This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2020.3003726, IEEE Access DigitalObjectIdentifier Decision and Reasoning in Incompleteness or Uncertainty conditions GERARDOIOVANE1,PATRIZIADIGIRONIMO2,MARTACHINNICI3,ANTONIORAPUANO1 1DipartimentodiInformatica,UniversityofSalerno,Italy(e-mail:[email protected],[email protected]) 2DipartimentodiMatematica,UniversityofSalerno,Italy(e-mail:[email protected]) 3EnergyTechnologiesDepartment,ICTDivision(DTE-ICT),ENEA,Casaccia,Rome,Italy(e-mail:[email protected]) Correspondingauthor:MartaChinnici(e-mail:[email protected]). ABSTRACT Whenwetrytosolveneworknownproblemstowhichwewanttogivenewsolutions,we createnewknowledgeandrealizenewdiscoveries.Todate,thescientificmethodshaveusedtheprobability in order to analyze problems, make inference and build forecasts. However, everyone agreed that most problemsdonotfollowstandardprobabilisticrules.Inthisstudywewillbuildanuncertaintylogicbyusing the concept of probability, with those of plausibility, credibility and possibility. We will provide several models which treats uncertainty information and allow to perform more reliable forecasts. After that, we willprovethemodelsreliabilitythroughafinalsimulationontheBiometricsandSportfieldsusingoneof themodels;thesesimulationarefullyreplicabileforeachfieldandforeachoftheprovidedmodels. INDEXTERMS Biometrics,Credibility,Incompleteness,Plausibility,Possibility,Probability,Sportsodds, Uncertainty. I. INTRODUCTION information not only uncertain, but also paradoxical1. This reasoning has been fully studied: Cuzzolin in [1] defines THE plausibility is a concept used in many inferential thepropertiesofplausibility.Friedman,HalpernandBarnett contexts and processes. A plausible inference theory calculatedandmeasuredtheplausibility[2][3].Thisconcept goesthroughapreliminarydefinitionofwhatis"uncertain" has been used in real world applications like [4] as well. and how it affects the inferential process. A plausible inter- While Shved, Wanga and Wen provide a way to measure ferencecanbeinvalidated,correctedorsimplyupdatedatthe uncertainty[5][6][7].Inthispaper,wearegoingtodefinea entryofnewdataorinformation.Inthissense,plausibilityis newperspectiveontheplausiblereasoning. notonlynon-monotonous,butoperatesinanopenconceptual space. The data, the available information and inferential rules are of a temporary nature and reviewable so, they are II. CONCEPTUALOVERVIEW constantlychanging. This section in inspired by the work of author in [8]. The Polya (1954) defines the plausible reasoning as opposed to formulationofatheoryofplausibilitywhichhasbeenknown thedemonstrativereasoning.Theplausibleinferencehelpsto in antiquity is that of Sextus Empiricus. Truthly, he at- acquirenewknowledge,usingtheinductionsandtheanalogy tributesthistheorytoCarneadeofCyrene,inhiscomposition which are not certain and do not transmit the truth. The "AgainsttheLogicians".Plausibleheremeans"topersuade". plausibleinferencehasafluid,temporaryandriskybehavior. Thecriteriaare: Rescher(1976),conceivestheplausiblereasoningasaform of deduction starting from uncertain premises, aimed at • somethingisplausibleifitseemstobetrue, the treatment of "cognitive dissonances". In this theory the • is even more plausible if it seems to be true and is compatible with other things that seem to be true (i.e. premises and conclusions are provisional, controversial and itisstable), fluid. For Dezert (2002), the plausible reasoning is an extension • isstableandistested. of the evidence theory of Dempster-Shafer, and therefore of the Bayesian probability. His theory allows to deal with 1admittingcontradictoryinformationthankstothe"hyper"set-power" VOLUME4,2016 1 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2020.3003726, IEEE Access Authoretal.:PreparationofPapersforIEEETRANSACTIONSandJOURNALS So it is not necessary that something be true or believed, Pr{E/H} and therefore "have a numerical value, equal to to be plausible. It simply has to satisfy some requirements, the probability that an event of the type predicted by E will so as to allow something to be assessed as in accordance or happen,computedonthebasisofthestatisticalhypothesisH" consistentwithrelatedfacts. although"credibilityandprobabilityaredifferentlydefined" Uncertaintycanbeoftwotypes: [10].WhenwedealwiththeplausibilityofA,wearedealing • Aleatory (also known as stochastic), when it concerns withthereliabilityofAandthestrengthofevidenceinfavor therandomnessoftheobjectofknowledgesystem; ofA,inotherwords,ourtrustinA.Credibilityisexpressed • Epistemic(alsoknownassubjective),i.e.whenitrefers in probabilistic terms, readjusting the Bayes’ theorem with tothestateofknowledgeofthesubjectoverthesystem. the substitution of probability (Pr) with credibility (Cr). In this view, plausibility and probability are indistinguishable, Theplausibilityuncertaintybelongstothesecondtype;it astheformercanbereducedtothesecond;howeveritisnot isanexpressionofasubjectiverelationship. always possible to compare numerically the plausibility of AccordingtoWalton[9],itisacontroversialquestionifthe twodifferentevents.TheprobabilisticconceptionofPolyais plausibilityisdifferentfromtheprobability,anditisdifficult affectedbyallthelimitationsoftheBayesianconception. to rule out that plausibility can be resolved in some special caseofprobability. In order to model the epistemic uncertainty with the proba- B. DEMPSTER-SHAFER bility,theCox’stheoremisused.Itestablishestheexistence To overcome the limits of the bayesian conception and the of a precise relationship between the notion of plausibility, relativevisionofplausibleinference,theevidencetheoryof plandthetheoryofprobabilitycalculation,pr.Thistheorem Dempster-Shaferhasbeendefined,hereafterdenotedasDS. proves that any system of plausible inferences, which is not Also for this theory the bayesian conception is the basis for isomorphictoprobabilitytheory,mustviolateatleastoneof theplausibilitynotion.Inparticular,it"includesthebayesain thefollowingconditions: theoryasaspecialcaseandthereforepreservesatleastsome