head 1.1; access; symbols; locks http:1.1; strict; comment @# @; expand @b@; 1.1 date 2004.06.03.14.37.40; author MarkusDolensky; state Exp; branches; next ; desc @none @ 1.1 log @Reqs for SED classifier by Adorf @ text @ࡱ> {}zM 6bjbj== 2HWW2sl 8, @@ ,-jx : :   Y-[-[-[-[-[-[- / @@1[-[-C p-CCCh  Y-CY-CCy*lY- l MjDB g+*Y--0-,J1C1Y-CAn SED Classifier for Planetary Nebulae a GAVO-AVO project Requirements Analysis and Project Plan HMA (GAVO), V1.1, 25-05-2004 Summary: The AVO team at ESO has suggested to the GAVO-team to consider a VO-collaboration resulting in an SED classifier service. This service would be initially used to solve the problem of classifying Planetary Nebulae by comparing observed SEDs with SEDs derived from a set of theoretical spectra. The classifier resulting from the proposed collaboration may possibly be incorporated into the next AVO-demo at the beginning of the year 2005. This document outlines the requirements, sketches a project plan and suggests a subdivision of the work. Background A member of the AVO Science Working Group (SWG) has proposed to the AVO team to consider implementing a classifier service that allows the classification of all sources in a published list of Planetary Nebulae. Classification is supposed to proceed on the basis of observed quasi-SEDs that have to be assembled via fuzzy spatial matching from published catalogues. The main goal is to classify all PNe with the help of a library of theoretical spectra (described below). A secondary goal consists in finding PNe that show an infrared excess, since these are potentially the most interesting ones. This document describes the requirements for the classifier. The requirements for a data download & matcher service, the result of which could be fed into the classifier, are described in a separate document. The corresponding projects required for implementing the download & matcher and the classifier services are obviously related. However, they are only loosely coupled. The only requirement is that the output of the download & matcher service can be fed into the classifier. Input Observed SEDs The input list of 1312 planetary nebulae can be retrieved from  HYPERLINK "http://vizier.u-strasbg.fr/viz-bin/VizieR-4" Vizier at CDS. Using this list, quasi-SEDs must be constructed by consulting various optical and infrared catalogues. For the remainder we assume that from a manual or automated process candidate quasi-SEDs are available, where each quasi-SED consists of photometric measurements in the different passbands, not necessarily in the same flux-units. Theoretical spectra A library of theoretical stellar spectra exists which is supposed to provide templates for the classification process. This library consists of hundreds of high-resolution spectra. Only the central star of the PN is modeled, but not the surroundings. For the purpose of the planned classification