1
00:00:00,299 --> 00:00:05,399
Okay, so I will just give you a brief
report on the most recent multiple,

2
00:00:05,429 --> 00:00:11,249
multiple measurements done by CMS
experiment. myself with that, just a very

3
00:00:11,249 --> 00:00:16,499
brief introduction. Before I go into more
details of the two most recent multi

4
00:00:16,499 --> 00:00:22,169
button analyses performed by CMS, which
are the W plus W minus and they try both

5
00:00:22,169 --> 00:00:28,949
and analysis, I will not talk about vector
scattering in this talk. So, if you're

6
00:00:28,949 --> 00:00:36,179
interested in that, you can follow Mario
talk on Friday afternoon. So, just a brief

7
00:00:36,179 --> 00:00:40,349
introduction and motivation while it's
important to perform multiple

8
00:00:40,379 --> 00:00:46,349
measurements, while the normal structure
of the standard model allows for

9
00:00:46,979 --> 00:00:52,289
interrupts telling directionality vector
by zone, which can be principal triple and

10
00:00:52,289 --> 00:00:58,169
courting any number of zones to connect it
to the vertex and a procedural measurement

11
00:00:58,199 --> 00:01:04,019
of the strength of this coupling is You
put on tests of the electric sector from

12
00:01:04,019 --> 00:01:11,279
the model itself. Moreover, as we already
are in the previous talks, the indirect

13
00:01:11,279 --> 00:01:17,219
sales for new features can be performed
with these analogies because if we see a

14
00:01:17,549 --> 00:01:21,779
difference between the predicted cabling
and what we measure which which which we

15
00:01:21,779 --> 00:01:26,549
call anomalous coupling, could hint to the
prevalence of new physics. Another

16
00:01:26,549 --> 00:01:32,039
important motivation to do these
analogies, of course, that many multiples

17
00:01:32,039 --> 00:01:38,429
on prophecies, actually backgrounds in the
study of new physics, or for example, in

18
00:01:38,879 --> 00:01:46,439
exit searches how these measurements are
very challenging because as you can see

19
00:01:46,439 --> 00:01:52,619
from the blog, the cross section four is
processing very low, generally smaller

20
00:01:52,619 --> 00:01:58,859
than peak about what to say. And so this
is why we need to use the full power of

21
00:01:58,859 --> 00:02:05,639
the full CMS atlases around two
statistics, which is about 114% of them.

22
00:02:06,569 --> 00:02:12,149
These are complex final state, because we
have a higher high particle multiplicity.

23
00:02:12,179 --> 00:02:17,279
And as I said before, these are high
precision measurements. So we need to have

24
00:02:17,279 --> 00:02:23,969
a good control on detector effects, kill
factors and so on to the precision

25
00:02:23,969 --> 00:02:29,639
measurements that we really want to the
basic idea of all the analysis is to

26
00:02:29,639 --> 00:02:35,279
measure the critical section, and
hopefully find in our significant seven

27
00:02:35,279 --> 00:02:40,379
evidence for an observation of prophecies
that have never been mentioned before. So,

28
00:02:40,799 --> 00:02:46,109
I will now talk about the first one to
focus on which is the W plus W minus

29
00:02:46,109 --> 00:02:52,319
analysis which has been published by CMS
experiments using the data collected in

30
00:02:52,319 --> 00:02:58,319
2016. So disease, an important background
source in searches for new particles but

31
00:02:58,319 --> 00:03:06,539
also, for example, indicating to WW. The
production comes mainly from tiki bar or

32
00:03:06,959 --> 00:03:13,229
intrusion if you consider higher order
interactions. And this analysis has been

33
00:03:13,229 --> 00:03:17,279
performed by CMS following two different
approaches. One which is the sequential

34
00:03:17,279 --> 00:03:21,989
cap one, which is let's say the usual way
to do the analysis, and the other one that

35
00:03:22,019 --> 00:03:26,639
exploits a random forest classifier and I
will show you why these two approaches

36
00:03:27,209 --> 00:03:33,869
somewhat somewhat complimentary and why
they can be different results. In the

37
00:03:33,869 --> 00:03:38,609
sequential exceptionalities. The selection
is pretty much the standard one. He wants

38
00:03:38,609 --> 00:03:42,599
to opposite charges related electronic
work since you have to do both on the cane

39
00:03:42,989 --> 00:03:47,129
and you want a trunk does not make a big
transformation coming from the neutrinos.

