BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Statistical learning theory for scientific applications: an overvi
ew
DTSTART;VALUE=DATE-TIME:20181004T143000Z
DTEND;VALUE=DATE-TIME:20181004T145000Z
DTSTAMP;VALUE=DATE-TIME:20201125T051603Z
UID:indico-contribution-5062-28910@agenda.infn.it
DESCRIPTION:Speakers: JESUS VEGA (Laboratorio Nacional de Fusión. CIEMAT)
\nThe statistical learning theory allows the estimation of functional depe
ndency from a given collection of data. It includes discriminant analysis\
, regression analysis and the density estimation problem. It is used to ob
tain data-driven models to relate heterogeneous quantities with the aim of
making predictions. In science\, statistical learning is an essential the
ory to find relationships among quantities whose formulation cannot be ded
uced from first principles. This talk summarizes the use of statistical le
arning methods in scientific problems: from unsupervised techniques to sup
ervised techniques\, from simple predictions to probabilistic predictions\
, from non-parametric estimations to parametric estimations\, from real-ti
me needs to off-line needs and from standard datasets of information to pr
ivileged datasets of information. Concepts and examples will be given.\n\n
https://agenda.infn.it/event/15217/contributions/28910/
LOCATION:INFN-LNF\, Italy Bruno Touschek Auditorium
URL:https://agenda.infn.it/event/15217/contributions/28910/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Causality detection methods for time series analysis
DTSTART;VALUE=DATE-TIME:20181004T153000Z
DTEND;VALUE=DATE-TIME:20181004T155000Z
DTSTAMP;VALUE=DATE-TIME:20201125T051603Z
UID:indico-contribution-5062-28894@agenda.infn.it
DESCRIPTION:Speakers: Teddy Craciunescu (National Institute for Lasers\, P
lasma and Raadiation Physics\, Bucharest\, Romania)\nCoupling and synchron
ization are common phenomena that occur in nature\, e.g. in biological\, p
hysiological and environmental systems\, as well as in physics and enginee
red systems. Depending on the coupling strength\, the systems may undergo
phase synchronization\, generalized synchronization\, lag synchronization
and complete synchronization [1]. Lag or intermittent lag synchronization\
, where the difference between the output of one system and the time-delay
ed output of a second system are asymptotically bounded\, is the typical c
ase of the fusion plasma instability control by pace-making techniques [2-
3]. The major issue\, in determining the efficiency of the pacing techniqu
es\, resides in the periodic or quasiperiodic nature of the plasma instabi
lities occurrence. After the perturbation induced by the control systems\,
if enough time is allowed to pass\, the instabilities are bound to reoccu
r. Therefore\, it is crucial to determine an appropriate interval over whi
ch the pacing techniques can have a real influence and an effective trigge
ring capability.\nSeveral independent classes of statistical indicators de
veloped to address this issue are presented. The transfer entropy [4] is a
powerful tool for measuring the causation between dynamical events. The a
mount of information exchanged between two systems depends not only the ma
gnitude but also the direction of the cause-effect relation. Recurrence pl
ots (RP) is an advanced technique of nonlinear data analysis\, revealing a
ll the times when the phase space trajectory of the dynamical system visit
s roughly the same area in the phase space [5]. RP refinement\, called joi
nt recurrence plots (JRPs)\, can be used to relate the behavior of one sig
nal with the one of another. Convergent cross mapping (CCM) [6]\, tests fo
r causation by measuring the extent to which the historical record of one
time series Y values can reliably estimate states of another time series
X. CCM searches for the signature of X in Y’s time series by detecting
whether there is a correspondence between the “library” of points\, in
the attractor manifold built from Y\, and points in the X manifold. The t
wo manifolds are constructed from lagged coordinates of the two time-serie
s. The Weighted Cross Visibility Graph (WCVG) method [7] starts from the i
dea of mapping the coupled time series into a complex network [8] and eval
uates its structural complexity by mean of the Shannon entropy of the modi
fied adjacency matrix (constructed by weighting the connections with the m
etric distance between two connected values in the time series). \nReferen
ces:\n[1] S. Boccaletti et al\, Phys. Rep.\, 366 (2002) 1.\n[2] A. M
urari et al\, Nucl. Fusion 56 (2016) 076008.\n[3] A. Murari et al\, Nuc
l. Fusion 57 (2017) 126057. \n[4] T. Schreiber\, Phys. Rev. Lett. 85 (2
000) 461.\n[5] N. Marwan et al\, Phys. Rep. 438 (2007) 237–329\n[6]
G. Sugihara et al\, Science 338 (2012) 496–500.\n[7] A. Murari et a
l\, Entropy 19-10 (2017) 569.\n[8] S. Mehraban et al\, EPL 103 (2015) 5
0011.\n\nhttps://agenda.infn.it/event/15217/contributions/28894/
LOCATION:INFN-LNF\, Italy Bruno Touschek Auditorium
URL:https://agenda.infn.it/event/15217/contributions/28894/
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Model Falsification Approach to Learning in Non-Stationary Envir
onments for Experimental Design
DTSTART;VALUE=DATE-TIME:20181004T145000Z
DTEND;VALUE=DATE-TIME:20181004T151000Z
DTSTAMP;VALUE=DATE-TIME:20201125T051603Z
UID:indico-contribution-5062-28890@agenda.infn.it
DESCRIPTION:Speakers: Andrea Murari (RFX Consortium and PMU)\nThe applicat
