没有合适的资源?快使用搜索试试~ 我知道了~
温馨提示
试读
29页
使用在已知物理过程中受过训练的变分自动编码器,我们开发了一种单边阈值测试,以将先前未见过的过程隔离为异常事件。 由于自动编码器训练不依赖于任何特定的新物理特征,因此建议的过程不会对新物理的性质做出特定假设。 基于此算法的事件选择通常会基于模型相关的假设检验来补充经典的LHC搜索。 这样的算法将提供异常事件的列表,与其他科学领域中的典型操作类似,实验协作可以进一步检查甚至发布为目录。 在此数据集中重复发生的事件拓扑可能会激发新物理模型的建立和新的实验搜索。 在大型强子对撞实验的触发系统中运行,这样的应用程序可以识别异常事件,否则这些异常事件将丢失,从而扩展了大型强子对撞的科学范围。
资源推荐
资源详情
资源评论
JHEP05(2019)036
Published for SISSA by Springer
Received: December 6, 2018
Revised: February 18, 2019
Accepted: April 18, 2019
Published: May 7, 2019
Variational autoencoders for new physics mining at
the Large Hadron Collider
Olmo Cerri,
a
Thong Q. Nguyen,
a
Maurizio Pierini,
b
Maria Spiropulu
a
and Jean-Roch Vlimant
a
a
California Institute of Technology,
1200 E California Blvd, Pasadena, CA 91125, U.S.A.
b
CERN,
Espl. des Particules 1, 1217 Meyrin, Switzerland
E-mail: olmo@caltech.edu, thong@caltech.edu, Maurizio.Pierini@cern.ch,
smaria@caltech.edu, jvlimant@caltech.edu
Abstract: Using variational autoencoders trained on known physics processes, we de-
velop a one-sided threshold test to isolate previously unseen processes as outlier events.
Since the autoencoder training does not depend on any specific new physics signature, the
proposed procedure doesn’t make specific assumptions on the nature of new physics. An
event selection based on this algorithm would be complementary to classic LHC searches,
typically based on model-dependent hypothesis testing. Such an algorithm would deliver a
list of anomalous events, that the experimental collaborations could further scrutinize and
even release as a catalog, similarly to what is typically done in other scientific domains.
Event topologies repeating in this dataset could inspire new-physics model building and
new experimental searches. Running in the trigger system of the LHC experiments, such
an application could identify anomalous events that would be otherwise lost, extending the
scientific reach of the LHC.
Keywords: Beyond Standard Model, Hadron-Hadron scattering (experiments)
ArXiv ePrint: 1811.10276
Open Access,
c
The Authors.
Article funded by SCOAP
3
.
https://doi.org/10.1007/JHEP05(2019)036
JHEP05(2019)036
Contents
1 Introduction 1
2 Related work 3
3 Data samples 3
4 Model description 9
4.1 Autoencoders 9
4.2 Supervised classifiers 14
5 Results with VAE 16
6 How to deploy a VAE for BSM detection 21
7 Conclusions and outlook 23
A Comparison with auto-encoder 25
1 Introduction
One of the main motivations behind the construction of the CERN Large Hadron Collider
(LHC) is the exploration of the high-energy frontier in search for new physics phenomena.
New physics could answer some of the standing fundamental questions in particle physics,
e.g., the nature of dark matter or the origin of electroweak symmetry breaking. In LHC
experiments, searches for physics beyond the Standard Model (BSM) are typically carried
on as fully-supervised data analyses: assuming a new physics scenario of some kind, a
search is structured as a hypothesis test, based on a profiled-likelihood ratio [1]. These
searches are said to be model dependent, since they depend on considering a specific new
physics model.
Assuming that one is testing the right model, this approach is very effective in dis-
covering a signal, as demonstrated by the discovery of the Standard Model (SM) Higgs
boson [2, 3] at the LHC. On the other hand, given the (so far) negative outcome of many
BSM searches at particle-physics experiments, it is possible that a future BSM model, if
any, is not among those typically tested. The problem is more profound if analyzed in
the context of the LHC big-data problem: at the LHC, 40 million proton-beam collisions
are produced every second, but only ∼1000 collision events/sec can be stored by the AT-
LAS and CMS experiments, due to limited bandwidth, processing, and storage resources.
It is possible to imagine BSM scenarios that would escape detection, simply because the
corresponding new physics events would be rejected by a typical set of online selection
algorithms.
