id stringlengths 36 36 | text stringlengths 1 1k | source stringlengths 42 191 | page int64 0 1.03k |
|---|---|---|---|
3bd7bd73-3e1b-440b-95c1-bac78cc72466 | pp. 19146–19158, Jun. 2023.doi:10.1364/OE.492228
[6]N. Hartmannet al., “Attosecond time–energy structure of x-
ray free-electron laser pulses”,Nat. Photonics, vol. 12, no. 4,
pp. 215–220, Apr. 2018.
doi:10.1038/s41566-018-0107-6[7]J.Lundquistetal.,“Multi-facilityvirtualdiagnosticforlongi-
tudinalphasespacepredictions”,... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6014.pdf | 3 |
2c455f2d-c509-48bf-84c1-cc4f7d6a2616 | 2022.doi:10.18429/JACoW-FEL2022-WEP22
[10]C. Behrenset al., “Few-femtosecond time-resolved measure-
ments of x-ray free-electron lasers”,Nat. Commun., vol. 5,
no. 1, p. 3762, Apr. 2014.doi:10.1038/ncomms4762
[11]K. Tiedtkeet al., “Gas detectors for x-ray lasers”,J. Appl.
Phys., vol. 103, no. 9, p. 094511, May 2008.
doi... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6014.pdf | 3 |
4c6b6a47-7bec-411e-b5f5-bb6f48c0770f | BAYESIAN OPTIMIZATION OF LONGITUDINAL
PHASE SPACE IN THE MAX IV LINAC
J. Lundquist∗, J. Björklund Svensson, S. Werin and F. Curbis
Lund University, Lund, Sweden
E. Mansten, MAX IV Laboratory, Lund, Sweden
Abstract
Reaching design performance in modern particle accel-
eratorsisachallengeinvolvingmanytaskswhicharetime-
c... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 0 |
e4e52544-9749-4ee5-a220-a343e8fec3a1 | in the optimization of longitudinal phase-space parameters
withtheuseofvirtualdiagnosticshasbeendevelopedand
implemented.
INTRODUCTION
Within modern accelerators there is a growingneed for
precise control of both transverse and longitudinal beam pa-
rameters. With linear electron accelerators providing beam
for light s... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 0 |
5be40755-f87f-4c5a-93b7-75f9901caed3 | rings [2]. The SPF houses two separate beamlines, one
experimentalbeamlineforlightproduction,andadiagnos-
tic beamline with a transverse deflecting structure (TDS)
setup for measuring the beam LPS [ 3]. The complexity and
practicalities of operating the TDS, along with the coupled
parameter space and non-linear control... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 0 |
66bc3954-89bb-4c55-aadd-a250e15af943 | ∗johan.lundquist@maxiv.lu.seMAX IV Longitudinal Phase Space Control and
Diagnostics
The longitudinal phase space of the beam is defined as
its extent in the ( 𝑡,𝛿) space, with 𝛿 =Δ𝑝
𝑝being the rela-
tive deviation in energy. Direct measurement of the LPS at
MAXIVisperformedusingtheTDSsysteminstalledatthe
endofthel... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 0 |
a22b3e60-1020-4197-8480-909e20ef3fc3 | arate stages of bunch compression, using an arc-like bunch
compression scheme [ 7]. The principal controlling beam
parameterofthecompressionistheenergychirpofthein-
comingbeam,i.e. thelinearcorrelationbetweentherelative
energy deviation and the longitudinal position within the
electronbunch. Theenergychirpresultsinlong... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 0 |
9615892a-732e-44be-931f-198336500228 | design energy for the bunch compressors [ 8]. To control
beam compression, one provides a new setpoint in SLED
filltime, and as the RF phase is changed by the feedback,
thebeamwillenterthecompressorwithadifferentenergy
chirp.
Bayesian Optimization
Bayesianoptimization(BO)isanefficientandeffective
way of optimizing envi... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 0 |
f79c0b2b-c79c-4e1d-a1eb-034f16f14bcd | tween exploration of unexplored areas and exploitation of17th International Particle Accelerator Conference, Deauville, France JACoW Publishing
ISBN: 978-3-95450-252-3 ISSN: 2673-5350 doi: 10.18429/JACoW-IPAC2026-WEP6015
2620 MC6.D13: Instrumentation: Artificial IntelligenceWEP6015 WEP: Poster session
Content from this... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 0 |
97a08c38-e296-49e1-b2f5-73d36663ea59 | themodeltoreachthehighestperformance. Forthisproject,
theupperconfidenceboundacquisitionfunctionwasused
in all cases shown [ 10].