of the attractions of that theory" [12] [13]. The DS theory renounces the one-dimensional representation of belief: in • pl(A|X)=0↔AfalseinX otherwords,itrenouncestoassigntheremainingportionof • pl(A|X)=1↔AtrueinX belieftothenegationoftheproposition.TheDStheorystarts • 0≤pl(A|X)≤1 from what is called a discernment or universe of discourse • pl(A∧B|X)=pl(A|X)pl(B|X) u, which is a series of mutually exclusive alternatives. The • pl(¬A|X)=1−pl(A|X), DS theory assigns a basic probability, a belief function and where pl (A|X) indicates the plausibility of A over X. If it a plausibility function to the u subsets which are called doesnotviolateanyoftheconstraintsitispossibletoreplace propositions.Wedefinethefollowingassignmenttoabasic pr a pl and use the probability calculation without losing probability: information. m:2u →[0,1]suchthatm()=0,(cid:80) m(A)=1 A⊆u A. POLYA which represents and expresses numerically the force of AccordingtoPolya,theprototypesofplausibleinterference evidence,theexactbeliefthatanagenthasinA. aretheanalogyandtheinduction;infact"theanalogyandthe The function m: 2u → [0,1] is called belief function Bel, particular cases are the most abundant sources of plausible whenitsatisfiestheconditions: arguments"[10][11].Inparticular"theinferencebyanalogy seems the most common and essential form of inference. It • Bel()=0, produces more or less plausible conjectures which may or • Bel(u)=1. may not be confirmed by experience or by more rigorous Plausibilitycannowbedefined.Weobservethat"aproposi- reasoning". The plausible argument, as defined by Polya, is tion is plausible in the light of evidence to the point where notdefinitive:itproducesprovisionalconclusions.Theplau- the evidence does not support its opposite" [12]. Then we sibilisticreasoningisnotsubjective:themodelsofreasonings introduce the dubious function, Dub, placing it as equal to areimpersonal,whileitistheforceoftheconclusionswhich Cr(¬A) where ¬A is the complement of A in u and we is of a subjective nature and therefore not representable by define the function of plausibility, Pl, as Pl (A) = 1-Dub means of quantity. As for the fundamental inductive model, (A) = 1 (Cr¬A). The plausibility function expresses what they have a correspondence in the demonstrative logic. In shouldbebelievedinpropositionAifallthenotknownfacts [10], the impersonal rules which establish some kind of at present state support A; it also expresses the maximum evidencearedescribed.Whileitispersonalandsubjectiveto probabilistic value that can be allocated to a proposition establish"whetheratesthasasufficientweightornot".For A: in particular "it measures the total mass of belief that Polya the probability is the core in the plausibility notion. can be moved to A" [14]. The lower limit corresponds to The Bayes theorem is the foundation of Polya’s concept of the belief function, and therefore the relation Cr (A) ≤ Pl plausibility. The credibility of B is the confidence of A if B (A) is valid. The difference between belief and plausibility is true. In fact, credibility can be a conditional probability is also reflected at the functional level: indeed the belief 2 VOLUME4,2016 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2020.3003726, IEEE Access Authoretal.:PreparationofPapersforIEEETRANSACTIONSandJOURNALS function"isoftenzeroforallatomicpropositionsincomplex propositionsare"candidates"tothetruthornot.Anargument domains, unless a large number of trials are available". The plausibility can therefore be defined as the maximum value otherbigdifferencebetweenbayesiantheoryandDStheory amongtheminimumplausibilisticvaluesoftheentemematic is the evidence combination, that is the belief updating rule integrations which allows a deductive conclusion derivated whenthereisanewevidence.Infact,intheDS’theorythe fromthepremises. combinationrulesubstitutesthebayesianrule.Thecombina- tionruleisalsocalledorthogonalsum.TheDScombination E. NOTMONOTONICITYANDPLAUSIBILITY rulesuffersfromsomewell-knownstructurallimitations:in Anothernon-probabilisticapproachaimedattreatingplausi- fairlysimplesituationsitcouldgenerateunexpectedandnon- bleinferenceisnon-monotonouslogics’one.Thisapproach intuitiveresults,whichstronglylimiteffectiveness[15]. is characterized by the violation of one of the strongest conditionsofclassicallogic,knownas"monotonicity".Ifthe C. DEZERT-SMARANDACHE conclusion ϕ is a consequence of a premises set Γ, then it TheDezert-Smarandache’stheory[16],hereafterdenotedas remains a consequence of any premises set ∆ that contains DSm, has born as an explicit attempt to overcome some Γasitssubset.Accordingtothiscondition,aconclusioncan difficultiesoftheDStheory.Plausibilityisanupperlimitof not be invalidated by the addition of new information, it re- the probabilistic value which a proposition can assume. Ds’ mainstrueonceandforall,regardlessofthepremisesthatwe plausibility theory conceives plausible reasoning as a con- canaddtothepremisessetΓ.Thevalidpropositionsnumber ceptualmodelforguidingdecision-makingunderuncertainty increasesmonotonouslytothepremisesaddedtoΓincrease. conditions,whichisbroadlydefinedrespectingtheDS’the- Ifthepropositionscanbeinvalidatedbytheadditionofnew ory.Inparticular,theuncertaintyreferredbyDezertextends premises,thevalidassertionsnumbermaynotonlyincrease, the one conceived in probability theory and DS’ theory. A but may also decrease. The non-monotonous logic is based situationoraninformationstateis"rational"whenthebasic on a reduced vision of uncertainty and it simply treats in- assignmentmisasumoneandtheclosureoftheoperators∩ ferences with exceptions. Inference forms modeled by non- and∪ontheuniverseelementsis0;itisstrictly"uncertain" monotonic logics are uncertain as they produce provisional when it is a sum one, and the closing of the operator ∪ conclusionswhichcanbeinvalidatedorreviewedinthelight can be different from zero; it is "paradoxical" when it is of new information available, and this is a big limitation. sum one and the closing of the operator ∩ can be different