– 1 –
JHEP05(2019)036
Establishing alternative search methodologies with reduced model dependence is an
important aspect of future LHC runs. Traditionally, this issue was addressed with so-
called model-independent searches, performed at the Tevatron [4, 5], at HERA [6], and at
the LHC [7, 8], as discussed in section 2.
In this paper, we propose to address this need by deploying an unsupervised algorithm
in the online selection system (trigger) of the LHC experiments.
1
This algorithm would be
trained on known SM processes and could be able to identify BSM events as anomalies.
The selected events could be stored in a special stream, scrutinized by experts (e.g., to
exclude the occurrence of detector malfunctions that could explain the anomalies), and even
released outside the experimental collaborations, in the form of an open-access catalog. The
final goal of this application is to identify anomalous event topologies and inspire future
supervised searches on data collected afterwards.
As an example, we consider the case of a typical single-lepton data stream, selected
by a hardware-based Level-1 (L1) trigger system. In normal conditions, the L1 trigger
is the first of a two-steps selection stage. After a coarse (and often local) reconstruction
and loose selection at L1, events are fully reconstructed in the High Lever Trigger (HLT),
where a much tighter selection is applied. The selection is usually done having in mind
specific signal topologies, eg., specific BSM models. In this study, we imagine to replace
this model-dependent selection with a variational autoencoder (VAE) [11, 12] looking for
anomalous events in the incoming single-lepton stream. The VAE is trained to compress
the input event representation into a lower-dimension latent space and then decompress it,
returning the shape parameters describing the probability density function (pdf) of each
input quantity given a point in the compressed space. In addition, a VAE allows a stochastic
modeling of the latent space, a feature which is missing in a simple AE architecture. The
highlighted procedure is not specific of the considered single-lepton stream and could be
easily extended to other data streams.
The distribution of the VAE’s reconstruction loss on a validation sample is used to
define a threshold, corresponding to a desired acceptance rate for SM events. All the
events with loss larger than the threshold are considered as potential anomalies and could
be stored in a low-rate anomalous-event data stream. In this work, we set the threshold
such that ∼ 1000 SM events would be collected every month under typical LHC operation
conditions. In particular, we took as a reference 8 months of data taking per year, with an
integrated luminosity of ∼ 40 fb
−1
. Assuming an LHC duty cycle of 2/3, this corresponds
to an average instantaneous luminosity of ∼ 2.9 × 10
33
cm
−2
s
−1
.
We then evaluate the BSM production cross section that would correspond to a signal
excess of 100 BSM events selected per month, as well as the one that would give a signal
yield ∼ 1/3 of the SM yield. For this, we consider a set of low-mass BSM resonances,
decaying to one or more leptons and light enough to be challenging for the currently
employed LHC trigger algorithms.
1
A description of the ATLAS and CMS trigger systems can be found in ref. [9] and ref. [10], respectively.
In this study, we take the data-taking strategy of these two experiments as a reference. On the other hand,
the proposed strategy could be adapted to other use cases.
– 2 –
JHEP05(2019)036
This paper is structured as follows: we discuss related works in section 2. Section 3
gives a brief description of the dataset used. Section 4 describes the VAE model used in
the study, as well as a set of fully-supervised classifiers used for performance comparison.
Results are discussed in section 5. In section 6 we discuss how such a procedure could
be deployed in a typical LHC experiment while relying exclusively on data. Conclusions
are given in section 7. Appendix A provides a brief comparison between VAEs and plain
autoencoders (AEs).
2 Related work
Model-independent searches for new physics have been performed at the Tevatron [4, 5],
HERA [6], and the LHC [7, 8]. These searches are based on the comparison of a large set
of binned distributions to the prediction from Monte Carlo (MC) simulations, in search for
bins exhibiting a deviation larger than some predefined threshold. While the effectiveness
of this strategy in establishing a discovery has been a matter of discussion, a recent study by
the ATLAS collaboration [8] rephrased this model-independent search strategy into a tool
to identify interesting excesses, on which traditional analysis techniques could be performed
on independent datasets (e.g., the data collected after running the model-independent
analysis). This change of scope has the advantage of reducing the trial factor (i.e., the
so-called look-elsewhere effect [13, 14]), which would otherwise wash out the significance of
an observed excess.
Our strategy is similar to what is proposed in ref. [8], with two substantial differences:
(i) we aim to process also those events that could be discarded by the online selection, by
running the algorithm as part of the trigger process; (ii) we do so exploiting deep-learning-
based anomaly detection techniques.