OptimizationoftheelectronbeamLPSisaclearuse-case
forBO.Quarryingtheenvironment,i.e. performingafull
TDSmeasurement,isexpensiveandtimeconsuming,and
the response is often non-linear and comple... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 1 |
4c885bf0-c643-49ea-bcb1-1ae3c99b52ca | allowing feedbacks to adjust the machine) was integrated
with the accelerator using PyTango. This full system was
integratedandtestedontheMAXIVlinacwithoptimiza-
tionsofFWHMbunchlength,energyspread,andthesumof
the two. A novel approach to the evaluation was also added,
leveraging a virtual diagnostic to non-destructive... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 1 |
d0a46858-ad5b-4ac9-90a6-f5ae99c03ae1 | bunchcompressors. Intotal,adatasetof1400imageswas
collected,alongwithrelevantmachineparameterstotrain
thenetworkon. Thenetworkwassetupusing TensorFlow
andconsistedoftwo64-nodedenselayersfollowedbythree
layers of convolutional neural networks [ 12,13]. Training
wasdoneusingtheADAMoptimizer,andleveragedasched-
uled learn... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 1 |
2b723e6b-fb16-47c6-8501-be6ce5830b32 | FWHM in both projections. The evaluation procedure now
consisted of collecting the relevant non-destructive diagnos-
tics used in setting up the network, normalizing their values
and inputing to the pre-trained network. With all of this
in place, live optimization on a VD proved efficient and
reliable.IMPLEMENTATION AN... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 1 |
a8435091-f52d-4a1c-bea2-2e7930ec8e1b | runs,theresultsofwhicharepresentedinFig. 1. Wecansee
howtheoptimizerimproveseachparameterinafewsteps,
andtheminimumencounteredduringtheoptimizationrunis
shown in each plot. It should be noted that this minimum is
notnecessarilytheoptimumaccordingtotheBOprocedure,
but it always found to be close to the minimum during da... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 1 |
dca79e86-2988-467e-a794-72c19c69382f | collected data isolated before training the model, i.e. on a
so called test dataset. In Fig. 2a we can see a fewexamples
of predictions of the LPS alongside corresponding measure-
ments, chosen based on extreme points in FWHM bunch
length. In Fig. 2b a histogram of the 𝑅2performance across
thedatasetisshown,withlinesd... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 1 |
f3f4d7ca-a63e-459f-86bc-1153264285b2 | Figure 2: Results from the VD LPS predictions. Figure (a)shows examples of the VD predictions along with the
corresponding measured images, sorted from top to bottom by FWHM bunch length. Figure (b)shows a histogram of the
𝑅2performance,withtheperformanceintheshowncaseshighlighted. Figure (c)showsacorrelationplotofFWH... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 2 |
c2fbb2a4-6406-4faa-a355-cf9559d87fab | acting as evaluator, could be performed. The results from
three such VD optimization campaigns are shown in Fig. 3,
optimizing on the longitudinal FWHM of the bunch 𝜎𝑠,
the FWHM energy spread 𝜎𝛿and the sum of these two
parameters. Along with the parameters as extracted from
theVDLPSprediction,ineachcasetheparameter... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 2 |
2f771561-36e1-45d0-b09b-61ede9067df2 | control has been implemented in the MAX IV linac. Us-
ingtheXOptframeworktoprovidetheBOalgorithmsand
PyTango forcontrol systemintegration, an automated opti-
mization procedure for key LPS parameters has been imple-
mented. The optimization was tested on liveTDS measure-
ments,minimizingtheparametersofFWHMbunchlength
a... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 2 |
facdd8d3-c568-419f-a45d-35a76cec65a5 | ing the optimization.
wassetupandtrainedtopredicttheTDSimagesbasedon
non-destructivebeammeasurementsandmachineparame-
ters. ThisVDachievedhighperformanceintermsof 𝑅2and
correlation with measurements in terms of FWHM bunch
length. Whendeployedinside theoptimizer,the VDrepro-
duced the optima observed with real TDS meas... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 2 |
6dfdfacc-3962-45ea-b7f3-b20151973158 | ISBN: 978-3-95450-252-3 ISSN: 2673-5350 doi: 10.18429/JACoW-IPAC2026-WEP6015
2622 MC6.D13: Instrumentation: Artificial IntelligenceWEP6015 WEP: Poster session
Content from this work may be used under the terms of the CC BY 4.0 licence (© 2026). Any distribution of this work must maintain attribution to the author(s), t... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 2 |
1d1a5980-d36c-4d52-8604-5daafa99512d | REFERENCES
[1]Z.HuangandK. -J.Kim,“Reviewofx-rayfree-electronlaser
theory”,Phys.Rev.STAccel.Beams ,vol.10,no.3,p.034801,
Mar. 2007. doi:1/zero.alt4.11/zero.alt43/PhysRevSTAB.1/zero.alt4./zero.alt4348/zero.alt41
[2]MAX IV Laboratory, “Detailed design report”, Aug.
2010, https : / / www . maxiv . lu . se / beamlines -
ac... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 3 |
396a63ff-2218-4330-992a-9aaa2ee984eb | Beams, vol.21, no. 11, p. 112802, Nov.2018.
doi:1/zero.alt4.11/zero.alt43/PhysRevAccelBeams.21.1128/zero.alt42
[5]S.Bettoni etal.,“Machinelearningbasedlongitudinalvirtual
diagnostics at swissfel”, Rev. Sci. Instrum. , vol. 95, no. 1,
p. 015110, Jan. 2024. doi:1/zero.alt4.1/zero.alt463/5./zero.alt4179712
[6]J.Lundquist,... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 3 |
4bb41deb-171d-4ac3-9b10-37dd42be616a | “SLED:AMethodofDoublingSLAC’sEnergy.”,in 9thInt.
Conf. High-Energy Accel. , SLAC,Stanford,CA,USA, May
1974,pp. 576–583.
[9]B. Shahriari, K. Swersky, Z. Wang, R. P. Adams, and N.
de Freitas, “Taking the human out of the loop: a review of
bayesianoptimization”, Proc.IEEE ,vol.104,no.1,pp.148–
175,2016. doi:1/zero.alt4.11... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 3 |
7d519b00-3b02-4d76-824b-4a7abcbb98d8 | doi:1/zero.alt4.18429/jacow-ipac2/zero.alt423-thpl164
[12]M.Abadi etal.,“TensorFlow:large-scalemachinelearningon
heterogeneoussystems”,2015, https://www.tensorflow.
