10:00:25 Hey man, I thought, I see you in the office. Well I'm in Richards attorneys Yeah, I see him in the background. Yeah.
10:00:35 And, and saying backs over here.
10:00:40 Oh cool.
10:00:47 Yeah.
10:00:50 I have to remember that life has very different configuration.
10:00:56 I'm having difficulty to
10:01:00 look into the device relapse off. Yeah.
10:01:07 Can you hear if Richard speaks hello yeah yeah I can't hear.
10:01:14 Okay, I'm just gonna stay muted on my laptop video.
10:01:42 Yun, can you hear us.
10:01:47 I can contacted Marianne Maria, Yes, be your media, but how have a chat tomorrow. Good, good.
10:02:00 I don't know if there's anything.
10:02:12 Still, she can do but
10:02:12 we'll see ya.
10:02:15 When you have time tomorrow to add one.
10:02:18 You have tantamount to that one. Yeah. Good. Great, I send you the Mossad.
10:02:22 Actually, I didn't. So I appreciate it may have gone to my spam for some reason you've been going to spam lately. Well then, check your spam because if I send it again and again.
10:02:34 I got it, I got it. There it is.
10:02:36 Okay, let's see who are we missing.
10:02:44 Whereas, Ethan I don't see Ethan.
10:02:51 He might be in, not in today he's getting a new car.
10:02:54 Oh, okay.
10:02:59 Okay, then why don't we start.
10:02:59 Even anything.
10:03:01 Yeah, let me give a brief update on what I've been working on.
10:03:14 Okay, so today I want to talk briefly about the project I've been working on basically using non parametric masters to extract the proton charged radius.
10:03:27 So, the proton radius puzzle is the discrepancy between the bionic hydrogen spectroscopy, and the Masters using our normal, normal electrons, like the EP scattering and the normal hydrogen spectroscopy.
10:03:44 And for the scattering experiments there's another level of difficulty in a way is that the way we define the radius of the proton is the taking its derivative of the electro form factor, sq squared equals zero, And the scattering experiment, we, there's
10:04:03 no way for us to extend to measure anything close to zero. So, the difficulty lies in the analysis on how we extract, our existing data at low key square to the Q squared equals zero, and gets the derivative.
10:04:20 So, I'm going to talk about to Colonel masters.
10:04:24 The first one is called the Gaussian process, and the one being used is a subset of the whole Galton profits family which is called the various in Provo the colonel masters.
10:04:34 The other one is the rich content refresh.
10:04:37 Before I dive into the details of these two Colonel methods. I going to talk a little bit
10:04:45 on the logic, we get to these methods. So there are two key points we get the colonel regression. The first one is what I caught that primal and tool form of regression.
10:04:59 So let's consider most familiar linear model. When you can write axe FX suppose x is the higher dimension, you can have multiple higher multiple access of.
10:05:14 So this is a linear regression, we're familiar with. And in the 1d case it's just a bias plus the way the slope time stacks.
10:05:24 So, in the most general form. If you try to minimize them. Main Square era of this progression model, we get the weight, the W as a function of. It's basically an expression of that.
10:05:38 Those matrix operations. And this, this expression here is called the primal form of regression and this is the form we are quite familiar with the most of the cases.
10:05:50 However, if you look into this expression and you can notice that we can rewrite it by adding one extra term of X transpose X. So that in the end it's, we can represent this all the weights, as a linear combination of all the training sex, you know, in
10:06:07 the capital x matrix.
10:06:09 So, this is called to form of regression.
10:06:20 And if you try to solve for the alpha here in this relation, you can find out that in the end on the model we got four new point at x star, the prediction we want for new point at x star after the regression.
10:06:27 If you look at all the way to the expression on the right, you can see that it involves two terms, the x, the matrix X times X principles and also the matrix x times the vector x star.
10:06:42 The key point for this form is that the expression that only contains the thought products of each pair of instances in that in that space. And that's the one key point for for for the regression.
10:06:59 Another point is that we can actually using a map, which maps the original X to another space, which I caught fire x. And by doing so, for example we have five x equals the vector of one x x square and execute.
10:07:17 Now we can transform a third or the polynomial regression into a linear one by writing this FX equals, and the five x times the weight.
10:07:27 So now, giving this mapping, we can apply it directly to the tool form, we extracted. Previously, and just replace them matrix X, with the matrix five after that transformation.
10:07:43 So, the key point is that we know this in the dual form solution. Only the inner products in the feature space it's used, we can. So, the idea of Colonel is that we bypass the mapping explicit form of the mapping five x, but instead we give only the form
10:08:03 of inner products.
10:08:13 The inner product of x one x two, which will call Colonel here directly.
10:08:10 So, the rich Connor regression, let's just use a current regression, plus regularization because for most of the cases, the colonel underlying mapping of the colonel will give this model a lot of flexibility.
10:08:25 So we need to add some level of regularization to prevent overfitting.