40
00:03:47,999 --> 00:03:55,229
The two main backgrounds come from Julian
and teba are single top events. And as

41
00:03:55,229 --> 00:04:00,539
usual, the single topic toolbar can be
reduced by providing a beachhead laser

42
00:04:00,869 --> 00:04:07,769
lights will Ditalion background and will
be used to discriminate between signal and

43
00:04:07,769 --> 00:04:14,579
background. As I said before, the analysis
used is also another approach which is the

44
00:04:14,579 --> 00:04:19,499
one of the run the first classifier. And
random forest classifier is basically an

45
00:04:19,499 --> 00:04:27,179
aggregate of boosting decision trees which
each percentage entry does not use the

46
00:04:27,179 --> 00:04:33,539
whole set of kinematic parameters, but a
subset of them. And in this way the

47
00:04:33,539 --> 00:04:39,059
overall results that it is you can get
from all the trees together mitigates the

48
00:04:39,059 --> 00:04:43,409
problem of overfitting. You can every
single BDT

49
00:04:45,180 --> 00:04:50,310
as before, it would mean backgrounds.
Jolyon and tiki bar to run the first

50
00:04:50,310 --> 00:04:54,660
classifier and been built for this
analysis. And as you can see from the

51
00:04:54,660 --> 00:04:59,970
blocks on the bottom left in the single
region, which is the one in the orange box

52
00:05:00,000 --> 00:05:07,080
So, you have a very high contribution from
WWF signal and a very low contribution of

53
00:05:07,080 --> 00:05:12,090
the Tiki Bar and this is after applying
both the score of the random forest

54
00:05:12,090 --> 00:05:19,470
classifier for the Julian and for the top
backgrounds. It has been shown that the

55
00:05:19,470 --> 00:05:24,360
random forest classifier analysis as a
signal efficiency impurity higher is

56
00:05:24,360 --> 00:05:32,040
compared to the center frequency counts
anomalies. The backgrounds are the Tiki

57
00:05:32,040 --> 00:05:37,050
Bar, the tread yacht and also the non prom
events which are basically from these

58
00:05:37,050 --> 00:05:43,980
analogies w plus Jessica and when one
where one jet is reconstructed as electron

59
00:05:44,460 --> 00:05:50,130
will not go into high detail, but
basically for all of these contour contour

60
00:05:50,130 --> 00:05:56,130
regions built in both the sequential
account or random forest and these

61
00:05:56,130 --> 00:06:00,720
backgrounds are obtained by normalizing
the contribution in the Conservation

62
00:06:00,720 --> 00:06:07,140
before estimate what is the contribution
in the signal region. This is semantics

63
00:06:07,170 --> 00:06:12,570
the telephone for the sequential caption
you can see on the table below the

64
00:06:13,230 --> 00:06:17,340
greatest uncertainty comes from the
analyzation of the backgrounds. So, this

65
00:06:17,340 --> 00:06:22,380
is the most delicate part of the analysis,
while there is also a contribution from

66
00:06:22,380 --> 00:06:28,290
the theoretical side coming from mainly
from Halo the QED calculation of

67
00:06:28,290 --> 00:06:32,730
certainty, the total systematic efficient
is about 6% while the statistical one is

68
00:06:32,730 --> 00:06:42,540
1%. So, this is a systematic dominated
analysis. So, the first measurements that

69
00:06:42,540 --> 00:06:46,200
came out from this analysis is the
measurement of the topic resection with

70
00:06:46,230 --> 00:06:52,290
the two different methods. So, from the
sequential data analysis, all the

71
00:06:52,290 --> 00:06:58,200
different possible channels are studying
separately and then put it all together

72
00:06:58,530 --> 00:07:03,900
where you can see The signal strength are
measured separately for Vito jet one jet

73
00:07:04,050 --> 00:07:08,820
sync same flavor or differently or
channels and all the signal strength and

74
00:07:08,820 --> 00:07:13,920
in agreement with each other with one face
or with the Standard Model prediction and

75
00:07:13,950 --> 00:07:18,900
this proves that there was a good
consistency of the model and the overall

76
00:07:19,170 --> 00:07:24,330
cross section is in agreement with the
theoretical prediction. random forest

77
00:07:24,330 --> 00:07:32,970
classifier is more sensitive to their
weighting of the debug on spectrum tables

78
00:07:32,970 --> 00:07:37,830
in terms of momentum spectrum that is used
to take into account higher the QC D

79
00:07:37,830 --> 00:07:44,880
calculation. And as a result, these
sensitivity to this correction is what you

80
00:07:44,880 --> 00:07:51,060
can see in the plot, which is that random
forest classifier as a higher efficiency