org/,
[13]K.Fukushima,“Neocognitron:aself-organizingneuralnet-
workmodelforamechanismofpatternrecognitionunaffected
byshiftinposition”, BiologicalCyberneti... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 3 |
88cc5f1d-4cb6-461d-bb9a-963131735626 | ISBN: 978-3-95450-252-3 ISSN: 2673-5350 doi: 10.18429/JACoW-IPAC2026-WEP6015
MC6.D13: Instrumentation: Artificial Intelligence 2623WEP: Poster session WEP6015
Content from this work may be used under the terms of the CC BY 4.0 licence (© 2026). Any distribution of this work must maintain attribution to the author(s), t... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6015.pdf | 3 |
3ea523d5-5b4d-4335-a6a1-7c5b5ead16f8 | DATA-DRIVEN OPTIMIZATION OF OPEN-LOOP CONTROL FUNCTIONS
AT THE CERN SUPER PROTON SYNCHROTRON
A. Menor de Oñate∗, M. Schenk, V. Kain, T. Prebibaj, I. Mases Sole, B. Rodriguez Mateos
CERN, Geneva, Switzerland
Abstract
To minimize beam intensity loss during a cycle in the
CERN Super Proton Synchrotron (SPS), several machi... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 0 |
ed78cc8e-8692-45ee-bf0f-6ac96c93ef65 | this method to an intentionally detuned machine develop-
ment beam in the SPS, as a step towards deployment on
the operational fixed-target beams. The approach is generic
and applicable to time-dependent parameter optimization
problems in other machines.
INTRODUCTION
TheSuperProtonSynchrotron(SPS)providesbeamstoa
range... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 0 |
9ff68f77-307e-4323-bc3b-5e97d817c4bd | bytheoptimization: thehorizontal( 𝑄H)andvertical( 𝑄V)
betatron tunes and the total RF voltage ( 𝑉RF).
Throughout the cycle, the beam is subject to a variety
of processes and beam-dynamics phenomena: injection,
RF capture, intensity-dependent tune shifts that increase
stepwisewitheachinjection,decayingeddycurrentscom... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 0 |
aad95c5e-231d-4f1d-8e41-3878e9138c42 | perimental needs vary, making manual retuning a recurring
operationalburden. Thisisparticularlyrelevantforthefu-
tureSearchforHiddenParticles(SHiP)[ 1]experimentatthe
SPS,whichwillultimatelybeahigh-intensityoperational
∗adrian.menor.de.onate@cern.ch
050100150200250300350400
Momentum(GeV/c)
0.00.20.40.60.81.01.2Intensit... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 0 |
a06ec7b7-ddf6-43c3-b527-acf0c2b93ef5 | beam equipment at acceptable levels.
Among the parameter functions, the betatron tunes and
RFvoltagehavethelargestimpactonintensityloss(orequiv-
alently, beam transmission) for the fixed-target beams under
study here and are the primary focus of this work. Other
parameterfunctions,suchasthechromaticities,willbead-
dres... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 0 |
afbcd33e-6934-441b-833d-481cf52d329d | ISBN: 978-3-95450-252-3 ISSN: 2673-5350 doi: 10.18429/JACoW-IPAC2026-WEP6016
2624 MC6.D13: Instrumentation: Artificial IntelligenceWEP6016 WEP: Poster session
Content from this work may be used under the terms of the CC BY 4.0 licence (© 2026). Any distribution of this work must maintain attribution to the author(s), t... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 0 |
b9571e00-f162-4527-abef-2728f1e316f1 | main magnets [ 3], as well as at other particle accelerator
laboratories [ 4]. This work focuses on the data-driven op-
timizationoftheopen-loopparameterfunctionsgoverning
beam transmission within an SPS cycle.
OPTIMIZATION PROBLEM AND
METHOD
Automatic tuning of parameter functions presents a func-
tionaloptimizationpr... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 1 |
54e6125c-2fd9-43f4-86b3-e3457d58c6b1 | rameter functions are correspondingly varied only within
this window. The optimization problem is then given by
max𝑄H(𝑡),𝑄V(𝑡),𝑉RF(𝑡)𝑇[𝑄H,𝑄V,𝑉RF], with
𝑇[𝑄H(𝑡),𝑄V(𝑡),𝑉RF(𝑡)] =𝐼(𝑡𝑓)
𝐼(𝑡0)(1)
Sampleefficiencyiscritical,sincemachinetimeisscarce.
Two strategies reduce the high-dimensional problem to a... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 1 |
8e3e0270-c945-4d0f-8594-c7db8b6309cf | inputfrommachineoperatorsand specialists. Second, since
the full cycle is too high-dimensional to optimize at once, it
issegmentedintowindowsthatareoptimizedsequentially.
Figure 2illustrates the actions on each of the parameters.
All parametrized values are concatenated into a single
action vector, which is modified wi... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 1 |
45205f08-25c3-4549-b84a-551fa454dc06 | skeleton points are constrained to Δ𝑄max
H,V= ±0.01 with
respect to the values present in the machine at the start of
the optimization, and the RF voltage is bounded between
0and7.5MV. Dependingontheparametrization,theaction
vector has up to 28 dimensions in the tests reported here.
The optimization uses BO with a Gau... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 1 |
248a7009-56e4-422e-b9f0-667987a70708 | are applied.
automatic relevance determination (ARD) lengthscales, and
LogExpectedImprovementasthe acquisitionfunction[ 5].
The GP hyperparameters are re-optimized at each iteration.
Theoptimizationiswarm-startedwith5to7randomactions
before BO begins proposing candidate actions.