10:08:29 And for this model you are too hyper parameters. One is the regularization constant and controls how much regulation we put our weight. And the other thing is, the corner parameters are the exact form and that expression constant in expression of the
10:08:49 colonel.
10:08:49 So here, I show you an example of the kernel, which we call the radio basis function.
10:08:56 It's basically an exponential. The dots product between these two variables. After the mapping is just exponential. And on the distance, it's more like a calcium function.
10:09:10 So if we do a tail extension on this expression, we see that this kernel is equivalent to the American which I showed in the bottom of the line of the slides.
10:09:22 So, as you can see, it's actually a transformation of the origin of x into infinite them.
10:09:33 Order of polynomials, but with an attenuation and set proportional to the magnitude of the x.
10:09:38 So, actually did. As you can see that this kernel actually transform the original space into an infinite space with infinite order of polynomials.
10:09:50 So that's the that's the magic trick of using the kernels that's you can use the kernel itself to do a mapping that's infinite that mentioned, the other model is called the calcium process, by definition, the Gaussian process is that collection of random
10:10:07 variables and, and they find out of them having a joint Gaussian distribution. But in my case, and in most of the cases in our machine learning work. When we talk about calcium process, we actually talk about what subset of it, which is called the base
10:10:24 improbability Colonel method.
10:10:27 The idea here is that we use, still the similar similar expression of our after the embedding map. But here, we assume a prior distribution for the weights w here to be normalized at zero with a covariance matrix of sigma, sigma p.
10:10:47 Now, because why is the linear combination of all the weights W, which is a Gaussian distribution, then the Y itself will also be a DLC and distribution, was the mean to be zero because we assume the weight to be centered zero, and a covariance, if you
10:11:07 take the expression can CF to do the calculation can see the covariance is a form of five times the sigma p, and five transpose, which we call k, this is a slightly different expression compared to the previous slides I put on a kind of original question
10:11:26 because here now the colonel contains the prior distribution of the ways.
10:11:39 So, now, we can use the wonderful property of the multivariate Gaussian distribution. And we know that after this assumption we know that our observed wise, and the points we want to predict will also form a joint Gaussian distribution with me at zero
10:11:58 and the covariance as this covariance matrix and the using the marginal distribution.
10:12:05 We get our Gilson distributions that gives us the most likely value for the, for the FX star we want to predict.
10:12:16 So, using these two models approach in my study is that I generate pseudo data on the lecture form factor, using several known models, and I add noise and a global normalization, and to the model.
10:12:33 And then I use my Colonel masters to fit those data to see how good my method can extract and the underlying radius, I put it into models that generates those students.
10:12:46 And now we'll talk slightly, a little bit about the choice of kernel and the hyper parameters, and how I calculate the radius after the fit.
10:12:55 So, in order to generate student as I use multiple existing models, although they might not be the ultimate truth of the, of the form factor function, but those are the models that fits the data relatively well, and each of the model actually has their
10:13:16 underlying radius.
10:13:20 So, this is how the pseudo, gentlemen. Can you go back. So Allah on, did you put in that radius.
10:13:36 No, it's, it's you, dear. In that code they provide the give the radius, as this, I think, I mean it's their minds they get a smaller radius, right.
10:13:44 They fit yeah they.
10:13:46 Yes, you're right, maybe I just get a number. I didn't copy the correct number but I use the default value for for for the mines data.
10:13:55 Okay, so I think eight five or something like that. Okay. Yeah, so maybe it's just a typo, okay yeah it's a it's maybe a typo. So, this number is way too large for that.
10:14:06 Yeah, yeah, you're right.
10:14:10 So on to add noise to the pseudo data. I generate all the pseudo data, because I want to compare the results was there was a period results. So I Genet all the pseudo data.
10:14:20 That's a queue square off those period of measurements.
10:14:24 And the noise added to the pseudo data is generated based on the statistical uncertainty. And I also add one term of uncertainty, on the global normalization of the form factor because there's normalization on the cross section, we measure.
10:14:43 So, on the bottom I shows weird period data and located.
10:14:46 So, the left one is the Q square on the linear scale, and the.
10:14:54 The middle one is cucumber on the large scale as you can see we curate a period has a lot of measurements very close to cubes great called zero, and the bottom right, it's a very soon mean, we're only the points that's very close to cure squared equals
10:15:17 and that this is how the data, the actual data looks like. And the pseudo data will be generated at the same to square, but maybe, but with random noise.
10:15:26 So, the choice of colonel in my study is just I introduced is the radio basics function. It has the properties that you can have a decay, a skill of elsewhere, so that the points are closer to each other, have a larger impact with with each other and
10:15:54 plot points are far away have much smaller impact. So the hyper parameters high choose the hyper parameter is that for the Gaussian process, because everything is joint calcium, we can actually maximize the log likelihood of all the data we see by adjusting
10:16:15 the hyper parameters, and those hyper parameters are decided by. By doing so, and for the colonel retrogression, I need to do a grid search off the minimum, I'll sample chi square, reduce the chi square for my tenfold cross validation station just to
10:16:28 avoid over 50 the data.