of PNe this library has to be preprocessed, i.e. the data has to be mapped from theoretical space to data space. To this end the theoretical spectra have to be down-sampled, ideally taking into account the spectral response curves for the filters used in the observations recorded in the catalogues. The number of values in each preprocessed theoretical quasi-SED has to be the same as the number of values in the observed quasi-SEDs. Weights Observed PNe may have an infra-red excess, which is not modeled by theory. Therefore it is advisable that the classification process (optionally) excludes the infrared pass band. A mechanism is needed that allows, at the users discretion, the exclusion of certain photometric data points from the classification process. This functionality can easily be accomplished with a vector of weight values, which may take the Boolean values true or false (1 or 0). The number of weights must equal the number of data points in a quasi-SED. Somewhat more generally, we may allow the weight values to assume any non-negative real value at the users discretion, thereby implementing a more general classifier, which can work with arbitrary weights. The weights might e.g. reflect the statistical uncertainty of the measurements. The default weight vector has weight = 1 for all weights. Output Deterministic classifier Initially a simple deterministic classifier will be implemented. This deterministic classifier will only use the observed and preprocessed theoretical quasi-SEDs, and take into account the weight vector. It will employ a nearest neighbour-algorithm to find, for a given observed quasi-SED, the closest theoretical quasi-SED. To this end a dissimilarity measure (e.g. the Euclidean distance) between quasi-SEDs needs to be defined. The fact that the quasi-SED contains logarithmic quantities (magnitudes) should be taken into account when devising the dissimilarity metric. The result of the classification process will be a table, which will show the IDs of the observed SEDs labeling the rows, and three pairs of columns giving the best, second best and third best class assignments along with the corresponding values of the dissimilarity metric. This would allow a coarse judgment on how unambiguous a classification really is. obsID theoID_1 d_1 theoID_2 d_2 theoID_3 d_3 The table should be formatted as an HTML table for viewing and as a CSV-file for later use. Optionally the classifier will output the complete classification matrix (table), which for each pair of observed quasi-SED and theoretical quasi-SED states the value of the dissimilarity metric. obsID d_1 d_2 d_2 Preprocessing of the observed data The computation of a dissimilarity metric requires some preprocessing of the observed quasi-SEDs, since the PNe are observed at different distances, and therefore the flux recorded varies. One possibility consists in normalizing