RESULTS
Tovalidatethe optimizationlogic,... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 1 |
d4e0d792-1eec-4605-9908-2e8f2fbcd1d4 | injection and a total intensity of 3.3 × 1012p, safe for the
machine in case of severe intensity losses. Note that the
cyclesalreadyhadacceptabletransmissions 𝑇 ≥ 0.97 and
werehencepurposelydetunedbyperturbingthetuneandRF
voltagefunctionsawayfromtheirnominalvaluestocreate
degraded starting conditions. This both provid... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 1 |
8a246751-e0b0-4afd-a061-ba75d94b7462 | iteration budget allocated for the tests. The two optimiza-17th International Particle Accelerator Conference, Deauville, France JACoW Publishing
ISBN: 978-3-95450-252-3 ISSN: 2673-5350 doi: 10.18429/JACoW-IPAC2026-WEP6016
MC6.D13: Instrumentation: Artificial Intelligence 2625WEP: Poster session WEP6016
Content from th... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 1 |
c23dc14e-f835-4a42-b5a6-a41ca17b0803 | 0 500 1000 1500 2000
CycleTime(ms)0.940.950.960.970.980.991.001.01NormalizedIntensity
Window0
[210,1600]msWindow1
[1400,1800]msProfile
Firstiteration
BestiterationProfile
Firstiteration
Bestiteration
0 5 10 15 20 25 30 35 40
NumberofCyclesintheSPS0.900.920.940.960.981.00Transmission(Iend/Istart)
OptimizationProgress
Wi... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 2 |
a3f68f64-4da1-4c65-900d-930a46374449 | interface between adjacent windows. We chose this overlap
empirically;itsvalidityneedstobefurtherinvestigatedfor
different configurations of parameter functions and cycle
regions. In this test, BO sequentially improved the transmis-
sionto≥ 0.98withiterationbudgetsof20and40cyclesfor
the first and second windows, respec... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 2 |
d638ef52-16e2-48b3-a9c3-04e6d2db3dda | and to optimization across the full cycle.
CONCLUSIONS
Thispaperpresentsprogresstowardsautomatingtheopti-
mizationofopen-loopmachineparameterfunctionsinthe
0.840.860.880.900.920.940.960.981.00
InitialTransmission0.840.860.880.900.920.940.960.981.00BestTransmission
N=54
Improved:45/54(83%)
MeanT:+0.023
MedianT:+0.021
In... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 2 |
516a272b-7c7b-4900-903f-7c88025cc786 | missionusingonlyintensitymeasurementsasfeedback,and
actingintheparameterspacedefinedbythechosenfunction
parametrizations (linear or spline interpolations, or constant
shifts).
ThemethodwasvalidatedonpurposelydegradedSHiP
cycles at intensities of 3.3 × 1012p, where the optimizer
recovered transmission by acting on seque... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 2 |
f4db32c6-9f9a-42d6-82c3-d1c75594e06b | multiple injections, higher intensities, and larger parameter
families has yet to be demonstrated.
Next steps will focus on incorporating additional domain
knowledge,improvingbeam-stateobservations,extending
the optimization to larger families of parameter functions
(includingchromaticity),andvalidatingthesystemathighe... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 2 |
603743ef-60b0-4a2a-bbc8-c4abe2bd6778 | REFERENCES
[1]S.Alekhin etal.,“Afacilitytosearchforhiddenparticlesatthe
CERN SPS: the SHiP physicscase”, Rep.Prog.Phys. , vol.79,
p. 124201, 2016.
doi:1/zero.alt4.1/zero.alt488//zero.alt4/zero.alt434-4885/79/12/1242/zero.alt41
[2]L. Foldesi et al., “Model-based optimisation for automated
multi-turnextractiontuningatthe... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 3 |
26a777c1-d2f0-4f83-b20e-4a96d30cefd9 | Aug. 2024.
doi:1/zero.alt4.11/zero.alt43/PhysRevAccelBeams.27./zero.alt4848/zero.alt41
[5]S.Ament etal.,“UnexpectedImprovementstoExpectedIm-
provement for Bayesian Optimization”, arXiv:2310.20708,
2025. doi:1/zero.alt4.4855/zero.alt4/arXiv.231/zero.alt4.2/zero.alt47/zero.alt4817th International Particle Accelerator Con... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6016.pdf | 3 |
80dd9975-032f-4eb0-925a-1d45a17f7e76 | TOWARDS FULLY AUTOMATED TRANSFER LINE COMMISSIONING AT
THE CERN SUPER PROTON SYNCHROTRON
A. Menor de Oñate∗, M. Schenk, V. Kain, B. Rodriguez Mateos, G. Dal Maso, N. Charitonidis
CERN, Geneva, Switzerland
Abstract
Beamcommissioningofslowextractedbeamsfromthe
CERNSuperProtonSynchrotron(SPS)totheNorthArea
experimentaltar... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 0 |
9f65cf24-a84e-43f8-8c2f-f53c7494cd65 | activesurface,andwhenthecorrectormagnets’polarities
are not known; common sources of uncertainty during com-
missioning.
We present an extended multi-stage optimization scheme
that addresses these critical limitations by automating beam
threadingwhenthetrajectoryexceedstheSEMs’acceptance,
systematically identifying cor... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 0 |
502f70cd-e39e-4602-8c5c-8c04123c8261 | icatedsteeringalgorithms. Finally,BOisusedtooptimize
the position of the movable SEM monitors at the targets’
locations, maximizing target intensity.
INTRODUCTION
The CERN Super Proton Synchrotron (SPS) sends DC
proton and heavy ion beams to the North Area (NA) experi-
mental targets through a number of transfer lines.... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 0 |
16316663-1dbc-4479-a2cd-81f488c62f0c | position can be inferred by using the intensity difference be-
tween foils and by fitting a Gaussian beam profile assuming
known emittance and nominal optics. However, when the
beam falls on one foil only, the position cannot be inferred.
∗adrian.menor.de.onate@cern.chIn this scenario, classical steering algorithms tha... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 0 |
607d1055-d717-4dde-af9a-8da5ff615144 | at the targets.