10:16:30 So this is how the face diagram for the two hyper parameters for the colorings regression look like. So, this is a heat map of reduce the chi square, I'll sample, just to show we're on the optimal choices of hyper parameters might be located.
10:16:52 So as you can see here.
10:16:55 Your is correlated region for the two hyper parameters, I have set have set gives low at a reasonable and a very low all simple reduced chi square and this correlation correlated region.
10:17:13 The reason I have to use that grid search for the optimal type of a minute parameters. Instead of using traditional using the traditional optimization methods, because there are strong correlation between these two qualities, and you can notice that there
10:17:33 are actually several points that has relatively small value.
10:17:39 are actually several points that has relatively small value Givens that have statistical fluctuations were adding noise to data, and in later I will show that those actually those several points are actually all valid, but choosing them.
10:17:50 there's a strategy behind choosing choosing them.
10:17:54 So, after it fits the model. The way I calculate the radius is that's given. I simply use the definition, because given our model can predict and the value at all the square, I can just use the definition by choosing on very small delta cute square and
10:18:11 and calculate the difference between RG at this delta Q square and G at zero, and just take the ratio. And this is numerically stable for q3 we're down to 10 to the next, 10, and the numerical accuracy for taking this difference that's quite was these
10:18:28 two values quite close to each other, there's also more than enough for our purpose to extract the radius.
10:18:36 So here are the results.
10:18:40 On this slides on the left plot, I show the period, the pseudo data generated by the P read fits with normalization era of 1%.
10:18:57 These points on the way.
10:19:00 We should read this graph is that the x axis is delta r, which is the radius, I extracted using my model. And the difference between this value and the value in the model itself when it generates the beta.
10:19:17 So we want them to be as close to zero as possible, which means our, then our. My way of extracting the radius has low bias. So here, for each kind of dots.
10:19:31 It's extracted the Delta are the difference between the extracted a radius and a true radius, for each underlying models that generates the pseudo data.
10:19:48 And the error bar just show the variants of Ortiz, and I saw them twice. So, on the right. It's basically the same plot, but in a different fashion. So the first.
10:19:56 On the left, the first.
10:19:59 The first one of the three is the bias, which is basically the offset here between the point and zero. So as you can see here.
10:20:08 All the executive radius is bias to the negative direction so my moto always earned the estimates exact radius, and the middle one is the standard deviation, which is basically how large the error bar here is.
10:20:24 And the last one is the means routes main square error, which is basically a square of bias, times the square of the standard deviation, and then taking the square root.
10:20:46 So it quantifies basically all the error. In this method, some of them come from the bias, and some of them from come from the standard deviation of the extraction. The variance.
10:20:48 So, as I just mentioned, Florida, Colonel.
10:21:02 Rich Colonel regression we have, we could have multiple choices of the hyper parameters. One of the hyper parameter is the legalization factor. So here I pick three different level of personalization, basically represents that the three level, different
10:21:10 level of suppression of the parameters of the weights, we put into the regression. So on the top left is the largest regularization factor, which means we have the largest suppression on all the flexibility in our model, and the right one is the middle
10:21:39 the bottom one is the one was small, the smallest legalization. So, our model has the largest flexibility. As you can see it's very obvious on the bottom left, left, once we give the model, a lot of flexibility we have large variants when we extract the
10:21:48 data, and we extract the radius, but on the top, and for the
10:21:54 on the top for the largest top left for the largest regular session factor. We have a huge depression on the model flexibility, you see a slightly larger bias, when we give it some flak that flexibility on the right.
10:22:12 So the conclusion is that in this scenario for this model, we want to choose medium for a legalization factor to give the model adequate amount of flexibility, while still control at the variance of the structure radius from this model.
10:22:33 So, If you are the bottom right, I show how the period, the final model the period choices, the original one one model fits the pseudo data.
10:22:45 So compared to the one on the right.
10:22:50 At the top right and pop bottom one, you see that my message can have slightly better variants, compared to the years, but my method has a consistent underestimation of the radius.
10:23:06 So, I'm not seeing what you notice is that a lot of traditional masters using a well defined function for extracted radius has actually the dependence on the cut off, energy, a cut off cute square.
10:23:21 So basically, the results were change where we change the largest and most q square values we feed feed into fitting their model. So I did another study on, and how my method, specifically that kind of regression method.
10:23:42 If it depends on the cutoff nq square we in our data sets and plot on the bottom showed stats for all the green, orange dots. I cut off all the points with q squared smaller than the points I mocked.
10:23:58 So for the first point here. Basically I generated pseudo data from curiosity square, but I leave everything below this to square out when I set up my model.
10:24:12 And for all the blue dots, I cut off all the queue square up above where I fit the model. So as you can see that because we have a non value of G equals one of two squared equals zero, and the model is actually insensitive if we cut off, cut away some
10:24:29 the low key square measurements. However, if we cut off some of the high IQ score measurements, because it actually moves information about the general shape for the electronic form factor.