Furthermore, the non-monotonous approach conceives the fromzero;itis"uncertainandparadoxical"whentheclosure plausiblereasoningasessentiallyaddressedtotheinferences of both operators ∩ and ∪ can be different from zero. The treatmentpropertoeverydayexperience.Thus,itneglectsthe assumptions’modificationproduceseffectsonthedeveloped wholepartoftheplausibleinferencedirectlyconnectedtothe conceptual model, but does not allow to exceed its limits. scientific disciplines, including Mathematics. As theorized AlthoughDezerttendstoamoreextensivediscussionofthe byPolyaandCellucci[18],theplausibleinferenceisfullof plausible reasoning, his approach contributes to leaving it experimentalreasoningthattheclassicallogicdoesnotallow anchoredtoastronglyreductionistview. tohandle. D. THERESCHER’SDEDUCTIVISTCONCEPTION F. THECOLLINS-MICHALSKI’SCOGNITIVEAPPROACH Among the non-probabilistic approaches to plausibility the- The cognitive approach to plausibility aims to show how ory,Rescher’sdeductivistapproachstandsout.Accordingto factors of a subjective and psychological nature can play Nicholas Rescher, the "plausibility theory aims to provide a decisive role in the process that leads to a choice. The a rational instrument to treat cognitive dissonances" [17]. cognitiveapproachtoplausibilityismodeledintheCollins- Rescherprovidesapreliminarydistinctionbetweenplausibil- Michalscki’s plausibility theory hereafter denoted as C-M. ityandprobability:thefirst"classifiespropositionsaccording It intends to formalize the plausible inferences which fre- tothestatusoftheevidentiarysourcesorthevalidatingprin- quently recur in people’s responses to questions for which ciples that guarantee them", while "the probability weights theydonothaveareadyanswer[19]. variousalternativesandevaluatesthem".Insteadofadopting Information is defined as common knowledge to a group of theprobabilitytheoryasabasisformodelingtheplausibility people if: all the group members know it; they know that notion, he bases his approach on the traditional Theophras- othersknowit;theyknowthatothersknowthattheyknowit; tus’ principle, pointing to a return to the historical roots of andsoon.Itisdifferentfromthesimplemutualinformation, the plausibility theory. It never develops calculations which "which implies only that specific possession of information merge quantities into new results, but proceeds only with andnoteventheawarenessoftheothers’knowledge".That comparisons between different degrees of data. A further mutual knowledge may become a common knowledge, in aspectofthedeductivisticapproachtotheplausibilitynotion presence of an "independent arbitrariness" to the various is the link with the entymatic argument where the common agents who make inferences and share information. These knowledge makes it possible to integrate the missing steps, reasoningformscreateplausibleinferencemodelswhich,al- that is the non-explicit assumptions within an inferences thoughdependentoncognitiveandpsychologicalfactors,are sequence.Theplausibilitytheoryallowsustoestabilishifthe notincludedortraceabletothetheoryofCollins-Michalsky. VOLUME4,2016 3 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2020.3003726, IEEE Access Authoretal.:PreparationofPapersforIEEETRANSACTIONSandJOURNALS Thistheoryleavesoutsomefactswhichtheplausiblereason- ingexplicitlyintendedtotreatandmodel. G. OPENSYSTEMSANDNETWORKS Thankstoitsrecentdevelopments,thenetworks’theorycan be viewed as an analysis source and plausible inference investigation. The knowledge representation in graph form is capable of detecting essential aspects about the plausible inference like nature and dynamics. We can consider an hypothesis as ’plausible’ when it is connected to existing knowledge and has a capability of self-adjustment and ro- bustness (resistance to error): the greater the degree of con- nection, the more likely the hypotesis to be plausible. Con- FIGURE1:Howaproblemisdefined nectivitywillmeasureitsintegrationandcompatibilitywith existing knowledge degree and robustness. The success of thisapproachdependsessentiallyonanadequateknowledge dependscouldbehencedescribedeitherbydynamicdefined representation in a reticular form, and in particular on the laws or by some dynamics which are out of the analyst’s chosengranularity.Plausibleinferencescanhaveverycom- control. plexforms,composedofmanystepsresultingfromdifferent rules application (deductions, inductions, analogies), which This indetermination could concern both the single vari- create equally complex graphs. It is possible to define plau- ableandthewayitaffectstheoutput.Inthefirstcasewehave sibilityintermsoflocalorglobalproperties.Inparticular,a adirectindeterminacy(causeindeterminacy)thatconcernsa nodecanbeplausibleifithasnumerousincidentedgesorif problemvariable.Inthesecondcasewehaveaderivativeor therearedifferentpathsleadingtoit. effectindeterminacy. ImagineanindividualthathastomovefromAtoBbybus. III. DECISIONINUNCERTAINTYDOMAINS The first indetermination concerne when the bus will arrive A. SCENARIO (temporalindeterminacy),soontheinputvariable"time".It In the knowledge and collective intelligence era, like ever, is obvious that reaching the B position could depend on the wearecalledtomakedecisions.Generally,weuseproblem trafficwhichrepresentsaneffectindeterminacygivenbythe solvingtoachivegoals(decisionmaking).Itisobviousthat constraint"traffic"orroadfillingcoefficient.Havingknowl- atargetcouldbereachedbyjoining"elementary"decisions. edgeofmultiplepathsandtheirroadstatus("traffic"orroad This highlights an important issue: in the modern decision- fillingcoefficient)itisanobviousadvantagetominimizethe making support systems, it is important to focus both on Breachingtime,alsopayingaroutecost,correspondingtoa the objectives and on the decisionaly path. The objective is longerjourneyintermsofspace. reachedbyjoiningsub-objectivesthroughanoptimalpath. Thisexampleallowstounderstandthesimplicityofdealing If we were in Physics, with a material system that has to with non-deterministic problems when they can be reduced