81
00:07:51,210 --> 00:07:58,470
at the low values of the momentum offers a
voting system and this means that it is

82
00:07:58,470 --> 00:08:06,090
more sensitive to Jessie banks and this is
why the devalue the central value of the

83
00:08:06,090 --> 00:08:11,580
measured cross sectionally leaking the
differential prior to the sequential

84
00:08:11,610 --> 00:08:16,380
Katelyn, it has taken to be taken into
account that also this radical uncertainty

85
00:08:16,380 --> 00:08:23,100
entire situation for the same reason for
these analogies, also the new show and

86
00:08:23,100 --> 00:08:26,280
differentiate perfection and be measured
and these numbers are just there for

87
00:08:26,280 --> 00:08:30,990
reference. And as you can see from the
prop on the top right, there is a very

88
00:08:30,990 --> 00:08:34,260
good agreement. For example, this is
differential perfection in the dialect in

89
00:08:34,260 --> 00:08:38,970
math, there is a very good agreement
between data and simulation. And the

90
00:08:38,970 --> 00:08:44,340
random forest classifier allows a lot over
to the measurement of just multiplicity.

91
00:08:44,670 --> 00:08:48,630
And it is very important because it's a
probe of theoretical calculation that

92
00:08:48,870 --> 00:08:55,530
ultimately go into the event generator.
And so it's important to know how the just

93
00:08:55,530 --> 00:09:00,780
multiplicity is precisely and as you can
see from the block on the bottom left Also

94
00:09:00,780 --> 00:09:07,020
you're going to live by the computer
simulation and the data and the lasting

95
00:09:07,020 --> 00:09:12,900
results with analysis also, believe it or
not most coupling have been measured. And

96
00:09:12,960 --> 00:09:19,620
as you can see from the block the highest
sensitivity is the highest value of the

97
00:09:19,620 --> 00:09:22,890
time with myself today electrons and

98
00:09:24,360 --> 00:09:29,940
limits on triple gauge Captain are
measured for the values because at the one

99
00:09:29,940 --> 00:09:35,940
presented on the Fineman diagram, and the
plus on the left represent a scam when

100
00:09:35,940 --> 00:09:40,980
you, Barney, just one parameter or capital
of them together. And the table on the

101
00:09:40,980 --> 00:09:45,900
bottom presented the results obtained by
this analysis for the limits on the

102
00:09:45,900 --> 00:09:52,800
anomalous capital. We talked about about
the other analysis. That is the most

103
00:09:52,800 --> 00:09:57,930
recent one we published, which is the
tribalism analysis that is done with the

104
00:09:57,930 --> 00:10:03,630
fool around to that effect. The goal of
these analogies was to measure both the

105
00:10:03,630 --> 00:10:09,030
inclusive tribalism production protection
and the the single line for the all the

106
00:10:09,030 --> 00:10:15,930
different processes. So, types mainly,
yes, thank you. Mainly There are of

107
00:10:15,930 --> 00:10:23,970
course, all the possible combination of
W's and D and the w w w production has

108
00:10:23,970 --> 00:10:29,040
been divided into two channels depending
on the chain. So if two of the W SDK

109
00:10:29,040 --> 00:10:35,340
electronically one and radically or all
three of them decay electronically, and

110
00:10:35,340 --> 00:10:39,990
for the flexion. The main backgrounds come
from what is called the last lesson. So

111
00:10:40,020 --> 00:10:44,970
three lesson processes, where one lesson
is not ever constructed for exception or

112
00:10:44,970 --> 00:10:51,540
efficiency problems, and the non prompter
background that I already described, while

113
00:10:52,020 --> 00:10:58,470
on the processes where at least one debit
spread in the background is lower, because

114
00:10:58,470 --> 00:11:04,560
it's possible to constrain The simpler
approach to electron mass within the value

115
00:11:04,560 --> 00:11:11,340
of the set mass and basically on the WWE
channel that main backbone comes from that

116
00:11:11,340 --> 00:11:16,650
production while on the other two channels
the backs on each individual even though

117
00:11:16,950 --> 00:11:22,530
the cross section and rushing ratio for
these processes are very, very low. So, as

118
00:11:22,530 --> 00:11:27,120
I said before, you know complex from
background is one of the main sources and

119
00:11:27,120 --> 00:11:35,700
has been estimated from data to boosting
decision trees have been used to better

120
00:11:35,700 --> 00:11:41,310
discriminate between signal and
background. And for example, in www and

121
00:11:41,310 --> 00:11:47,010
actually this one was a decision tree is
used to enhance the discrimination between