the flux Either by dividing by the mean (if the data space is linear), Or by dividing by the mean square. Alternatively, flux ratios of magnitude differences could be used in order to remove the effect of variable distances. Statistical classifiers A more advanced implementation of the SED-classifier would automatically take into account the measurement uncertainties of the photometric quantities that make up the quasi-SEDs. Assuming a multivariate Gaussian distribution of the photometric measurements, a family of likelihood-based classifiers can be implemented, which are based on the Mahalanobis distance. Each classifier needs to be augmented by a classification strategy, such as maximum-likelihood (ML), maximum-a-posteriori (MAP), Bayes, or Neymann-Pearson. The classifier described here is quite generic, and could also be used for classifying astronomical objects other than Planetary Nebulae. Dataflow The classifier shall be implemented as a client-server application, and be accessible via the Internet (see  REF _Ref73249150 \h Figure 1). The purpose of the classifier is to compare the observed SEDs with the templates from the theoretical SED library, and to produce a classification.  Figure  SEQ Figure \* ARABIC 1: The data flow for the SED-classifier. The user indicates the type of data (observed, theoretical, weights) to be uploaded, along with the desired output. The classifier computes the best class assignments and outputs them along with the values of the dissimilarity metric. Optionally the complete dissimilarity matrix can be output. File-based In its initial implementation the classifier service is supposed to offer the following functionality: Upload of a list of observed SEDs into the classifier. Upload of a list of theoretical SEDs derived from theoretical high-resolution spectra. (This functionality will require suitable authentication and authorization, if the uploaded SEDs are to be permanently stored on the server.) Upload of a vector of weights. (This is an optional step, only required if the user wants to deviate from weights equaling 1. Classification by a deterministic nearest neighbor classifier based on a suitable dissimilarity measure between the observed and the theoretical SED. The input data format of the classifier service will initially be restricted to blank- or comma-separated (CSV) ASCII files. The number of spectral channels and their central wavelengths and passbands need to be the same for the observed and the theoretical SEDs. The initial implementation of the classifier is to be operated from a standard Web-browser. A prototype of the deterministic SED classifier described so far has been implemented by GAVO and is available for demo purposes, further requirements analysis, and actual SED classification. VOTable-based In a next step the classifier might accept VOTables as input, and produce a VOTable, probably