transitfromAtoBinnfinitestepsandthroughmdecision- todeterministiconeswithaseriesofchoices(constraints). making paths, it would be easy to estimate the problem If we think about how to choose which path to take, the complexityandreduceitwithprinciplesofminimization,i.e. answer leads to another dualism: infocompletness and in- the problems of transaction energy, the transaction disorder foincompletness. With the previous example it is simple to ortimeminimization. understand how the knowledge of a free path allows the In the case of a logistic transport problem, in which the driver to easily choose which road to follow to reach point objectiveismovingmerchandiseorpeoplefromAtoB,the B(goalofthetrip)intheshortesttime.Withthismethodwe optimal solution may be achieved by maximizing merchan- cantransformanondeterministicprobleminadeterministic disetransportedinagiventimeinterval. onethankstoachoicebasedonthescenarioknowledge. Whentherearemoresolutionsoratargetsolutionreachable In this way, we have introduced complex systems and throughdifferentcompetitivestrategies,theseexamplesand decision theory characterization: the info completeness- theirgeneralizationsgiveustheopportunitytoformalizethe incompleteness. problemin(fig.1): From the following three characterizations the stochasticity • initialconditionsdefinition, isdefined2: • constraintsfixing, 1) variablesdeterminism,variablesindeterminism; • targetindividuation, 2) functional link between variables and their impact on • solutionstrategydefinition, thetargetdeterminism,targetindeterminism; • solutionoptimization. Afirstproblemclassificationistoestablishwhetheritisde- 2Thatisthenon-determinismofmanyproblemsinmostofthecognitive terministicorstochastic.Thevariablesonwhichtheproblem areaswhereitisnecessarytomakedecisions 4 VOLUME4,2016 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2020.3003726, IEEE Access Authoretal.:PreparationofPapersforIEEETRANSACTIONSandJOURNALS 3) completeness and incompleteness of the starting sce- methodologies to solve and to describe the uncertainty in a nario(inputvarables),thearrivalscenario(outputvari- rigorous framework. An example is the price distribution of ables), and the constraints (functional models which afinancialinstrumentaroundasamplepriceinwhichanot- linkinputstooutputs). normaldistributionisobserved,withheavytails(largerthan We give an example in the financial context where the Gaussiantails). targetisthereachingofapricetarget: This means that some improbable events can still happen, therefore it is important to estimate whether the event is 1) firsttypeindetermination:tickinstant; plausible,credibleorpossibleandestimatereliability. 2) secondtypeindetermination:tickssuccession; Info-incompleteness or the input indecision generate emo- 3) third type indetermination, info incompleteness: ex- tionsandbeliefsthatmakedecision-makingprocesssubjec- changedvolumesandpricevariation. tive. It is important to specify that the word stochastic could So we understand why the rational operator becomes not mean full random, because it is possible to build mod- emotional-affected and takes sentiment and subjective els which manipulate stochasticity and make predictions. In choices. otherwords,takingthepreviousexample,wewillbeableto The purpose of decision-making support systems must be give an estimation based on known constraints or analyst’s to build simulated scenarios, like those of game theory. In needs, so we will be able to make predictions on the price gametheory,differentsolutions/targetscooperateorcompete in a time frame. Stochastic dynamic systems solve many tothebestchoice,wherethebestwordisbasedondifferent problemseveniftheydonotprovideaperfectlydeterministic parametersofdecision-makerinterest(targetconstraints). answer. Thanks to this methodology, related to technological help It is important to define the right position of the problem. andthesupportofsimulatedscenarios,anemotional-affected This problem concerns the certainty of the target. About operatorwillbebroughttorationally(inducedrationaloper- the previous examples, we must ask if the target price is ator)decidebeforetheresultachievement(decisionforward reachable, or if the considered bus line stops at point A etc. intime). Sothequestionisifthetargetcanbepossiblyachievedorit Informationincompletenessandthevariables/information,or isimpossible. functional bonds indetermination, generate emotions in the Therefore, a set of linguistic nuances emerged between the decisionmakerwhichinducebeliefs. certaineventandtheimpossibleone. Thankstothisworkwebringbacktheoperatortoberational In fact, other terms have emerged linguistically: possible, on a larger cognitive basis (expanded knowledge); we use plausible and credible. These terms need to be formally other approaches like sentiment analysis or emotional anal- defined. Shortly, our study highlights that decision-making ysistocontrolthisbeliefs. problemsincomplexitycontextsaregeneratedbydualisms: Thenewrealizationmomentwillleadtothehyperconscious- • determinism,indeterminism; nessstatuswheretheoperatorisconsciousoftheaugmented • completeness,incompleteness; scenariocalledhyperscenario. whichgenerateanotherdualism:certainty,uncertaintyofthe target which have two extremes of solution. This target is B. CONTINUOUSOCCURENCEOFUNLIKELY betweencertainevent,impossibleeventanduncertainevent, POSSIBLE whichisrealityanditisinthemiddle. Howmanytimeshavewewitnessedtheoccurrenceofhighly Uncertaindualismisfacedbyquantificationof: improbableevents? • plausible,implausible; How many times was a football match conclusion plausible • credible,incredible; andthingshavegonedifferently? • probable,improbable; Moreover,howmanytimesthepricedynamicsofafinancial • possible,impossible; instrumenthavemadeustobelievethatwehaveunderstood The goal of our work is to understand how to use, quantify thedominantdirection,butassoonasweenteredthemarket andgeneralizethesedichotomiesinordertobuilddecision- ithadanoppositebehavior? making support systems which are not only probability- These are some