With this in mind, the following sequential automated
procedures have been tested during commissioning in the
horizontalplaneoftheline: first,steeringfromseverebeam
trajectoryoffsets using Bayesianoptimization(BO);second,
estimatingcorrectorpolaritieswithkickresponsemeasure-
mentsthatarecomparedagainsta... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 0 |
26863db6-6e31-4b42-a486-857dfab9e1b6 | backtoanacceptablesymmetry(heretargetedat0.6),the
transferline isdivided into segments mapping one or more
correctormagnetstothedownstreamSEM(s),locatedbefore
the next corrector. Each of these segments is sequentially
steered by first taking quasi-random actions that explore the
actionspaceofthemagnet,generatingspace-f... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 0 |
bda248fb-ee16-4c33-8c99-a50a2ab18cbf | using two magnets or more, corresponding to the magnet
combinations that result in a good objective value:
Objective =𝑁SEM
∑
𝑗=1⎧{
⎨{⎩𝑆2
𝑗if𝐼tot
𝑗≥ 𝐼min,
0otherwise.(2)17th International Particle Accelerator Conference, Deauville, France JACoW Publishing
ISBN: 978-3-95450-252-3 ISSN: 2673-5350 doi: 10.18429/JACo... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 0 |
7af2885e-1921-4375-8da6-0fca6651e374 | −60−30 0 30 60
Magnet 1 [µrad]−50050Magnet 2 [µrad]Optimization Landscape
Sim optimum
−60−30 0 30 60
Magnet 1 [µrad]RBF ARD Kernel
Random samples (50)
GP optimum
−60−30 0 30 60
Magnet 1 [µrad]Linearly Transformed Kernel
Random samples (50)
GP optimum
0.1150.4630.8111.1591.5061.8542.2022.5502.898
Symmetry ObjectiveFigur... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 1 |
1177bd69-31ff-4f40-98b2-c1473d95ba2b | slope also changes with the polarities of the magnets.
ModelingthisproblemusingaGaussianProcess(GP)(to
beusedassurrogatemodelforBO)withastandardkernel
with axis-aligned lengthscales (ARD) fails to capture the
ridgestructure,assuchkernelscanonlyrepresentanisotropy
along the coordinate axes (see mid plot in Figure 1). To... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 1 |
e82c4dad-0dee-4e28-98e7-300504fe6b23 | across it (the lengthscale governs how rapidly the kernel
correlationdecayswithdistanceineachinputdirection,so
a large lengthscale implies the function varies slowly along
that direction, while a short one implies rapid variation).
The transformation is applied directly to the kernel inputs
as shown in Equation 3 .
𝑘(... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 1 |
5eea0dd7-f613-4f55-be87-cc7c0988d188 | Figure3showsthethreadingresultsusingthisalgorithmic
logic in the real machine, in a large section of the transfer
line referred to as TT20, consisting of 8 SEM devices and 7
correctormagnets. Thetransferlinewasdecomposedinto
differentsegmentsthatweresteeredfromano-beam-reading
5 10 15 20 25 30
BO iteration0.40.50.60.70... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 1 |
ae3dd3c4-66ff-47ec-ad9c-d8012449ddda | value that is safe for the machine, but which results in noisy
readings at the SEMs, representing the most challenging
optimization scenario, and at the same time the most useful
one from an operation’s perspective.
POLARITY ESTIMATION
Once the beam is detected in all SEMs using the logic
describedabove,ahigh-symmetrys... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 1 |
9ec28ae5-8a5a-41fd-8645-670018a3b4f7 | corrector magnet’s polarity 𝑝𝑗∈ {+1,−1} relating each
corrector 𝑗to the accelerator optics model by sequential
perturbation testing. For a given corrector, perturbation
actions of alternating sign 𝛿𝑘(𝑛)
𝑗(where𝑛is the iteration in-
dex) are applied, and the expected response 𝑒𝑛,𝑖= 𝑅𝑖𝑗𝛿𝑘(𝑛)
𝑗17th Inter... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 1 |
8586a4e5-8be5-400d-8d8d-cb19a48d991f | 0.00.20.40.60.81.0SymmetryBeam Symmetry After Threading
Target symmetry
SEM 1
Magnet 1
Magnet 2SEM 2 SEM 3 SEM 4
Magnet 3SEM 5
Magnet 4
Magnet 5SEM 6
Magnet 6SEM 7
Magnet 7SEM 8Figure3: BeamthreadingoftheTT20linesegmentconnect-
ing the SPS to the North Area experimental hall. Corrector
magnetsusedforeachSEMsegmentaresh... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 2 |
667caa62-8fd1-459e-8871-72588953422a | correlation coefficient 𝑟, which is then converted to the test
statistic𝑡 = 𝑟√(𝑁 −2)/(1−𝑟2)and evaluated against the
Student’s 𝑡-distribution to yield a confidence 𝑐. The itera-
tion terminates when 𝑐 ≥ 0.95, and the polarity is accepted
as𝑝𝑗=sign(𝑟), encoding whether the physical corrector
response agrees w... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 2 |
4e982e06-ce18-44c8-903e-23fe70c9bfa0 | lators arranged in these beam lines. These scintillators were
specificallyinstalledforthepurposeoftargetalignment,and
are not a permanent fixture of the beam line. Since the pre-
vious commissioning, the targets’ position may have drifted
orbeenmodifiedcomparedtothepreviousinstallation,re-
sultinginmisalignmentandhence... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 2 |
66c27f16-9701-460d-9807-08e181f2be3a | the resulting beam-target interaction quality. An outer-loopoptimizer then proposes the next SEM position based on
these scintillator readings, and the process repeats.
We optimize the outer loop using BO (where convention-
ally,a grid scan method is used, significantly hindering the
sampleefficiencyoftheoptimization)t... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 2 |
778c5525-2677-4a26-80cc-d77a1925f304 | tion at the H2 line only (a) or both H2 and H4 lines simulta-
neously(b). AxesshowthemovableSEMpositions(mm),
whilethe objectivevalue(proportionalto scintillatorinten-
sitycounts) is color-coded. Data are fittedusing a GP with
RBF kernel and ARD. 𝑥0is the initial objective value.