10:24:45 The model will perform worse and worse as we lose more and more points at haikus career, basically we have less information on the shape. The actual shape of the form factor function.
10:24:57 Yeah, I think that's it. So, both approach includes because we give the model, a lot of flexibility, by using the colonel, and got both approached me to include inevitable legalization just keep it from overfitting the data.
10:25:16 And we noticed that there's regularization always resulting underestimation of the port on track rate is the Gaussian process is easy to train, but the leaves no room for adjustment for the good kind of rich regression.
10:25:30 We have to use a grid search, which is computationally very expensive. But the regularization constant gives us a handle to trade bias and varies, and the model using larger organization has comparable, or even slightly better results compared to the
10:25:50 traditional function function fitting methods.
10:25:55 And this approach is insensitive to the cut of energy, given that we have enough points in the data set, like I showed you may have we cut some of the input data so we can still get a relatively close value.
10:26:11 Until we lose, like the majority parts of the data.
10:26:15 Yeah. So, that's it.
10:26:21 Okay. Thank you man questions to you men.
10:26:29 Instead of the grid search. Could you also do just hit the sand or something.
10:26:36 No, I would imagine that if I give it enough number of some enough number of tries then it's possible.
10:26:47 Because when I try to extract the optimal hyper parameters, I have to let it tree on the noisy data, which means that for those points I will have fluctuations, and they are so close to each other and with some correlation it's very hard for for the gradient
10:27:06 descent masters to work. Yeah. Yeah, that makes sense. Yeah, yeah. Okay.
10:27:12 While we're on this slide, maybe I missed this.
10:27:18 Should I be considering this as being fairly smooth, with the the the Yeah, Apparently choppy.
10:27:28 on a, on the color bar.
10:27:44 DC region are relatively smooth, but there is some structure in it sets the middle in the middle, it's slightly higher end up like slightly smaller at the two edges.
10:27:58 This is what I observed. And this, the fluctuation here is mostly from the fluctuation of the noisy data.
10:28:09 So this is moves but on this scale.
10:28:13 We see fluctuation in the noisy data.
10:28:16 but we also see us some structures that at center the value is slightly like at the center of the, of the diagonal line, the value is slightly larger compared to more places
10:28:36 as Richard, your understanding of the bypass the other underestimation of the radio.
10:28:42 Where does that come from.
10:28:54 It's come from the regularization so basically we want to suppress the model have a lot of flexibility. but in the process of doing so, we, we, like, inevitably, give us some suppression on the, on the slope, we want it to be as flat.
10:29:05 So, basically we don't want it to fluctuate a lot.
10:29:10 And then the penalty will put into the model basically make the model has a slight preference of flat curve, which will reduce the slope.
10:29:22 So even if we use the correct model to correct, and to use the correct model fits the beta, we still have an estimation on that slope is the method, ultimately they're not very useful, you know as buyers.
10:29:42 I mean, the model has flexibility to fit the data pretty well. And the facts we know that the estimation is from the regularization we can, as I showed that we can change the legalization to a slightly just to control their own estimation.
10:30:02 Once we like.
10:30:05 Even if we have an estimation it's not as I show here it's not performing worse compared to the traditional methods because they also have, like, bias, based on different underlying model set generates the pseudo data.
10:30:20 So the fact, we have a bias that's from the realization
10:30:28 is like.
10:30:31 It's not worse than the ultimate answer or the bias, can you quantify the bias, and then can you correct the answer. I mean in the end we want to determine what the protons heart rate is.
10:30:44 Yeah.
10:30:46 It's hard to quantify the bias, we all we know it's it's small, and it's biased towards the negative direction.
10:30:55 The thing I think I can draw from this mastered is that remote theoretically almost theoretically it's biased.
10:31:19 Under estimates, as the fact that it's put a use it on some of the existing experiment data to extract the radius will give us underestimation of the actual radius.
10:31:14 So basically give us a lower box for that. Yeah, there's no bonds being larger than the viewer hydrogen expect trust. Trust complete data can tell us something.
10:31:28 So, the idea is that My,
10:31:33 My way of ultimately using this matter is that it just provides a lower bar, and the lower bound can be helpful, given that our EP scattering data is EP scattering radius is larger than the real hydrogen spectroscopy.
10:31:53 So you know, worst case scenario my model and the estimated radius, but it's still larger than the than the real hydrogen spectroscopy, then the thing still exists the discrepancy still exists, it's not from the analysis itself.
10:32:13 So what do you get, I guess.
10:32:18 So what is the proton radios that you get from this work from your work what what I haven't put it onto the data but I can get the number, pretty quickly.
10:32:30 I can update you with that.
10:32:34 So you would do it for each experiment.
10:32:46 Yeah, I think I will do a similar study for each experiment, just to get get an estimation of how much bias my model will have. Then after that I will use the model directly fit experiment data, to get a number.