examples of how complex it is to make based. previsions under uncertainty conditions (input, models and Someone could be confused and wonder why to use so resultsuncertainty)orinformationincompleteness. many different concepts and not rely on probability theory Overcome determinism and entered the stochastic or un- orBayesianprocesses. certainty phenomena context we find the terms: probable, There is a simple reason, some phenomena which consider plausible,credible,possible. complexity or complex systems are not Gaussian (or other In literature there are different research lines and thousands known distributions) distributed, but they have heavy tails works which describe, distinguish and model these terms whicharenotGaussian. sometimesinaconflictualway. When someone uses the term heavy tails, he is putting Inthisstudy,wepresentagradedorganicvisionoftheterms, a patch on a problem, because he does not have suitable starting from the best modeled concept, rather probability, VOLUME4,2016 5 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2020.3003726, IEEE Access Authoretal.:PreparationofPapersforIEEETRANSACTIONSandJOURNALS using that to model probable event, but that extend it to With this assumption it is obvious that in uncertainty or in- describeotherconceptswherepossible. completenessinformationconditions,theimprobableevents In other words, while György Pólya3 or Arthur Dempster occurrence opens the space to emotions which significantly - Glenn Shafer4 have tried to extract,in probabilistic terms demarcatethepassagefromstrictuncertaintymethods(prob- thebelieffunctionfromthedescribablecomponent.Herewe ability) to the uncertain events treatment under emotional acceptthe differenceamong theterms; wehierarchizethem effectsaction. instrenghttermsandwedescribethemwithamethodthatis In this scenario the closer concept to probability is plausi- moregeneralandnotprobabilityexclusive. bility, even if plausibility and credibility both concern an Inourvisiontheconcepts: estimation theory under emotional coditioning. For what concerns affectivity it tends to provide a subjective answer probable(P ), that aims to being objective. So we assign a probability to r an event based on the several people opinion, in the belief plausible(P ), case this comparison is not required, there is no subjective l evaluationobjectification.Beliefbringsoutthepersonaland credible(C ), subjective evaluation to be shared, surely possible, but not r necessarilysharable. possible(P ), In other words, for the belief it is sufficient that the propo- o sition to which a subjective probability is to be assigned is haveadecreasingstrength,eveniftheyalldescribeuncertain possible. events.Soa: From these considerations it emerges that as the concept of probability is more stringent than plausibility one, in the • possibleeventisnotnecessarilycredible; first we assume that even if there is uncertainty, there is • credibleeventissurelypossible; noemotionallydrivenassessment;insteadintheplausibility • credibleeventisnotnecessarilyplausible,butitissurely definition, we associate an emotional component to uncer- possible; tainty,butitrepresentsthemostpeopleopinion.Inthebelief, • plausible event is not necessarily probable, but it is emotionalitymakebringstheegoout,thatisapersonaland surelycredibleandpossible; subjective vision that does not require comparison with the • probableeventissurelyplausible,credibleandpossible. community,butonlyneedsthepossibility. Fromnowon,itwillbenecessarytoindicateacharacteristic Wesummarizethisintheconceptualprovocation: multiple function for a process or phenomenon that is a "when the improbable is highly possible we should not be combinationofmorefunctions(i.e.distribution): surprisedifitisrealized". Formulated as the same way as a Edward Murphy6 law it F1probability shouldbeascientifictruthtoall. In other words the current price p(t) is a certain form of F2plausibility uniformity, while the trend p(t-1)-p(t) is a more or less reliable,butnotsuresource. F3credibility Using the Dempster-Shafer’s (denoted as DS) theory rather thanbayesianmodelallowstoexceedthemultidimensional- F4possibility itylimittosumone. In other words, price growth possibility with a subjective whereF aremultiscalesingle-modefunctionsto1sum5. i probability estimated, does not imply that the decrescence We conduct an analysis that starts from a different perspec- hasthecomplementto1,soadecreasingtrendprobabilityis tivewhichnaturallyemergesfromartificialintelligencestud- equaltooneminusuptrendprobability. ieswhenitisdesideredtoextenddecision-makingcapability Thanks to the DS theory it is possible to give a probability ofanartificialagentwhichoperatesinuncertainconditions, toanEeventandanEnegation(¬E),thatcouldbeanother with increasing cognitive capacities which try to tend to probabilitydifferentfrom1-P(E)withP(E)+P(¬E)≤1. humanones. InDStheory,C (E)≤P (E)meansthatcredibleislessstrong r l than plausible, because the plausibility is the probabilistic 3Pólya’stheory 4Dempster-Shafer’stheorymanagesthedistinctionbetweenuncertainty maximum value that can be allocated to E proposition, and ignorance and calculates a proposition evidence, in their work it is while belief is the probabilistic minimum assignable value calculatedthebelieffunctionBel(X),thatistheprobabilitythataproposition to E, above all because for plausible we mean something issupportedbyevidence. believableformorepeople. 5WhenanEeventishighlyimprobablei.e.P(E)≤1%,P(E)≤0,1%, P(E) ≤ 0,01% we are moving in the plausibility of improbable events. Then it is not important how the event is improbable, but how much the improbableeventisplausible.Issimilarforcredibilityandpossibility. 