ion of machine learning and advanced statistical tools to complex physics
experiments becomes very problematic\, when the i.i.d. (independent and id
entical distribution) hypothesis is not verified\, due the varying conditi
ons of the systems to be studied. In particular\, new experiments have to
be executed in unexplored regions of the operational space. As a consequen
ce\, the input quantities used to train and test the performance of the to
ols are not necessarily sampled by the same probability distribution as in
the final applications. In the present study\, a new data driven methodol
ogy is proposed to guide planning of experiments and to explore the operat
ional space. The approach is based on Symbolic Regression via Genetic Prog
ramming to the available data\, which allows identifying a set of candidat
e models. The confidence intervals for the predictions of such models per
mit to find the best region of the parameter space for their falsification
\, where the next set of experiments can be more profitably carried out. T
he procedure is repeated until convergence on a satisfactory model. Extens
ive numerical tests and applications to the scaling laws in Tokamaks prove
the viability of the proposed approach.\n\nhttps://agenda.infn.it/event/1
5217/contributions/28890/
LOCATION:INFN-LNF\, Italy Bruno Touschek Auditorium
URL:https://agenda.infn.it/event/15217/contributions/28890/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Adaptive Learning for Disruption Prediction
DTSTART;VALUE=DATE-TIME:20181004T151000Z
DTEND;VALUE=DATE-TIME:20181004T153000Z
DTSTAMP;VALUE=DATE-TIME:20201125T051603Z
UID:indico-contribution-5062-28877@agenda.infn.it
DESCRIPTION:Speakers: Michela Gelfusa (Department of Industrial Engineerin
g\, University of Rome “Tor Vergata”\, via del Politecnico 1\, Roma\,
Italy)\nAccurate prediction of catastrophic events is becoming an importan
t area of investigation in many research fields. In Tokamaks\, detecting d
isruptions with sufficient anticipation time is a prerequisite to undertak
ing any remedial strategy\, either for mitigation or for avoidance. Tradit
ional predictors based on machine learning techniques can be very performi
ng\, if properly optimised\, but tend to age very quickly. Such a weaknes
s is a consequence of the i.i.d. (independent an identically distributed)
assumption on which they are based\, which means that the input data are i
ndependent and are sampled from exactly the same probability distribution
for the training set\, the test set and the final actual discharges. These
hypotheses are certainly not verified in practice\, since nowadays the ex
perimental programmes of fusion devices evolve quite rapidly and metallic
machines are very sensitive to small changes in the plasma conditions. Thi
s paper describes various adaptive training strategies that have been deve
lop to preserve the performance of disruption predictors in non-stationary
conditions. The proposed techniques are based new ensembles of classifier
s\, belonging to the CART (Classification and Regression Trees) family. Th
e improvements in performance are remarkable and the final predictors sati
sfy the requirements of the next generation of experimental devices.\n\nht
tps://agenda.infn.it/event/15217/contributions/28877/
LOCATION:INFN-LNF\, Italy Bruno Touschek Auditorium
URL:https://agenda.infn.it/event/15217/contributions/28877/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Determining the causality horizon in synchronization experiments
DTSTART;VALUE=DATE-TIME:20181004T155000Z
DTEND;VALUE=DATE-TIME:20181004T161000Z
DTSTAMP;VALUE=DATE-TIME:20201125T051603Z
UID:indico-contribution-5062-28876@agenda.infn.it
DESCRIPTION:Speakers: Emmanuele Peluso (University of Rome Tor Vergata)\nS
ynchronization can be defined as the coordination of events to operate a s
ystem in harmony. It is important in the operation of manmade systems and
in the investigation of natural events. In the last decade or so\, synchro
nization of multiple interacting dynamical systems has become a lively fie
ld of study for control. In Magnetic Confinement Nuclear Fusion\, various
pacing concepts have been recently proposed to control various instabiliti
es\, such as sawteeth and ELMs. Some of the main difficulties of these exp
eriments is the quantification of the synchronization efficiency and its r
ole in the understanding of the main physical mechanisms involved. In this
paper\, various classes of independent of statistical indicators are intr
oduced to address these issues. In metallic Tokamaks\, one of the most rec
ent applications of ICRH heating on JET is sawtooth control by ICRH modula
tion\, for avoiding triggering dangerous NTM and counteracting impurity ac
cumulation. Various forms of ELM pacing have also been tried to influence
their behaviour using external perturbations and the one studied in this p
aper\, injection of pellets\, seems the most promising in the perspective
of future devices. In the application to JET experiments with the ILW\, th
e proposed indicators provide sound and coherent estimates of the efficien
cy of the synchronisation scheme investigated. They also confirm the inter
pretation that the fast ions play a fundamental role in the stabilization
of the sawteeth\, in both L and H mode.\n\nhttps://agenda.infn.it/event/15
217/contributions/28876/
LOCATION:INFN-LNF\, Italy Bruno Touschek Auditorium
URL:https://agenda.infn.it/event/15217/contributions/28876/
END:VEVENT
END:VCALENDAR