10:32:52 So you've got a number, you'll get a lower bomb for each experiment. Yeah, yeah.
10:32:59 I think I can get a lower volume for for p read and the minds.
10:33:07 Okay.
10:33:07 Okay. Other questions to you man.
10:33:16 Okay. Anything else you men you wanted to race. So I talked to you, Mike. Last Wednesday, about this project. And he's pretty happy with that. So, so basically I think that's enough for for the joint degree I'm going to have.
10:33:36 So, I probably will just wrap up this project. He's kind of interested in.
10:34:01 Ah, Indiana won't get it published, he's kind of interested in using he's, he's studying on different neural networks structure and he was interested in seeing how it will be applied into this proton radius puzzle.
10:33:56 So after that, after that meeting he gave me some source code of your neural network, and I tried it, so on, then I will just talk to him about that. This.
10:34:13 Other than that, I heard from mines yesterday. That's all the current plan is to restart things.
10:34:23 On September, the 10th.
10:34:26 Yeah.
10:34:27 Okay.
10:34:29 As I understand it, Mike comes back to MIT on around the 15th of June.
10:34:37 Yeah.
10:34:41 No, it's good that he's interested.
10:34:44 Okay. Anything else.
10:34:48 Then, on Tupac says, I haven't heard anything about the test Spain.
10:34:55 We will have a team meeting on Thursday
10:35:01 Can't remember what time 1145 I think was the time.
10:35:08 So that's it for two packs, unless Victor Do you have anything to add yeah I I exchanged the mathematic fingers.
10:35:28 Fingers Fingers generated the truth. Okay, great. So I haven't looked at the paper recently but I'll do that later today maybe.
10:35:37 Okay. And they also adjusted the captions, because I had these left right fingers but no our fingers allow one is on top of each other. So, one is about one stone.
10:35:54 Ok.
10:35:55 Okay, now will we should look at that Ethan's not here at the moment, but we should try and get that out and send it out to the group for comments.
10:36:08 If there's no okay.
10:36:11 I mean, we can also talk, do we want to also add some other figures concerning like this test with late in front of the bus calorie meter or not.
10:36:31 I don't know why don't we talk about that offline.
10:36:35 It seems to me that they those tests, we did were fairly inconclusive. weren't they, they didn't
10:36:45 was just because of the energy struggling of the beam.
10:36:50 Yeah, that was the problematic thing. Right.
10:36:55 Okay. Anything else on feedbacks, not from my sight. Okay. Patrick Did you have anything.
10:37:09 I'll assume not.
10:37:13 Okay, dark light.
10:37:17 We're going to have a meeting this Thursday at one o'clock.
10:37:23 So those are responses. And we've had a lot more people sign up for the doodle poll, and that still remains the optimal
10:37:36 number not as many as you might like there's all over and Thomas.
10:37:45 Yeah, yeah.
10:37:50 Oh and Bob Blackstone is there is trying as well.
10:37:55 So, yeah, so one o'clock.
10:38:00 Anything else you wanted to add about dark later.
10:38:04 We had a meeting last week I guess and we're getting going there is aspects of the relations, and the test environment.
10:38:15 The test environment. So there's, if you're not on the list and want to be part of it just like.
10:38:21 We're going to be every week.
10:38:28 Yeah.
10:38:31 Okay.
10:38:35 Anything else he I see a che anything new.
10:38:44 Yours. I don't understand
10:38:48 these calculations.
10:38:52 Okay, yeah, look at that next week when Hagar is available.
10:39:00 Um, The chats.
10:39:05 Now Patrick says there's mics not working.
10:39:10 And no updates. Okay.
10:39:14 So unless there's something else go on to class 12.
10:39:23 All right. Bye Bye then. Good day, y'all. Talk to you later. Bye.
10:39:36 Talk to you later by
10:39:36 the end.
10:39:36 So we've released a poster Richards released the poster on DNP anyway the poster for the fall meeting so you can start thinking about coming to Boston now.
10:39:50 Anyway, the poster for the fall meeting so you can start thinking about coming to Boston now. Okay.
10:39:53 So class 12 are the dates of the GNP it's like October 11 to 14 Is that right, right. Yeah, great website to
10:40:06 just do DNP 2021. Okay, sure.
10:40:10 And the deadline for abstracts is July 129 days away.
10:40:26 Okay so, class drove updates, Bobby.
10:40:41 transferring over to new computer, which is great.
10:40:44 So a little bit of work, just for, I don't know, technical point, Windows has full compatibility now with Linux. So if you like, Linux, you can do everything on windows that you can on Linux and I mean like actually everything.
10:41:02 So, that's nice. Anyways, besides so it's nice to have a computer that seems to be entirely working now, I may still have to send back the old computer back to Lenovo tech support.
10:41:14 I'm kind of waiting, a week or so to make sure I don't need anything that I forgot to transfer over, because I have to kind of try and, you know, really wipe it before I send it out.