6Murphylawisapseudo-scientificparadoxessetwithanironiccharacter realisedbyArthurBlochwithdidacticphrasesinstatistical-mathematical form 6 VOLUME4,2016 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2020.3003726, IEEE Access Authoretal.:PreparationofPapersforIEEETRANSACTIONSandJOURNALS Another difference element between our theory and DS We suppose that on this dynamic system we want to make theoryis:whiletheDStheoryprovidesamethodtochange a position and speed (or momentum) measurement. Make previousopinionsbecauseofnewbeliefs,equallytreatingthe a measurement means to send one or more photons (i.e. a newandpreviousevidence,ourdynamicsystemsexperience beam)inordertoknowineachinstantwheretheelectronis leadsustogivemoreweighttothenewevidencethanold,to andtheitsspeed. refiningthecognitiveprocessandagradualapproachtothe Feynman’s observation is that by making this measure we solution. are changing the state of motion of the electron system; we In financial markets context, this means: if we do a price are perturbing it, because, to provide information about the average for assuming a future price, then an average that electronspeedandtheelectronposition,thephoton,bumping weighs recent prices more than older ones will be more itwillchangeitsspeedanditsmotion. appropriate than an average that equally weighs new prices So, the effect of the measure represents a perturbation. The andolderones. more we are sure of the position, having sent more photons tomeasuretheposition,thelesswecansayaboutthespeed NicholasRescher’snon-probabilistictheoryofplausibility because doing the different measures we have perturbed can be useful when there are equal reliability conflicting (if not distorted) the electron motion and vice versa. The information(newsorindicators). Heisenberg principle says that in the combined measure of Asanexample,thinkthatasetofnewsorindicatorstellsus position and momentum it is impossible to be more precise thatthepriceofafinancialinstrumentisdestinedtoriseina than a certain threshold value given by the reduced Planck timeframeandtofallinanotherone. constant9. Thanks to the Rescher’s theory, which is not sum 1, we can In the macroscopic scale reality, we understand that what associate a value of bullish belief, for example 0.9, and a explained by Quantum Mechanics does not respond to our bearerofthesamevalue,0.9too,sincewehavenotbounded commonsense.Letthemotionofacar,andweconsiderthe it by sum one total like it happens to sum one total theories measurement process link to a beam of photons that collide astheprobabilityone. with the vehicle to know the position and the speed. It is Consequently,theresultisthatthepricewillnotcertainlybe evident that the photons can not disturb the vehicle motion stationary,soafinancialinstrumentwithalateralbehavioris state. tobeexcluded. In this case the observer or measurer is unrelated to the vehiclemotionandanyuncertaintyisgenerated. C. INFERENCEMETHODSFORFINANCIAL ForthisreasonthemacroscopicrealityisdescribedinClas- INSTRUMENTSDOMINATEDBYUNCERTAINTY: sicalMechanics,startingfromtheIsaacNewton’sprinciples INTERACTIONBETWEENFINANCIALOPERATORAND ofdynamics(inertiaprinciple,actionandreactionprinciple). MARKET Manymacroscopicdynamicalsystemshaveasimilarbehav- iortoquantumsystems. When a financial operator joins the market, he performs an Considering the price dynamics of a financial instrument, it analysis (or someone does it for him) in order to decide isclearthatwhatFeynmanexplainedisincompleteanalogy. whethertoenteronafinancialinstrument,andwhatkindof If we replace the electron position with the price level of a positiontotake,intermsofsizeandofupwardordownward financial instrument, we formulate an uncertainty financial directions. markets principle similar to Heisenberg’s principle. In this Thefinancialoperatordoesnotremainanobserver,interacts wayitisimpossibletocertaintypredictthepriceandvolatil- withthefinancialmarketanddisturbsit. ityofafinancialinstrumentinafuturemoment. Inotherwords,theanalysisbecomesaninteractionwiththe This because of the decision taken by the financial operator market,oncetheoperatordecidestoenter. This reasoning recalls that of Richard Feynman7, where he and therefore the entry to the market automatically could imply the market transition from one state to another one. explainsthequantumsystemsmeasurementprocessandthe consequentWernerKarlHeisenberg’suncertaintyprinciple8. A state could be different from the previous one depending ontheoperatorsize,(hisliquidityintroducedtothemarket) About this Feynman explains that the fluctuations or uncer- givingadirectiontoit. tainties that introduce a description in terms of expectation Think of the effects caused by market mover (such as the and probability values to the state of a dynamic system, effects of the decisions carried out by FED, ECB, BOE etc proposes the following example. Suppose that the dynamic ...). systemunderstudyisanelectron,oratleastaparticleofthe What has been described explains the need to place the microscopic world described through Quantum Mechanics. financial markets in the complexity theory and to describe 7Feynman’sreasoningsaysthatonlyaquantumsystem,wherethereis them with increasingly accurate tools which can look at uncertaintyorevenworseambiguity,cansimulatethebehaviorofanother quantumsystem. 9Planck’s constant, also called action quantum and denoted by h, is a 8The Heisenberg principle establishes the limits in the knowledge of physicalconstantthatrepresentstheelementaryactionquantity,determining valuesthatphysicalquantitiesassociatedwithoperatorswhodonotinteract that fundamental physical quantities do not evolve continuously, but are witheachothertakeonaphysicalsystematthesametime. quantized,thatis,theycanonlyassumevaluesmultiplesofthisconstant. VOLUME4,2016 7 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2020.3003726, IEEE Access Authoretal.:PreparationofPapersforIEEETRANSACTIONSandJOURNALS uncertaintyfromdifferentperspectivesandnotonlyfromthe this model, as in the previous one, all the Pi have the same probabilistic,plausibilistic,credibilisticorpossibility. importance.Letusmakeanotherweatherforecastexample: Becauseallthesevisionsinaninstant,indifferentquantities, • P1=militaryairforcerainforecasts; contributetotheformationoftheprice. • P2=sentimentforthewind; We intend to provide a further enrichment level for the • P3=experts’forecastforthetemperatures; MRQF theory [20] as a complementary work. For this aim, • P4=users’forecastsforthehumidity. weworkonthemodelingofuncertainty,andonhowthevar- Thismodelcanbeusedforcalculatetheexpectationfunction iousprobable,plausible,credible,possiblemodelscontribute fortheevent"itwillrain,thewindwillblowat31km/h,there totheconstructionofamarketstatefunctionwhichabsorbs willbe18C°andthehumiditywillbe80%".Itisimportant theindividualprobability,plausibility,credibility,possibility togiveadifferentimportancetotheProbability,Plausibility, functions.Afterthatharmonizesthemtoallowtheanalystto CredibilityandPossibility;so,wedefineweightsforeachof make inferences about future market conditions and to help theseconcepts. him to make decisions about entering the market, in what dimensions,andwhethertobullorbear. 