with several sub-tables as output. Web-service based An interesting further step would consist in encapsulating the classifier described above in a Web-service. It is an open question which form the input and output data should take. VOTables do not appear appealing at this point. Responsibilities It would be the responsibility of the AVO team to prepare the list of PNs with associated SEDs, and to prepare the library of theoretical SEDs. This work would be carried out in cooperation with the proposing SWG-member. The resulting library would have to be kept non-public by GAVO until further notice. It would be the responsibility of the GAVO team to provide the AVO team with the proposed classifier service. Initially the service would be offered via the GAVO Web on a restricted basis, requiring authorization/authentication. Alternatively, the classifier software could be delivered by GAVO to AVO in form of a war file that can be locally executed by the AVO-team using a servlet container such as Apache Tomcat. Next steps Test data For testing the AVO-team needs to provide theoretical and observed data to the GAVO-team. Development and testing The GAVO team needs to make the classifier prototype and subsequent versions available to the AVO-team, and it needs to be clarified, how. For the time being the AVO-team could work with a war file that is occasionally replaced by a new version. Data storage It needs to be decided how and where the classifier may store the library of theoretical templates. Authorization and authentication If the functionality for user-directed template uploading is included, then an authorization and authentication mechanism needs to be defined. Alternatively, each user may upload his/her own templates that would be available as long as the session lasts. Later Phases The scope of later phases is not yet defined. Below a few options are discussed in approximately the order in which these could be implemented. Down-sampling It might be interesting to create a separate down-sampling service, so that users with theoretical spectra could map their data into observational space. Maximum likelihood classification Other generalizations considered so far would consist in using the photometric uncertainties of the observed SEDs. (VOTables might provide a handy way of storing the photometric uncertainties along with the SEDs.) If each SED is augmented by its photometric uncertainties, a more sophisticated statistical classifier could be used. Instead of computing the Euclidean distance between the observed and theoretical SEDs, the Mahalanobis distance would be used, which uses the variances (and covariances if provided) of the SED data elements for statistical weighing. The Mahalanobis distance is related to the chi-squared metric and the resulting statistical classification is similar, if not identical, to a maximum likelihood