10:41:28 Anyways, I'm still doing a little bit of work on filtering for my ready to try not generator.
10:41:34 Andre Kim, who's post a postdoc or staff scientist at j lab I'm not sure exactly which has been in touch with me a little bit about some work I did a few weeks ago on beam spin a symmetries he seems to believe that, you know, we have kind of some small
10:41:49 differences and he is confident that that's just you to me missing a side band subtraction. In my work.
10:41:58 But I'm going, you know, some ways back and forth with him on that.
10:42:02 And the main thing is getting ready to submit some radiative generated events to the system Oh, I guess I'm also working a bit with Maury to.
10:42:16 There's been some like queuing issues on our, you know, getting software simulations to run so there's a small kind of technical aspect I'm working out with more to kind of have that automatically handles because over the past couple weeks jobs have a
10:42:29 like, 100,000 jobs were submitted something like this, and the whole system kind of crashed and had to be cleared out, I think st Beck was affected.
10:42:36 Anyways, so nothing major I'd say that's about it.
10:42:41 Also I guess we all know the collaboration there's a collaboration meeting going on now. Actually, and I kind of had a weekend away from work and I kind of got lost some emails, so actually kind of missed the first part of the meeting, but I think anyways
10:42:58 I think that's all for me anyways.
10:43:02 Yeah, I, I just saw that over the weekend I got an email from john son, they don't seem to advertise it mean what does he advertise the meeting I mean I don't get any notice of these meetings.
10:43:25 I'm not. Anyway, I'll join in, I can actually add a touch of a question for you people who are at MIT.
10:43:35 So I'm vaccinated now.
10:43:38 Do I have to ask, Karen TAO to be on Colbert pass daily to come into MIT or what's the situation of accessing MIT.
10:43:50 I was when I was a TA, but due to some touching the reason that I forget this episode Karen switch me to like a one time access sort of thing back in December.
10:44:04 And that is what I'm still on, I believe, so I don't know if that needs to get changed.
10:44:09 Bali does. Yeah. Yeah. Well, yeah, you need to code paths right now anyway, and we're getting tested, like something weekly, just as it always was pretty vaccination so yeah well I got an email from Karen about that she.
10:44:34 Well, she messed up something, some configuration of Michael Vick path. After my QA but it will it didn't happen my network they'll probably should have said that earlier.
10:44:49 You were in the same situation of TA, the same course and, yeah, that's right. Yes. Okay. I'll investigate.
10:44:59 Do you have a new office, Bobby.
10:45:03 Oh well I guess that was another question which is, I mean I see you and Richard are in the same office together. So I think I'd be in the office that same back is in.
10:45:12 In, for 21 I think
10:45:16 it's so I'm assuming seen as you two are together that two people can be in the same space at this point on campus. I think it depends still, you know, we're just here for this meeting, I think, yeah, I still got my office.
10:45:34 You know two doors down.
10:45:36 That is not limiting any, any advocacy. After that, our archive. So, so, one of my tasks. To get started today was to actually do a survey of all the space and a new roster where everybody is.
10:45:58 Okay. For the last 1214 months so. So, I will.
10:46:00 Oh, I guess, I wouldn't be doing that hoping the next week. Now, know where everybody is.
10:46:07 Yeah but, in principle, you can have two people in an office now, or after June, so long as you're back so, yeah.
10:46:18 As long as your immune. Believe me,
10:46:22 Yo and follow the rules. I think roles still are to wear a mask around.
10:46:27 Yeah.
10:46:37 On the card or.
10:46:34 Okay, same back, anything. Yeah, I can
10:46:43 see I'm actually saying back is here with us so okay so
10:46:57 panel is actually.
10:47:01 So, well I'm starting, still, still looking at one day, example.
10:47:12 It was just a section.
10:47:10 So, let me start with this small thing, it's very compatible with classic form for fermented data, I use RJ for 2018 embedding configuration embedding peripheral already and I use overruns
10:47:44 were simulation. I thought we did about 500 million, million events, but didn't have a chance to look at all these events but I only use 10% of that. so I use 50 million in all possible, you can ask, Steve.
10:48:06 Just during Facebook's.
10:48:06 So, well this it's kind of messy plot so pack and kind of showing how I vector got to back right definition ratio. The black one is actually referred to data right one as signal green one aspect one, and there's some mission in blue, as the simulation,
10:48:31 if these are visit blood flowing, removing signal, and background.
10:48:37 So, only showing the farm today that simulation interested.
10:48:41 Again, cannot find any disagreement to finish.
10:48:46 Because limitation of statistics.
10:48:53 And also, these are multiple other disciplines service positivity variable, read as signal and green one aspect run.
10:49:08 Everything is so closely variable 60 servers, but the thing they are critical.
10:49:13 But looks quite good.
10:49:19 But then if I have
10:49:28 more quicker parts of the cross section of blood one as data entry one red one as soon all green on the background, and I plotted. These all simulation business.
10:49:37 And then I have to purchase builds and if I divide that by.