122
00:11:47,010 --> 00:11:52,770
signal and non from background and one for
all the other backgrounds. And you can see

123
00:11:52,770 --> 00:11:56,820
the result of the BDP score on the plot on
the bottom left you can see that in the

124
00:11:56,820 --> 00:12:03,210
signal region, there is a higher signal to
noise ratio was for the full electron

125
00:12:03,840 --> 00:12:08,640
factory, the main background contribution
comes from the dead end. Of course

126
00:12:08,640 --> 00:12:12,900
background especially in the case where
it's well known that electrons are

127
00:12:12,960 --> 00:12:18,360
opposite side fine. And as you can see
from the plot of the transistor, the

128
00:12:18,360 --> 00:12:24,210
mission transfers momentum. This is a good
viable where place a cap to get a higher

129
00:12:24,750 --> 00:12:29,460
sensitivity to the signal. They may
systematic uncertainty as for the previous

130
00:12:29,460 --> 00:12:34,080
analysis come from the background
estimation and the limited statistics in

131
00:12:34,080 --> 00:12:41,970
the control region estimation. This is the
results of the analysis but there's a lot

132
00:12:41,970 --> 00:12:49,290
of information. But basically this is
these are all the categories for all the

133
00:12:49,290 --> 00:12:54,990
different channels and the decay and the
decay mode or all of them. And as you can

134
00:12:54,990 --> 00:13:02,610
see, there are some channels that have a
very high significance that is The plot in

135
00:13:02,610 --> 00:13:07,980
the middle the brownish one. And as I said
before, the main background for the www

136
00:13:07,980 --> 00:13:13,680
channel comes from the last lesson, while
for the WWE that comes from the

137
00:13:14,520 --> 00:13:18,270
background, you can see that there are
many channels where the sensitivity is

138
00:13:18,270 --> 00:13:25,620
very high to the presence of these
presence of these processes. So, these are

139
00:13:25,620 --> 00:13:30,840
the results of this analysis, all the
production as we mentioned, and you can

140
00:13:30,840 --> 00:13:34,770
see them on the table on the left and so,
both the

141
00:13:36,059 --> 00:13:41,459
signal cross section and the big red one,
and they are in agreement with the

142
00:13:41,849 --> 00:13:47,429
Standard Model prediction. And you can see
on the table on the bottom right, what is

143
00:13:47,429 --> 00:13:51,269
the significance of obtaining with the two
different methods for getting counts and

144
00:13:51,269 --> 00:13:55,709
the boosting decision tree. And you see
that usually that was the decision to

145
00:13:55,739 --> 00:14:00,119
hire. I think music as compared to the
cutting count one and another thing The

146
00:14:00,119 --> 00:14:06,119
test to be noticed is that the sensitivity
for a company www Chanel is much higher

147
00:14:06,119 --> 00:14:10,079
when using the full run to that effect, it
has been done for this analysis compared

148
00:14:10,079 --> 00:14:17,639
to the 2016 only data Finally, going to
add slightly older analysis which is the

149
00:14:17,639 --> 00:14:24,089
www analysis potential statistics 16 with
turning 16 data Sorry, I wanted to include

150
00:14:24,089 --> 00:14:29,909
this because besides the limits on almost
cabling, it started also the craving of a

151
00:14:30,449 --> 00:14:35,009
possible action like political production
in associate in association with the W

152
00:14:35,009 --> 00:14:38,909
boson, which is another thing that can be
done in multiple alizee cities, which is

153
00:14:39,029 --> 00:14:43,709
test precise models. And as you can see
from the Brazilian plot on the bottom

154
00:14:43,709 --> 00:14:48,389
right, these production the progress of
this particular has been included up to

155
00:14:48,389 --> 00:14:55,709
480 gb approximately. So, this is my
summary. With the LLC run to data

156
00:14:56,609 --> 00:15:02,009
properties are accessible even though they
are rayar the perfection of the measure

157
00:15:02,009 --> 00:15:06,539
for many processes, not just the two
represented here, but for many others, and

158
00:15:06,539 --> 00:15:10,619
the limits on the animal's capital is
becoming more and more stringent. Also, in

159
00:15:10,619 --> 00:15:16,469
the better buttons capturing processes,
some efficient models can be tested. And

160
00:15:16,469 --> 00:15:22,709
also another thing is that information of
prophecy that critical level, for example,

161
00:15:22,709 --> 00:15:27,689
they want to prevent obesity can be
extracted from the internet. Thank you.