Applying deep learning at the trigger level has been proposed in ref. [15]. Recent
works [16–19] have investigated the use of machine-learning techniques to setup new strate-
gies for BSM searches with minimal or no assumption on the specific new-physics scenario
under investigation. In this work, we use VAEs [11, 12] based on high-level features as a
baseline. Previously, autoencoders have been used in collider physics for detector moni-
toring [20, 21] and event generation [22]. Autoencoders have also been explored to define
a jet tagger that would identify new physics events with anomalous jets [23, 24], with a
strategy similar to what we apply to the full event in this work.
Anomaly detection has been a traditional use case for one-class machine learning meth-
ods, such as one-class Support Vector Machine [25] or Isolation Forest [26, 27]. A review
of proposed methods can be found in ref. [28]. Variational methods have been shown to
be effective for novelty detection, as for instance is discussed in ref. [29]. In particular,
VAEs [11] have been proposed as an effective method for anomaly detection [12].
3 Data samples
The dataset used for this study is a refined version of the high-level-feature (HLF) dataset
used in ref. [15]. Proton-proton collisions are generated using the PYTHIA8 event-generation
library [30], fixing the center-of-mass energy to the LHC Run-II value (13 TeV) and the
– 3 –
JHEP05(2019)036
average number of overlapping collisions per beam crossing (pileup) to 20. These beam
conditions loosely correspond to the LHC operating conditions in 2016.
Events generated by PYTHIA8 are processed with the DELPHES library [31], to emulate
detector efficiency and resolution effects. We take as a benchmark detector description the
upgraded design of the CMS detector, foreseen for the High-Luminosity LHC phase [32].
In particular, we use the CMS HL-LHC detector card distributed with DELPHES. We run
the DELPHES particle-flow (PF) algorithm, which combines the information from different
detector components to derive a list of reconstructed particles, the so-called PF candi-
dates. For each particle, the algorithm returns the measured energy and flight direction.
Each particle is associated to one of three classes: charged particles, photons, and neutral
hadrons. In addition, lists of reconstructed electrons and muons are given.
Many SM processes would contribute to the considered single-lepton dataset. For sim-
plicity, we restrict the list of relevant SM processes to the four with the highest production
cross sections, namely:
• Inclusive W production, with W → `ν (` = e, µ, τ ).
• Inclusive Z production, with Z → `` (` = e, µ, τ).
• t
¯
t production.
• QCD multijet production.
2
These samples are mixed to provide a SM cocktail dataset, which is then used to train
autoencoder models and to tune the threshold requirement that defines what we consider
an anomaly. The cocktail is built scaling down the high-statistics samples (t
¯
t, W , and
Z) to the lowest-statistics one (QCD, whose generation is the most computing-expensive),
according to their production cross-section values (estimated at leading order with PYTHIA)
and selection efficiencies, shown in table 1.
Events are filtered at generation requiring an electron, muon, or tau lepton with p
T
>
22 GeV. Once detector effects are taken into account through the DELPHES simulation,
events are further selected requiring the presence of one reconstructed lepton (electron or
muon) with transverse momentum p
T
> 23 GeV and a loose isolation requirement Iso <
0.45. If more than one reconstructed lepton is present, the highest p
T
one is considered.
The isolation for the considered lepton ` is computed as:
Iso =
P
p6=`
p
p
T
p
`
T
, (3.1)
where the index p runs over all the photons, charged particles, and neutral hadrons within
a cone of size ∆R =
p
∆η
2
+ ∆φ
2
< 0.3 from `.
3
2
To speed up the generation process for QCD events, we require
√
ˆs > 10 GeV, the fraction of QCD
events with
√
ˆs < 10 GeV and producing a lepton within acceptance being negligible but computationally
expensive.
3
As common for collider physics, we use a Cartesian coordinate system with the z axis oriented along
the beam axis, the x axis on the horizontal plane, and the y axis oriented upward. The x and y axes define
the transverse plane, while the z axis identifies the longitudinal direction. The azimuth angle φ is computed
from the x axis. The polar angle θ is used to compute the pseudorapidity η = −log(tan(θ/2)). We fix units
such that c = ~ = 1.
– 4 –
剩余28页未读,继续阅读
资源评论
weixin_38748555
- 粉丝: 6
- 资源: 934
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功