10:49:42 Let me not interrupt integrated luminosity I will help section. So differentiate perfection in.
10:49:52 Observe connect to spirit of stuff very wise D and transferring the file.
10:50:00 So the red one issue by scandal data.
10:50:07 Well the skill.
10:50:10 Do the brown one is actually the FBI tiller protection estimate from theory and divided into one is extremely abusive.
10:50:21 This is will depend on some may argue that it will though is not working quickly.
10:50:29 Well this is extra committed to previous previous experiments.
10:50:40 The effects.
10:50:40 The purple one working so my data is kind of below.
10:51:06 Yeah. Picture below that.
10:51:01 Well clearly there is.
10:51:05 So, I don't believe.
10:51:08 I think the race between the purple and and the data or friend.
10:51:14 Yeah, the beta seems to fit.
10:51:19 Yeah, I don't want to get the older, Kurt
10:51:28 all is actually present in Bobby's channel.
10:51:39 Okay, so they are either is calculated separately.
10:51:39 Yes.
10:51:40 Yes. Well they're, they're all from the same generator is a horse horse.
10:51:49 Well, yeah.
10:51:53 Yeah, so, by BBC I mean, the TV show was a dental or scared, so. Well, technically they can be Lord, that, that definitely.
10:52:19 But, yeah, so I should have noted that if it's if it's if it's a fourth beast.
10:52:17 You guys.
10:52:17 Give me a dental or scared.
10:52:23 So,
10:52:23 the BBC, I added at the opposite level. Right, right. Yeah, you square the opportunity to write some good BBC us contribution to bed. But then you get interference.
10:52:34 Yeah, I just don't know you like what you're comparing fairness is going to be negative.
10:52:41 Right, right. So I think last.
10:52:49 Yeah, yeah, yeah so sorry. This is not sure if it was just a dental and supposedly has a book out of here. Okay.
10:53:02 So,
10:53:02 yeah, so.
10:53:06 Wow.
10:53:08 Well, I don't know, probably, sincere. Seuss issue.
10:53:17 This is a narrow bed is this is right in there. In really where is this under the fact that we're
10:53:31 going to for texture and protein angel is going in.
10:53:38 And
10:53:41 how many events will be in this
10:53:46 in some
10:53:49 first signal I get for 50,000 60,000.
10:53:57 Yeah, because I collect around the hundred thousands at that hundred thousand defense that pass this is Jeff guts, but 30% or 40%.
10:54:18 So,
10:54:25 Actually, Christian.
10:54:41 This is such a
10:54:38 I can calculate.
10:54:44 Chapter angle to wait for Jeff is one is a using front end viable, the other one is using the quantum side.
10:54:52 And the thing is to check. I'm assuming that these two agreed in generic normal for the
10:55:01 reconstruction reconstructive events.
10:55:10 And it was, well, this is actually diverse, just define my feelings. Okay.
10:55:23 I found around 4000 events in one run.
10:55:30 5303.
10:55:30 Well, this one has four multiple triggering events, there's limited to five inverse, gentle barn.
10:55:43 I know this is right so I have to calculate integrated the somebody okay got over all detective region. And I hope that's 800 in the wild, so that they can direct them to one times 500% to buy, it's pretty lofty events.
10:56:05 I hope this isn't working because I've seen disparity between material.
10:56:12 Do it, but I think it's also good not on your on your own lot of people are working on this I'm going to talk.
10:56:23 I was in contact with Voltaren Johnson and elastic pika many people at work.
10:56:29 Yeah.
10:56:31 But I think it's good that you look at it.
10:56:36 Yeah, for sure right to deeper. Well me yeah he's named him I think we want to get him to do the talk.
10:56:46 Yeah.
10:56:50 Yeah, right. So, I'm also curious about is momentum friction. Yeah, but all. Yeah, so this is a different story but close to six USD momentum friction.
10:57:05 Electronic.
10:57:07 I, they perform the lecture and moment impression by trusting front and product because we're delighted.
10:57:25 But imagine just ending production boring reconstruction is actually worth that collection construction. So,
10:57:35 Well, I don't think that's making sense courting electronic music producer which is worth. Holding classic, they did and will probably last year's last year's celebration day, they presented their plans to perform the same thing but they make a full presentation
10:57:50 full presentation after they figured out how
10:57:55 to correct electric trusting putting actually electronics worked with this know.
10:58:08 Actually, electronics, watch this. Now, remind me so yeah I think looking at last it so the idea is to look the normalization. Yeah, I agree. I think
10:58:17 really the cross check yeah you know I have to do it to 5%, you know,
10:58:24 I mean at least check if things are off by a factor of 10 or something.
10:58:31 Yes, you know, one of these points away.
10:58:54 So
10:58:47 hey, Patrick, anything, did you get your mic fixed.
10:58:57 So he wrote, not much of an update, going back and forth with Brandon on simulation, have been wrapping up work on class 12 streaming readout simulation traveled back to Philadelphia last week, so have been catching up with family and friends that haven't
10:59:19 seen since the pandemic.