162
00:15:29,639 --> 00:15:33,239
Thank you, Alison, for this very

163
00:15:34,620 --> 00:15:38,100
concise summary. So we have time for some
questions.

164
00:15:43,799 --> 00:15:45,749
Are the question on page 11?

165
00:15:48,450 --> 00:15:48,960
Yes,

166
00:15:49,830 --> 00:15:55,290
so it's not your villain talking. So
here's what you have on the top, top left,

167
00:15:55,290 --> 00:16:01,140
you have the signature where you have to
send sign leptons and then To neutrino to

168
00:16:01,140 --> 00:16:09,210
jets. So actually this final status
exactly vector wasn't catalysts inside. So

169
00:16:09,210 --> 00:16:15,750
I'm wondering how do you track it echoes
and scattering contributions? Because in

170
00:16:15,750 --> 00:16:19,500
principle you cannot distinguish an
experiment. So they also contributed to

171
00:16:19,500 --> 00:16:25,650
the same order. Yes. Do you treat a signal
or background or you checks it is

172
00:16:25,680 --> 00:16:27,180
irrelevant into space space.

173
00:16:28,409 --> 00:16:33,689
There are some cats that did not report
here which are that the nGj mark is

174
00:16:33,689 --> 00:16:38,759
required to be below 500 gV. And data
preparation is more than 2.5 which is

175
00:16:38,759 --> 00:16:43,439
basically the opposite of what you do for
a vector both an analysis has been done to

176
00:16:43,439 --> 00:16:47,009
reduce the contributions as much as
possible.

177
00:16:50,370 --> 00:16:52,920
And then in this case, this is the
technical

178
00:16:54,330 --> 00:16:59,250
sorry. And in this first case, to set
tripod on Facebook, this is the critical

179
00:16:59,250 --> 00:17:02,400
channel. You could Do just consider this
as background though.

180
00:17:06,120 --> 00:17:09,840
I think the contribution is very small
compared to the

181
00:17:11,639 --> 00:17:17,939
the picture by production text

182
00:17:20,879 --> 00:17:22,259
question on slide eight

183
00:17:26,550 --> 00:17:33,300
don't understand correctly this plot on
the right that at the low PT, random

184
00:17:33,300 --> 00:17:37,710
forest gives you better performance, but
at the high PT is your worst performance.

185
00:17:39,750 --> 00:17:42,960
Yes, so did the IPT of the WWE system.
Yes.

186
00:17:43,680 --> 00:17:47,790
Yeah. Can you can you describe a bit how
do you construct uncertainty when you

187
00:17:47,790 --> 00:17:51,510
train BDT in one region and you apply it
in another region? What are the

188
00:17:51,510 --> 00:17:56,100
uncertainties? What are the contributions
don't say? How do you estimate

189
00:17:59,070 --> 00:17:59,610
okay.

190
00:18:01,500 --> 00:18:05,520
Noticing person so I don't know if someone
is connected again answer to this question

191
00:18:06,480 --> 00:18:10,200
because this is beyond what I know about
this.

192
00:18:17,940 --> 00:18:18,450
Sorry.

193
00:18:19,500 --> 00:18:21,780
Can you point me to the reference in the
talk when

194
00:18:22,650 --> 00:18:28,500
it's on like to leave a link to the
documentation is clickable.

195
00:18:30,840 --> 00:18:36,150
I can I can tell them a little bit on this
if you want. I think this is, um,

196
00:18:36,840 --> 00:18:41,340
shouldn't necessarily be interpreted as
where the random forest is better

197
00:18:41,340 --> 00:18:47,730
performing. It's simply showing which
events it selects. So while the while in

198
00:18:47,730 --> 00:18:54,480
principle, the random forest is trained on
on inclusive events, because it tries to

199
00:18:54,480 --> 00:19:00,990
find regions, but it's most sensitive to
separating the WW from the teeth. bar, it

200
00:19:00,990 --> 00:19:07,230
ends up selecting the lower PT events. So
this is just showing where the events come

201
00:19:07,230 --> 00:19:12,030
from. And then as to how the uncertainty
is calculated. The biggest impact of the

202
00:19:12,030 --> 00:19:19,320
uncertainty in the random forest is when
you look at the PG rating of the WWE

203
00:19:19,320 --> 00:19:22,980
system that comes from the higher order
calculations, and its most sensitive to

204
00:19:22,980 --> 00:19:25,140
this topic collectivities

205
00:19:38,190 --> 00:19:44,310
Okay, if there are no more questions,
let's thank Alessandra again. Thank you