3) Thirdmodel:Weightedaveragemethod ToextendthepreviousmodelsweassumethatnotallP have i D. UNCERTAINTYTREATMENTBYINFORMATION thesameimportanceintheexpectationfunctionconstruction. FUSIONANDEXPECTATIONFUNCTIONMODELING So,webuildthefollowingweightedaveragetypemodel, WtoafienPedd,ePfifrno,emCtha,nePae,pxsppurecochptartihtaiaotetnafuf(un∅nc)ct=itoion0nalaanc:dom2(cid:80)Ap2os→itiao(nA[0i,n)1=t]ero1mb.s- a3 = ((cid:80)(cid:80)4i=4i=11ααiPii), r l r o i=1 i whereαiareweightsrelativetothedifferentPi. Anexamplecouldbetogivea50%ofweighttoP thatis: 1 1) Firstmodel:Averagemodel α = 0,5; 25% to P , so α = 0,25; 15% to P , so α = 1 2 2 3 3 We indicate with P1 = Pr, P2 = Pl, P3 = Cr, P4 = Po. 0,15,and10%toP ,α =0,1; 4 4 Theexpectationfunctionsimplestmodelwecanbuildisthe inthiswaytheexpectationfunctionis followingaveragemodel a1 =a1(P1,P2,P3,P4)= 41(cid:80)4i=1Pi. a3 = (cid:80)(cid:80)4i=4i=11ααiPii =0,5P1+0,25P2+0,15P3+0,1P4. Evidentlythismodelassignsthesameimportancetothedif- 4) Fourthmodel:Weightedproductmodel ferentdistributionsoftheprobability,plausibility,credibility In analogy, to extend the product model for the expectation andpossibility,soitisbothanadvantageandadisadvantage function,weconsiderthefollowingweigthedproductmodel dependingonhowmuchexperttheanalystisandhowmuch wthee pwroanbtabtiolitymathkeeoreyxpiserutsethdefosrycstaelmcu.laTtehimsomreoadletle,rnaastivine a4 = (cid:81)(cid:80)4i=4i=11ααiPii. eventsexpectation.Letusmakeaweatherforecastexample: Also in this way, we will have the possibility to weigh the different contribution of the probability, plausibility, credi- • P1=placeAforecastsofthemilitaryairforce; bilityandpossibilitytotheexpectationfunction.Inboththe • P2=placeBforecastsbythesentiment; weightedmodelswehavetheconstraintof • P3=experts’forecastforplaceC; • P4=users’forecastsforplaceD; (cid:88)4 a =1. Thismodelcanbeusedforcalculatetheexpectationfunction i fortheevent"itwillrainintheplaceAorintheplaceBor i=1 intheplaceCorintheplaceD". Thisisbecausewhenweuseanexpectationfunctionmodel, we are trying to rebuild a probability which had uncertain 2) Secondmodel:Productmodel inputs. Another way of estimating the event expectation is through thePiproduct,formally 5) Overlapmodelwithshiftbasedonprobability a =a (P ,P ,P ,P )=(cid:81)4 P Firstandthirdmodelsrepresentoverlapmodelsandweighted 2 1 1 2 3 4 i=1 i overlap of the different P contributions to the expectation i whereP :2A →[0,1]. function. i Ingeneral,weobservethat A further possible generalization could include a P hierar- i (cid:81)4 P ≤(cid:80)4 P . chicaluse. i=1 i i=1 i Let P be the probability of the event, In detail we can r Tobeabletosaythata ≤ a ,itisrequiredamoredetailed see probability, plausibility, credibility and possibility as 2 1 studythatincludestheP functionalmodels,consideringthat conceptstobeusedindescendingorderofimportance. i the function a has the normalization factor equal to 1. In Forexamplewecanconsider 1 4 8 VOLUME4,2016 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2020.3003726, IEEE Access Authoretal.:PreparationofPapersforIEEETRANSACTIONSandJOURNALS mass (m) function, the belief(bel) function and the Demp- ster’splausibility(wecallitDpl)foreachP: P if1%<P ≤100% P1 if0,1%<rP ≤1% • m(P1),relativebel(P1)andDpl(P1); a5 = 2 r • m(P2),relativebel(P2)andDpl(P2); PP34 iiff0P,r0≤1%0,<01P%r.≤0,1% •• mm((PP34)),,rreellaattiivveebbeell((PP34))aannddDDppll((PP34));; Thiswouldmeanthatuntil1%oftheprobabilitydistribu- And finally compose it with the Dempster’s composition tion we will use only the classical probability(P1); beyond rules. that probability we will define the improbable event, with a plausibility (P2) coefficient between 0% and 100% in the E. RELIABILITY intervalinwhich1%≤P ≤0,1%;similarly,however,0.1% Inclassicalreasoning,themethod,theelementorthesystem r < P ≤ 0.01% we define the improbable and implausible reliability,istheprobabilityofnotbreakingforacertainpe- r event, with credibility (P3) between 0% and 100%; finally, riodt,underdefinedoperatingandenvironmentalconditions; for even lower values of probability we will have unlikely i.e.reliabilitydefinitionisinfrequentisticterms: andimplausibleevents,withacoefficientofpossibility(P4) R= nf, between0%and100%. n At the ends of these evaluations there is the certain event withnf numberoftimesthatthemethod,elementorsystem P1 =100%andtheimpossibleoneP1 =0%: hasworked,andntestsnumberandsamplesusedforthetest. By using our approach, reliability is obtained by changing P :2A →[0,1]. 