) er more or less hypothesis you can't make some measurements on on the evolution because the time-scale in measurement is so small but evolution here is (xx) , i don't know exactly what is a PAM unit that you need for for the matrix PAM it was some (xx) it's not that secondary level (xx) , i think that the whole field is very difficult maybe it's that that you don't have enough ground (xx) enough enough data and , then the comparisons are more or less (xx) comparisons , do you have other opinions or other impressions are you <S5> no i [mean] </S5> [very] happy after reading this very carefully </S2>
<S5> er when i read it was let's say , clear except @about that thing@ when you try to implement this , the problems may appear (you know when you) try to </S5>
<P:14>
<S2> so other other comments <P:11> er thanks for the presentation and we conclude today thank you </S2>