classification. Maximum-a-posteriori classification If photometric uncertainties are provided, further generalizations of the classifier are conceivable. For instance, the a priori probabilities of PN types I and II might be provided as input. Then the maximum likelihood classifier could be generalized to a maximum-a-posteriori classifier, which produces the correct classification more often than the simpler maximum likelihood classifier. Further generalizations are conceivable that would drive the classifier away from objective classification into a goal-directed subjective Bayesian operational mode, which might become very useful during proposal preparation for producing interesting candidates for further observation. Operating characteristic Another useful option would be to complement the statistical classifier with an operating characteristic (OC) diagram indicating the hit rate and the false alarm rate of the classification. Such a diagram could also be used for controlling the free parameters that determine operating point of the classifier. Web service It is also under consideration to augment the simple interactive HTML-client by a SOAP-based Web-service, so that the classifier could be invoked programmatically from remote locations. References H.-M. Adorf, 2004, Data download and matching for Planetary Nebulae a potential GAVO-AVO project, Requirements Analysis and Project Plan. GAVO-AVO collaboration proposal  FILENAME SED classifier requirements.doc Page  PAGE 6 of  NUMPAGES 6 e  F G Mmv`i  $$Y%k%P&a&:)O)))**5\ jUjEUmH sH  mHnHujUmH sH OJQJmH sH  0JmH sH jUmH sH jUmH sH 5\mH sH CJaJmH sH mH sH =)>e#$  K L  LM66MTmEvw< & F`  p#``i s ! """## $ $$$Y%k%P&a&& & F$$a$&&''':)E)O)))**(+I+H,U,,,--00`3y344u55 & F & F **(+I+G,U,,-V/j/00`3y344u55556.6/606:6;6Z6[6b6c6i6j6k6l6p6q6{6|6}6~6666mHnHsHujUmHsHmHsH6]mH sH mH sH mHsH*5 6 66.6/66666 01hP. A!"#$n% DyK yK Xhttp://vizier.u-strasbg.fr/viz-bin/VizieR-4{DyK  _Ref73249150xUDde"lm  C bAJfigures\deterministic classifier.jpg"RT ?Qbp;}lTFdT ?Qbp;}JFIF``C    $.' ",#(7),01444'9=82<.342C  2!!22222222222222222222222222222222222222222222222222K" }!1AQa"q2#BR$3br %&'()*456789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz w!1AQaq"2B #3Rbr $4%&'()*56789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz ?;SPL4vwC21H 8$!< ۷=^i2}p[|%ܳ\Kdº 4UxwO* (#ÿTi?(?#IQG⫠9B<;@@'G!F4UxwO* (':%̪GІȤm6ASq\ ̚}faʑ`ފĩ5@@YAYAyk tFS?Oūگ}RctQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQED/ƺ "^\c]W $«s O-q8>cdm)x& ty{ }כzbj+\ʪLȤuXjآOژCQ.-FiNktҢ2T}>oׂm'C=E?4 Z9='uP+{ҭ-ǒIU =I9$ R}RCv%Z0U;yz(m-Qkw l"qT aCg=h_$mIԣYQ.YF @@Hpx5@@4>=+L rp<*P5[yuddwd#+} bYϤU(,(A~X+} bYϤU(,(A~X+} bYϤU(,(A~X+} bYϤU(,(A~X+} bYϤU(,(A~X+} bYϤU(,(A~X+} bYϤU(,(A~X+} bYϤU(,+QZϊ4TQFBsşHh(((((((((((((_7q-ZȽI`b+[uM"4ݶ d !?L-S͏J@@eϮ+3QZɝP9pRHVpH#>ƀ:J+K]sWfL?E{hL?E{j8|sܡx"e@@셓D`X"Gb2\-/}u^(n趶\\E j]DUUI&,ReϮ+3@@ev ?G󊑮M}M֙dKwrLEryGe=/U֧;Dyb+m%PGj0@@(((((((((((((((sşHkgc+Z((((((((((((((+JѳWA\?P?: (  ; _[5t7@@w?