10:59:24 Actually I wanted to ask, you know, Brandon hasn't been in the last few meetings is is a not getting the notices or is he just on available at this time.
10:59:42 I came last week late. Okay. Okay.
10:59:51 Um, any Richard did you want to say something. In general, or.
11:00:00 Oh, yeah, no, but, Georgia. I thought you were going to comment something on calculating cross sections or.
11:00:07 Yeah. So,
11:00:16 you see my email.
11:00:16 Yeah.
11:00:18 So I started this by saying, well, it comes on and off for our DB up meeting has become bonked.
11:00:29 I reported on our discussion last week make it every two weeks, we'd like the presentation of lasting peace Gatorade intrinsic motivation and ego or wanted presentation models course we can present updates as promises made.
11:00:59 So this is the reply from Bulker I agree we should have died in meetings general the main goal is for students in their PhD project, time to time to get a theorist or senior science presentation with physics topics on the schedule and then you present
11:01:01 as relates four months or so.
11:01:04 for next four months or so. Each week one topic so elastic scattering, the mark off
11:01:11 to load to abuse was realistic event generators with VCs and the event BMP you can ask it back row on the VCs generator, Larry presentation, students using some generators presentation, number three was re a corrections your various DVD pre processes,
11:01:37 we could ask for more senior person present how they're implementing those processes.
11:01:38 He was convenient should be responsible for three of the meetings I'll prepare those students and recently presented so you get a true presentation.
11:01:48 And then comes on came in his group is graduating graduate, PhD, getting going. I've been following the mockups inclusive analysis work quite well so far no seriously working the last cross section national.
11:02:05 And it was opposed to my group we worked on last 12 on the conditioning day with summer to the last one was student Brandon clarity did some work with
11:02:17 five thesis, the elastic cross sections are important.
11:02:20 Finish.
11:02:21 It was not a simple job songs as well for moves to work.
11:02:27 Regarding reaction mechanism pretty good also agreed and very important topic, probably the most important my personal opinion.
11:02:49 Score range
11:02:41 to separate the energy data by angle spectrum from plan the production code shows consent, right dominance of course I'm fine.
11:02:51 already strong indicator transfers virtual proton dominance is also consistent.
11:02:56 Things may change.
11:02:57 Anyway, going back and forth, so I think we should try to start this thing up next week, can maybe you can think about when you might want to present
11:03:15 on what's the last picture.
11:03:22 Let's talk about, you know, it's not that you, that's what I'm. That's why I'm saying like, we started in and you think, I mean, this could occupy the next year, you know, last Saturday I don't think that's what we want.
11:03:33 So, I think we have to be smart about this and see what's been done what can we do what should certainly be distributed.
11:03:42 I don't think you can just need gas analysis as well as elastic.
11:03:49 part of the
11:03:52 day the whole inclusive electron part of that should be the same you know for everything is that being is that all the same for everybody.
11:04:01 And so, whereas the complication.
11:04:06 I don't know, you know,
11:04:09 you each of you to check the proton as well was not elastic proton, so you actually do have, in some sense, the elastic analysis that, but elastic you just have a complete correlation proton and electron.
11:04:24 But elastic you just have a complete correlation proton and electron. So, I don't know, you know i mean i think that all people look at this in a focused way, you can probably make progress.
11:04:34 So, anyway,
11:04:41 I assume there's nobody here wants to talk next week.
11:04:44 Is that right,
11:04:50 yeah I think yeah that's right Bobby, Patrick.
11:04:57 You are not here so I don't think I would have.
11:05:04 So I figure.
11:05:07 Sorry, so I'll send it. I feel for this everybody should have done by now.
11:05:19 So you can look at it, we can
11:05:22 proceed. Now, Patrick says. Next week is too soon for him. Yeah.
11:05:31 Okay, I'm searching, anything you want to.
11:05:37 Yeah.
11:05:38 Yeah. Yeah, so I totally moved to Jeff set up at this weekend.
11:05:50 And, theoretically, I should be fully reading to the lab, but I checked the online status off my registration to shut up.
11:06:00 Looks like no updates since last week. So two things to be changed. Why is the host for. I believe should be from James is still showing that in process.
11:06:18 And the second one is my passport.
11:06:23 Inspiration issue.
11:06:25 I need to contact so to ask her whether she can update that item for me.
11:06:31 Given eyes, I have proved that I'm in process of my passport renewing.
11:06:41 I'm so I did the safety training online.
11:06:47 Looks like was the registration forms, was the are all completed shouldn't be any more issue for me to get into the lab.
11:07:10 Yeah, so
11:07:15 for the computing resources work.
11:07:21 I turned to another direction which is to learn about how Nathan who is another computing with another job staff in charge of the computing since.
11:07:36 Learn from his script about some similar tasks.
11:07:43 I think I will have another discussion with him today with Morrie to see how to fit in our situation.
11:07:54 Without else.