CONCLUSION
When commissioning the SPS-... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 2 |
34679bd3-a836-49f5-8889-a9fe2408066c | commissioning steps. A BO-based steering algorithm, com-
binedwithquasi-randomsearch,sequentiallyrecoversthe
beamfromano-signalstatetoasymmetryof ≥ 0.6across
all SEMs. Then, the line is steered to high symmetry val-
ues using specialized steering algorithms that rely on the
knowledge of the corrector magnets’ polaritie... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 2 |
bafc4a62-59ea-46d1-9d7f-42bc7fa931fd | criteria to discriminate beam presence from SEM noise.
REFERENCES
[1]A.MenorDeOñate etal.,“TrajectorysteeringforDCbeamsat
theCERNSPSusingreinforcementlearningbasedonintensity
measurements”, in Proc. IPAC’25 , Taipei, Taiwan, Jun. 2025,17th International Particle Accelerator Conference, Deauville, France JACoW Publishin... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 2 |
bd641781-9423-40db-b9af-b089ba232534 | pp. 2928–2930.
doi:1/zero.alt4.18429/JACoW-IPAC2/zero.alt425-THPM113
[2]K.Bernier etal.,“Calibrationofsecondaryemissionmonitors
of absolute proton beam intensity in the CERN SPS North
Area”, CERN, Geneva, Switzerland, CERN-97-07, 1997.
doi:1/zero.alt4.517/zero.alt4/CERN-1997-/zero.alt4/zero.alt47
[3]B. Autinand Y.Marti... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 3 |
2ba8793a-4a13-4680-8728-4aed88a62e19 | theCERNNorthandEastAreas,”CERN,Geneva,Switzerland,
Rep.CERN-ACC-NOTE-2020-0043, 2020.17th International Particle Accelerator Conference, Deauville, France JACoW Publishing
ISBN: 978-3-95450-252-3 ISSN: 2673-5350 doi: 10.18429/JACoW-IPAC2026-WEP6017
MC6.D13: Instrumentation: Artificial Intelligence 2631WEP: Poster sessi... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6017.pdf | 3 |
5bd40eeb-baff-4262-81b4-f3e6c5e93049 | HIGH-DIMENSIONAL BAYESIAN OPTIMIZATION FOR SPARSE
OBJECTIVES: AN APPLICATION FOR AUTOMATED BEAM
COMMISSIONING IN THE LOW ENERGY ION RING AT CERN
B. Rodriguez Mateos∗, T. Argyropoulos, V. Kain,
M. Schenk, M. Slupecki†, European Organization for Nuclear Research, Geneva, Switzerland
Abstract
Recent advances in high-dimen... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 0 |
be8a0f55-d799-44b6-9e9c-970203c98ee3 | axis-alignedsubspacepriors(SAASBO),nestedembeddings
for mixed spaces (Bounce), and length-scale-adapted priors
inregularBayesianOptimization. Ourresults demonstrate
therelativestrengthsofeachmethodinthecontextofpar-
ticle accelerator optimization, where sample efficiency is
critical,theobjectivefunctionexhibitssparsity... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 0 |
930a7265-2402-4270-8461-5d4bf3933df2 | missioning often requires the simultaneous tuning of tens
of parameters spanning transfer lines, injection elements,
radio-frequency (RF) systems, and correction magnets, a
tasktraditionallyperformedmanuallybyexpert operators
through iterative adjustments. As facilities become increas-
inglycomplexanddemandever-higherp... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 0 |
8e53db45-79d8-4b19-9c20-d797ba45c023 | when each evaluation corresponds to a machine cycle [ 7].
∗borja.rodriguez.mateos@cern.ch
†also at Department of Communications and Computer Engineering, Uni-
versity of MaltaHowever, the performance of standard BO with Gaussian
Process (GP) surrogates is known to degrade rapidly as the
dimensionalityofthesearchspacegr... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 0 |
33bb877d-440e-4e5b-898c-ac8f7581e4e3 | nellengthscales,assumingonlyasmallsubsetofdimensions
isrelevant. NestedembeddingmethodssuchasBounce[ 12]
progressivelybuildlow-dimensionalembeddings,originally
designed for combinatorial and mixed spaces but adapt-
able to continuous problems. Recent work [ 13] has also
shown that standard BO can be made competitive in... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 0 |
f2eae8d3-b6d6-4171-be2b-88efc52990d8 | Proton Synchrotron [ 16]. Its high repetition rate, low beam
energy, and the regular need for re-tuning due to Linac3
stripperfoilaging[ 17]makeitapracticalplatformfordevel-
oping and validating data-driven tuning strategies. Previous
work at LEIR demonstrated that standard BO with Upper
ConfidenceBound(UCB)andLogNoisy... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 0 |
455dc450-20da-4cba-ad28-ccff83234669 | unoptimizedinitialconfigurationandwithoutpriorknowl-
edge. We consider two operationally relevant beam types,
the NOMINAL cycle with 43 tunable parameters and the
EARLYcyclewith38,bothoptimizedwithrespecttotheac-
cumulatedbeamintensitybeforeRFcapture,measuredwith17th International Particle Accelerator Conference, Deauv... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 0 |
68f3fad8-ff77-4695-9a76-d2d2e9ab1150 | theringbeamcurrenttransformer(BCT).Allalgorithmsare
evaluatedundermatchedinitializationbudgetsandidentical
objective functions to provide a fair comparison, offering
practical guidance for selecting HDBO methods in accelera-
tor commissioning.
METHODOLOGY
Optimization objective
LEIRbeamintensityismaximizedbyadjustingth... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 1 |
908208da-bb68-4eea-8712-aa3e53b75b04 | the intensity of the j-th Linac3 pulse measured at the end of
theLinac3-to-LEIRtransferline. Normalizingbythetotal
deliveredchargeisolatesthecontributionofthecontrolled
parameters from upstream fluctuations originating in the
ECRISsource[ 18]andalongthelinac. Thisisoperationally
critical: ionsourceafterglowintensity,tr... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 1 |
ea5be7b5-4043-404c-9b0d-cc326e95ba03 | correctors, the fourorbit bump settings around the electron
cooler, the injection bump settings, the electron gun voltage
program defining the electron energy, and the LEIR tune
and chromaticities at flat bottom.