Կj(u"??k_  (((((((((((((((((+((((((((((((((~(w F]sCT6j(+o+lW?WR٨(> |ȃ(((((((((((((((W=4[IlEv(pI2~?2o5Gj렠77ώ<ώ(7}Ak:[ox!~s}XqEsw?Cɿw?CɿtP?x|t? Qx|t? WAEsw?Cɿw?CɿtP-5GaM {UQ@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@sCT6j+JѳPAEP\sa]KKfº@@Q@@A tD>A@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@j렮5o uPE=Mzx :DQ0`fS擻7Xx-Am^bkh.sדqA\A tQEQEQEQEQEQEQEQEQEQEQEQEQEQEOFa[ 'V] [xN Kr++0wfzlu[Pqkm/Ԗ'ٞ`7@@Au͆uxfI)Eܗ$ѰY;+d;FD]֟;^[i{st#LK݊7'ОCiv7d6fK=E?)O~@@.m緗2%k ,kK> dbI~U99=_2J,sqoqbd1`xb9P+ddg+hxi\^6^dH{Y ÑZ<-tz^hV;VKKee`bqN;((((((((((((+JѳWA\?P?: (  ; _[5t7@@w?Կj(u"??k_  (((((((((((((('V]rv#^ѦaC[9*0FAs__+@@k( ?PAEs"_' ?EBOQ\WЯtW?/~?__+@@k*Agh|0q?utW?/~?__+@@GLN¯n*QEQEQEQEQEQEQEQEQEQE%P٫~(w F@@Q@@s u/-  ; _[5tQEs:@@5W?|A`]QEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEOFa[ 'V@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@sCT6j+JѳPAEP\sa]KKfº@@Q@@A tD>A@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@sQVC տ+uAEPEPEPEPEPEPEPEPEPEPEPEP\?P? y`?lQEW?WR٫PAEP?|A`]s:@@5PEPEPEPEP\5}Wz6mG6>wF]IkFĿDz mmKo7]a6%[-tP?}n=ouQ@@Cm[?oAEsz mmKo7]a6%X8{auU.}HPMEQEQEQEQEQEQEQEQEOFa[ 'V@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@sCT6j+lo>#Cd)lK)# @@#7? MwO&:  ; _[5C?-.6zPGxd ,N22p23Vi}Zx4MF3<6̭bK[([r=7? MwO&ȃ+|c>m~D$WRv`"@@qaI-^7#:3i"#kEs|XA=|7? MwO&: +ÿ24o?(C !dhM;A'Qo#,H98v:JoK]mJכx6Ů:t,'Ώz0kiZm[oG:]{umgNͨ_f|H,W .{njjjOiy xbXVȠf)7iY[j 42wrdYp3p'5? hi{(l 2M4kz5xZ&1I5C1Tpp9N95^:U/n ѬKo&&0Cs``q@@K SEle95wy[YǷǥ ܐG dp0ɻӤR74߽baX'/4w3xtER8*{s=+@@Z<)>T'A IQ *#UVJk,-omH"mU:c8QEa H&,lQsE +ÿ24o?(C !dhM;A'G&A7? MwO&: +ÿ24o?(C !dhM;A'Zzfa$}ryn@@ `y ЊEP\j렮5o tQEQEQEQEQEQEQEQEQEQEQEQEQEQEW?WR٫PAEP?|A`]s:@@5PEPEPEPL5ԩǡ@@K_Mmȸyg3dgFĿDt5x蕠 =cRFou9;HfUA8Zj$7ZUmb/Ⱥ, z]eݬsZ]F%92?*Ga}B]^Xib ҁ wg\Fmơ mok(BD;y$fǎur%Z4bSF+~ѐ\Z|1PS}Kȭ 3FleO YFӭɺA2 Bl;}ߗn6Pt^.ia[K$rF3M_i_fF-&ȟ~+3'k3i\B%iϙ$̭#7%җUw.#Cri<;b8N[|3Ouw V"9>eH3ޮK,znmk wG ̆7$Y+zOqKyrI4s9K-DŽ"Q0-R^7P@@ h*K w6{dZM1eWjm`ヸN=+[!:7Lf iʬr0G8ȧC-2;{䑮%ʞ{iJ Cg'qV .{ {쩝sӢ(((((((gc+Z+ | ( ( ( ( oK]mJA\}k[-F)PzPCEs)gVtW?V)gAEs)gVtW?V)gAEs)gVtW?VmZƗS$Ӂ#@@QEQEQEQEQEQEW?x±t{C,Ek@@Q@@sQVC տ+uAEPEPEPEPEPEPEPEPEPEPEPEPEPEP\sa]KKfº@@Q@@A tD>A@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@Q@@s?+ZA\9 VQEW??n:+[·_tQEQEQEQEQEQEQEQEQEQEQEQEQEQE7@@w?Կj+o+lQED>A\A tQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEOVSH/5)!hai8.ʣ'h$p=k˼gD׼eiV XmbC8J~^ɃR}<1zmNDϢƫOA@@!F?#IWAEsxwO*B<;@@']i?AH@@$Nh?#IQG⫠9B<;@@'H<馁& k #wQsCy6w<2ehϡSMtP=/QUKU%qԕdopێ*s7W;u6{{z;GtQEQEW'ivώm41i"G2c+p~UW>PT5#ÿTi?(?#IQG⫠?2h8##ÿTi?ܹpds֖9B<;@@'G!F4UxwO* (#ÿUrvЬ0ۺFLwJ((((WA\wvVOmV>tkMI"uo9_Es&k(T3\_@@A t1-yei}#G{vK]ExXnqh73Y6?+[Oe `9hSsYE&k(HuO5eCЙ/H!?L?$Q Bf": +T3\_Q]xUKI|#؆3Yq_F[ kg(%A_}k<+u=\Xc3nZfFt>J(H.Wy_k&8L@@֨wxy5Tfr&5Jq(δiu4[ߕoSc$b[| qWBHnRRѠڤ(a̐G?6pkj/@@`ɜ3SEҢRM[nVqqXMHNŬ nD2H n ۋӊ gföw{y>fgHH HMr7#MHkl0A {u?EO N@@.