1 the probability P with the expectation function a. Similarly wecandefineunreliabilityasFault 6) OverlapmodelwithshiftbasedonPihierarchical F=1-R. As an alternative to the previous model, without making to the probability play a pivotal role, but generalizing the Pi Soweareina1-dimensionalone-sumlogicandthereforeR andevaluating,withownimportance,alsotheemotionaland andFaretimefunctions. cognitiveelements,leadstotheformulationofanexpectation Thereliabilityislinkedtothenumberofsuccessesorfailures functionwhichcanbedefinedas obtainedafterattime. FormallythereliabilityRisthefollowingtimefunction: R:T →[0,1]. P ifP −3σ ≤p ≤P +3σ P12 iafnPdPPp12t−>3PσPPP121 +≤3pσttP≤1 PPP12 +3σPP12 Wmqueeannatt,listsyoysdteefimn,eoarvmereaagne toimpeerattoingfaitliumree o(mfttfh)ethmeetfhoolldo,weilneg- a = P ifP −3σ ≤p ≤P +3σ 6 3 P3 P3 t P3 P3 P4 iaafnnPddPpp4t−>>3PPσPP42 ++≤33pσσtP≤2.PP4 +3σP4 wInitohuλrsntuudmybwereocfofnasuidltesrm/edftaFRilu==reRs1/a(Tλt)t,;tiimnge.eneralR=R(T,C, t P3 P3 A)wheretisthetime,Cisthemethodforjudgingsuccesses Where P is the average price on the considered set of data andfailuresandAareenvironmentalconditions. and σ is the standard deviation; the subscript Pi indicates Thisreliabilityconceptcanbeappliedasanestimatortothe withwhichdistributionP andσarecalculated. probability,plausibility,credibilityandpossibilityfunctions, From the provided models, we understand how general the inwhichbacktestingcanbeseenhowthedifferentinferences issueofdecisionsinuncertaintyconditionsisandhoweasy worked. it is to build models using the expectation function which Whenwedonotrefertothemethodbuttotheresult,weneed describe the inference processes in info-incompleteness and tointroducethetrustworthinessconcept. inputuncertaintyterms. The proposed reasoning theory is a meta model where F. TRUSTWORTHINESS defining expectation function models makes it possible to In some contexts, simple reliability and trustworthiness, as constructnewandmoreopportuneoperativemodelscontex- well as probability, plausibility, credibility and possibility, tualisedtoacognitiveandinformativedomain. can be considered synonyms, but this is not the case of domainsandeventswherecomplexitytheoryisrequired. 7) ModelbasedonDempster’scompositionrules Aftersettingthemeasurementconditions,thetrustworthiness Anothersignificantwaytoconstructtheexpectationfunction indicatesthecostancyofasetofresults. istousetherulethatDempsterhasdefinedfortheplausibil- With this definition we understand how in an uncertainty ity, extending it to the application of Pi case calculating the context,intermsofprobabilitiesanddistributionswhichare VOLUME4,2016 9 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2020.3003726, IEEE Access Authoretal.:PreparationofPapersforIEEETRANSACTIONSandJOURNALS Fingerprint Face Iris Retina dependent on emotion (such as plausibility and credibility), Accuracy 90 76 93 94 thereliabilitycanpreciselyexpresshowthesamedynamics aredifferentlyconsidered. Vascul. Ear Voice Signature For example, a trader with the same trend at different times Accuracy 96 74 66 67 mayfeeltheneedtoact/reactdifferently. Hand Palm BodyMot. DNA At this point the result trustworthiness acquires a different, Accuracy 67 88 78 98 butequallyimportantrolefromthepreviousconcepts. Trustworthiness is not reliability, accuracy or validity of a TABLE1:Biometricsaccuracy data synonymous; but we will say that an estimate is trust- Fingerprint Iris Face worthyiftheresultsremainconstantovertime. Accuracy 90 93 76 Ifthisdoesnothappen,wewillsaythattheestimationorthe resultisuntrustworthiness. TABLE2:Selectedbiometrics There are several methods to estimate trustworthiness; we will use the Pearson10 correlation index which is the basis ofthevariousreliabilityestimates. • probability:thebiometricsstatisticalreliabilitythatisin ThePearsoncorrelationindexbetweentwovariablesxandy the[0,1]interval, is • plausibility:thebiometricssentimentthatisinthe[0,1] r = σxy with -1≤r<1 interval, σxσy • credibility:theawarenessandreliabilityoftherecogni- whereσ xydenotethecovarianceandσ xσ ythestandard tionalgorithmthatisintheinterval[0,1], deviationsofxandyrespectively. • possibility:thesystemfaultstatistic. Wedistinguishthefollowingcases: InSportodds,allthefeaturesareinthe[0,1]interval: r <0:relatedvariables • probability:thebookmakerquotesaverage, r =0:nonrelatedvariables • plausibility:theexpertsopinions, r >0:anticorrelatedvariables • credibility:thesentimentanalysis, 0<r <0.3:weaklycorrelatedvariables • possibility:theusersforecasts. 0.3<r <0.7:moderatelycorrelatedvariables Thesimulationhasbeendevelopedwithrstudio;wecreated r >0.7:highlycorrelatedvariables thetwosimulateddatasets11 andtheyhavebeendisturbedin andsimilarlyforanti-correlationcases. ordertosimulateuncertainty. Generallyinthecaseofmorethantwovariables,thePearson indexwillbecome A. BIOMETRICSSIMULATION r =r . ij Whenwefusebiometricsinamultibiometricsystem,weare We have a correlation matrix of r rows and r symmetric dealing with a decision-making problem. The decisions can column r = r with coefficients on the diagonal equal to be joined by several methods: voting method, where, each ij ji 1,asr =σ /σ2. classifier votes for a class, then the pattern will be assigned ii ii i Forexample,inthefinancialcasewecanestimatetherelia- to the most voted class, we focused on this method; Gupta, bilityoftwoormoreperiodsofresults. Singh Walia and Sharma [22] propose a score based fusion method;Divyakantetal.[21]describeotherfusionmethods. Weinitializeeachbiometricwiththerelatedaccuracy: IV. SIMULATION In this section we will describe a simulation based on the Delacetal.[23]andTripathi[24]calculatedthebiometrics describedtheory.Thissimulationhasbeendevelopedintwo reliability;wehaveextractedthebiometricsaccuraciesfrom fields: these works (table 1). Suppose to build a multibiometrics systemthatusethemostplausiblebiometricsintermsofac- • Biometricfusionchoiches, curacyanduser-acceptability(useracceptabilityisdescribed • Sportodds. byTripathi[24]andRui[25]).Weselect: For each field, we suppose to have different features to be appliedatprobability,plausibility,credibilityandpossibility. • Fingerprint, Ouraimistocalculatehowmuchthesefeaturesweightinthe • Iris, formula: • Face. Inthetable2,wecanseetheselectedbiometricsaccuracy. a3 = (cid:80)(cid:80)4i=4i=11ααiPii. After that, we initialized three algorithms for each bio- Inparticularinthebiometricsfieldwehave: metricsandrelativesuccessrate:randomhigh,mediumand 10ThePearsoncorrelationindexbetweentwostatisticalvariablesisde- lowerrate.Thesuccessratecanbefoundintable3. finedastheircovariancedividedbythestandarddeviationsproductofthe twovariablesandexpressesapossiblelinearityrelationshipbetweenthem. 11onefield,onedataset 10 VOLUME4,2016 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.