Bayesian Optimization framework
All BO-based methods considered in this work share the
sameskeleton[ 2,7]. ... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 1 |
4e61a802-6483-4d6b-b91c-9c5f4bf18efc | function. Throughout this work we use LogNEI [ 20] unless
statedotherwise,foritsnumericallystablegradientsinnoisy
regimes. Initialization is common to all GP-based methods:
half of the budget is drawn from a Sobol low discrepancy
sequence[ 21]andtheremaininghalffromapureexploration
acquisition 𝛼PE(x) = 𝜎(x), followin... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 1 |
09e205c2-d02b-40fb-8652-218c76883bb6 | normal prior on each ARD lengthscale,
𝑙𝑖∼ ℒ𝒩(𝜇0+log𝐷
2, 𝜎0), (1)
whereℒ𝒩denotes the log-normal distribution. This
makestheprior-expectedlengthscalegrowwith √𝐷,counter-
acting the shrinkage of fitted lengthscales as dimensionality
increases. Weuse 𝜇0=√2and𝜎0=√3asrecommended.
The method retains a single global ... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 1 |
52a668ce-fe42-4004-9d11-0e3d726c8d0c | keeps a single global GP but imposes a hierarchical spar-
sity prior on the per-dimension inverse squared lengthscales
𝜌𝑖= 𝑙−2
𝑖, with𝜏 ∼ ℋ𝒞+(𝛼),𝜌𝑖∣ 𝜏 ∼ ℋ𝒞+(𝜏), where
ℋ𝒞+is the half-Cauchy distribution and 𝜏a global shrink-
ageparameter. This horseshoe-style prior concentrates pos-
teriormassonconfigurati... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 1 |
6c6c2c20-fcf8-425e-826e-f17fe9f64b9a | heretothepurelycontinuouscasewithinitial 𝑑𝑒= 25based
on preliminary scans.
EARLY cycle optimization (38 parameters)
Figure1showsthetracesforthe38-dimensionalperfor-
mance optimization of the EARLY cycle. CMA-ES con-
vergesmostrapidlyinthefirst ∼25cycles(inpartbyskip-
ping the initial exploration budget), reaching an ... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 1 |
409a2185-4d7d-4e36-855d-3d4f1b7ff017 | ISBN: 978-3-95450-252-3 ISSN: 2673-5350 doi: 10.18429/JACoW-IPAC2026-WEP6018
MC6.D13: Instrumentation: Artificial Intelligence 2633WEP: Poster session WEP6018
Content from this work may be used under the terms of the CC BY 4.0 licence (© 2026). Any distribution of this work must maintain attribution to the author(s), t... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 1 |
775bf248-75f9-4f4c-af4a-9189ac924925 | initialprogressonsmoothlandscapesbutlackstheglobal
probabilistic model needed to escape local optima. TuRBO
andBouncereachthehighestfinalefficiencies( ≈0.42),with
TuRBOcrossingtheoperationaltargetof 2×1010charges
in under 100 cycles. Lengthscale-rescaled standard BO is
competitivebutconvergestoaslightlylowerplateau( ≈0... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 2 |
3ad63231-f0c7-472f-ae0c-cab57deafe43 | transverse, longitudinal, and cooler-bump parameters all
contribute meaningfully to the objective.
Figure 1: Highest observed injection efficiency (dashed,
right axis) and ring intensity (solid, left axis) vs. cycle num-
berforthefivealgorithmsontheEARLYcycle(38param-
eters). All runs were recorded under similar experi... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 2 |
2757985d-f430-408f-9c11-0a1a5b9f0aed | the sense described above: each per-pulse trim can indepen-
dentlysteerthatpulseoutoftheLEIRacceptance,producing
aflatzerocontributiontotheobjectiveoverasubstantialfrac-
tionofitsrange. Figure 2showsthecorrespondingtraces.
TuRBOclearlyoutperformsallotheralgorithms,reaching
theoperationaltargetof 12×1010charges(or0.27in... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 2 |
1d91581f-a257-453d-8930-8fb77bb1bccf | mixing of a sparse direction with non-sparse directions can
effectivelyblockexplorationoftherelevantsubspace. Asthe
numberofsparsedirectionsgrows,theprobabilitythatthe
embeddingalignsfavorablywith allofthem dropssharply,
making Bounce increasingly fragile in higher dimensions.
CMA-ES again converges quickly in the earl... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 2 |
5523a306-0a9d-414e-81ef-7c3650fb0bce | other methods, on the order of minutes per step on our hard-
ware. Machinecyclescontinuetoelapseduringinference,
so SAASBO effectively operates at a lower duty cycle in
wall-clock time. The cycle counts in Fig. 2therefore un-
derestimate its real-time disadvantage; the wall-clock gap
toTuRBOisevenlarger. Thismattersope... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 2 |
0edcf5e8-5c8b-4440-b0cb-4e4405015354 | computational overhead, and is retained for LEIR beam
commissioning. Its trust region structure suits the injec-
tionlandscape: fittinglocalGPsonlywithinacontracting
hyper-rectangle avoids the global identifiability problems
caused by sparse directions while still exploiting the ob-
jective’s smoothness in the high-rew... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 2 |
429c0bf1-2e67-4e8e-bc12-b9ab7bc58f20 | REFERENCES
[1]A. Edelen et al., “Opportunities in Machine Learning for
Particle Accelerators”, Nov. 2018.
doi:1/zero.alt4.4855/zero.alt4/arXiv.1811./zero.alt43172
[2]R.Roussel etal.,“Bayesianoptimizationalgorithmsforac-
celerator physics”, Phys. Rev. Accel. Beams , vol. 27, no. 8,
pp. 084801, Aug. 2024.
doi:1/zero.alt4... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 3 |
8b313c17-adc6-42d2-a11e-f72561eb104a | doi:1/zero.alt4.18429/JACoW-IPAC2/zero.alt425-THPM1/zero.alt49
[5]J. Duriset al., “Bayesian Optimization of a Free-Electron
Laser”,Phys. Rev. Lett. , vol. 124, no. 12, pp. 124801, Mar.
2020. doi:1/zero.alt4.11/zero.alt43/PhysRevLett.124.1248/zero.alt41
[6]S. Hirlaender and N. Bruchon, “Model-free and Bayesian
Ensemblin... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 3 |
4994e182-4a59-4d31-8ba5-ea6b0ec62a58 | dimensional Bayesian optimization of discrete sequences”,
Nov. 2024. doi:1/zero.alt4.4855/zero.alt4/arXiv.24/zero.alt46./zero.alt44739
[9]R. Bellman, Dynamic Programming . Dover Publications,
1957.
[10]D. Eriksson et al., “Scalable Global Optimization via Local
Bayesian Optimization”, Feb. 2019.
doi:1/zero.alt4.4855/ze... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 3 |
d889a2b6-7125-42ca-a8bf-5e5915e8f454 | pp. 1764-1793.
[13]C. Hvarfner, E. O. Hellsten, and L. Nardi, “Vanilla Bayesian
Optimization Performs Great in High Dimensions”, in
Proc. 41st International Conference on Machine Learn-
ing, vol. 235, PMLR, Jul. 2024, pp. 20793-20817. https:
//proceedings.mlr.press/v235/hvarfner24a.html
[14]N. Hansen, “The CMA Evolutio... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 3 |
61b7186b-fcc9-45e2-b2e0-10f455215e97 | V-3, CERN-2004-003, Dec. 2004.
doi:1/zero.alt4.517/zero.alt4/CERN-2/zero.alt4/zero.alt44-/zero.alt4/zero.alt43-V-3
[17]S. Hirlaender et al., “LEIR Injection Efficiency Studies as
a Function of the Beam Energy Distribution from Linac3”,
inProc. IPAC’18 , Vancouver, Canada, Apr.-May 2018, pp.
758–760. doi:1/zero.alt4.184... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 3 |
5a89fc72-9f57-46f7-803e-ddec26e88c1a | Improvement for Bayesian Optimization”, Jan. 2025.
doi:1/zero.alt4.4855/zero.alt4/arXiv.231/zero.alt4.2/zero.alt47/zero.alt48
[21]I. M. Sobol’, “On the distribution of points in a cube and the
approximateevaluationofintegrals”, USSRComputational
MathematicsandMathematicalPhysics ,vol.7,no.4,pp.86-
112,1967. doi:1/zero.... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 3 |
a22cfe9d-44c5-4597-9f1c-e1a1a555e539 | ISBN: 978-3-95450-252-3 ISSN: 2673-5350 doi: 10.18429/JACoW-IPAC2026-WEP6018
MC6.D13: Instrumentation: Artificial Intelligence 2635WEP: Poster session WEP6018
Content from this work may be used under the terms of the CC BY 4.0 licence (© 2026). Any distribution of this work must maintain attribution to the author(s), t... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6018.pdf | 3 |
d4a52ca7-2625-4d3d-b723-17e8fb0dffee | ONLINE REINFORCEMENT LEARNING FOR STRIPPER FOIL AGING
COMPENSATION AT THE CERN LOW ENERGY ION RING
B. Rodriguez Mateos∗, T. Argyropoulos, V. Kain, A. Lu, A. Menor de Onate,
M. Schenk, M. Slupecki†, European Organization for Nuclear Research, Geneva, Switzerland
Abstract
Stripper foil degradation at the CERN Low Energy ... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6021.pdf | 0 |
47ec7b30-cc27-4c72-bf2d-e9df2dd96e98 | degradation and maintain ring performance. The controller
observes longitudinal Schottky spectra encodings and time-
of-flight measurements from the linac to adjust the ramping
and debunching cavity phases, and electron cooler gun and
orbit bump in real time. We demonstrate that pre-training
theagentin adata drivensurr... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6021.pdf | 0 |
a6325003-3e8f-4a59-8a02-391ec5b72c03 | high-dispersionregion. Betweeninjections,electroncooling
compressesanddragsthecirculatingbeamlongitudinallyto
freemomentumspaceforthenextpulse. Afteraccumulation,
the coasting beam is captured into bunches and accelerated
to72MeV/uforextractiontowardtheProtonSynchrotron.
The quality of this process depends critically o... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6021.pdf | 0 |
bc53887c-8fc4-42f9-ad3b-2b8f03c713d2 | extreme cases, accumulated intensity drops by more than
a factor of two, requiring foil replacement approximately
every two weeks [ 1]. Historically, these drifts have been
∗borja.rodriguez.mateos@cern.ch
†also at Department of Communications and Computer Engineering, Uni-
versity of Maltacorrectedmanuallybyadjustingth... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6021.pdf | 0 |
7b826972-4731-4bb3-9ba8-8691016369d7 | intoacompactlatentrepresentation,whilemulti-layerper-
ceptrons learned both an intensity model and a dynamics
modelpredictingstatetransitionsasafunctionofparameter
changes. Several RL algorithms were trained offline on this
surrogateandshowntorecovernominalintensityinfewer
than20stepswhenthemachinewasmoderatelydetuned,... | D:\Daily_ihep\Conferences\IPAC2026\WEP\WEP6021.pdf | 0 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.