#O-00(ϮuRnPJԙC,2'IPkKo t4<≉oWiP65i;l.llZ(n%J*`v6q Yi4$SИޔQ:dͧ_3Z#VR`;p1灑ZFaд1to3hHq|sޯ[AiAm pƒ jW%ß]b42#Yc+zӃEs&k(T3\_@@CЙ/HSsYEtWJ񖿤C"=fE!frr8r-jd¯(;Z e}؜~QVCXjRܳE!RK%IĊ(׼?YG{,3zV|'Zxoo%&v@@@@TY ݁9ִ{ ,l])!ToNC RoKkr.oo./&w#PG?.|9cx[(b=R8RT)!) Hj5d<)Mٓ;osס9rAnt ]}+ŹUE26MBI: GmnR]5شU@@<%훉mgYxMSh$89j fYVAˮntakM>Dytwrwq: ummM1y#iyA@@6~:+<:e[ipWK}*!.?0I"so~sPdx,ʯl@@j嗂lt.;6-9ي",0 yqiOu$R3._/,)R0GP{`Pe׊4OX݃DTdAdpEOc`s-p设j(ww69Cn^DcWQcQpAmۅU14\V,6ZOm Ȅi%ڭF+4QE:xZ/]qR񾺶Ԗx'eCf!?L?$PAEs&k(T3\_@@CЙ/HSsYEtϧ9zM7_a]4_KQ-V,\F <u5ԯ5osLTytfZΊ(ggDcCYTA=z¨zFG}Fr%pG@@ ^Z=M> X+#:!]8܁S V_SLplw0_)?;ܶQD6QQHDaP1#;{薑Ȗ=K* lzxrG\֞{.Qot]2cn`G /6%iY Eub +p ScLO{,E[ogޮ[X%V&vUp945Q@@Q@@Q@@Q@@Q@@Q@@Q@@A tD>A@@Q@@Q@@Q@@#*` Z(K;[ eB~qX7^%%kFĿD^6OReI U$gָoZ8D\D il(#9^@@e*F=ºM-ͭlm嶎X$ϧҀ2ijVp4e!H0yK*P UqO\:&oځ;ZFa\f'$q]|~Т=OKy-pH/#>-!!lXc#@@gP˟P;}A&F+m 6*Yhi!a 3 Ed,'Qa+l-CīX. \ fi"_i]YiqXMmo-?T¢5Z`w4"=`1\c Ͻrw]Z? n\'W` :oxWS,o] 1,EPEPEPk:UkMBf* <@@540o CIH0T{O9 տ+uA\j렠(YSsycwue,v$վx ̝9 fǍO-M|S%p-!0N8sj=jnHb4r,{}p1F_o=Z0\YGg>Ռ"G1RT9Ro͢gE!3G1ouYax]πp7tbYU.ᲆ[ hѝ6g,0pF=.WD352}3!2P # lc2Ŝ%x66Z}oE;Gh(ʼn 7A746K;i04H'"+?Daдk]2+tڭ&7IZР(KAz+tr^ ((]m-`9wFsI8Oc'9z+O(s*?WxQMlKg{u%<tGOsL }Yx{Ed,3L\98F>Ҁ9D{mbV oFF9zP-6k 64߰y)yddʍ0GJE~t/7\u//ܬEʔG?v ׈WSEKqP9( g "|*>Xö%Yl+:3wB 8 3FNNMp-{kkqͥi(f(+H )a]č3+Oyowi8I#40T0 b;9$22;9]*OYO+~GR:],േ;A@@OLws@@h(((((u"??k_  ((((+ѿj/u+]b^xZQ\6-+ gP#kAMsT#kAMsTQ\"6?k?G"6?k?@@#kAMsT#kAMsTQ\"6?k?G"6?k?@@#kAMsT#kAMsTQ\"6?k?[VikI,i29QEQEQEsdI%U7,Һ 럶&49IGb2A$U_J4}jQSO8Q\&Z_"4eϮ+3@@eϮ+3G&Z_"4Q\&Z_"4#Hmp*p8 2\-/}u^(|Km:^d>8Ѯ`x"eUwdDX \&Z_"4ljZt6"1؅"NbAlՒ=͏ں iiv1[ؙ%坉%v$OTQEQEW>PT5렬KÖz^5+HK&sO@@W?5Q5PAEsS\US\UtW?5Q5PAEsS\US\UtW?5Q5PAEsS\U\BV1y8_Vc%R(((((_  u"??k((((((((((((((((("C#1<:߃<}w'-$7;vv+1$#G"`ʸox/zg-" Eb$k8 h(9_ȋ[襭_ȋ[襠 ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (9ȃ+> | ( ( ( ( (C>nw1\z15ס*NC^(ס*NC^(ס*NC^(ס*NC^(ס*NC^(ס*.cQ9ap (((((+ | ( ( ( ( oK]mJA\#W뭿Z((((((((((((+PT5,>;x\cKcʨr qT>үtm*?{[)<}p}񞁨OmXS+:$@@Tp^ L XE+DRe8#qSEAmw٘BŌ2ʕ zG$$0 { }JWfY"9]J2Y(y(IXIr-ʉgUAGv ( ( ( ( ( (9ȃ+;_Z!]K)T<s(h7h7s<s(h7h7sچ|m˞{@@5s7^%%k `kM쭗+ټ*/ƷM̷M2ǟ%~bN>5|GZiwZv4HTaR1ҭK;K|/gnsӭpzS^@@ضrWBs}wO5k]Z}Zi8WaLW/7+=6 1eX [,p8*geDgcQ}ˬBoKAj4`n>hWsB6}si oH$ƻ-jI!b c!JwHӢ"M Sn Ay-" He@@e# ¼Kfmni!1) c[-|,Ӧ 'Y'P 0>Ϫ\j6݋PT5,TgxngWxm*[2#0qk,{I^XW 2Ȼb# ;8#`d U6m}d̅OtS* <njL8P$WڟⴼP14 bJ*~Zaki>nSQ夞W%#Ys(pUHXqR%$7ko][,"taS[}RY{+. z -d;BjZiddwFݏVfbKڀ.EQEQEQEQEQEQEQEQEQEQEW?_oV oK]mJAEPX~o#,aGㄶwBW9=r((((((((((((((((((}?`kA\CQנ(4.-'}O̺ ݼ]JBe^Ey[V7FqjI%ƒ.i ͅntڈofW$"$cvO@@a}o?௒(g?pG+(T2A 5瀿DѿZ袊((((((((((((+iYR0t^:_B=}~H>}o?௒(g?pG+(Q? "}~H>}HYe=_"WOE? i8@@8 NormalCJ_HaJmHsHtHZ@@Z Heading 1$<@@&&5CJKH OJQJ\^JaJmH sH T@@T Heading 2$<@@& 56OJQJ\]^JmH sH N@@N Heading 3$<@@&5CJOJQJ\^JaJ<A@@< Default Paragraph FontN>@@N Title$<@@&a$5CJ KHOJQJ\^JaJ >J@@> Subtitle$<@@&a$ OJQJ^J,@@, Header  p#, @@", Footer  p#.U@@1. Hyperlink >*B*ph8"@@8 Caption xx5CJ\aJ2H )>e#$ KL L M MTmEvw<`i s  Y!k!P"a"""###:%E%O%%%&&('I'H(U((()),,`/y/00u11 2 22.2/222000)0)0e0e000000000000 0000 000000 00 00 0 00Y0`0`0`0`0`0`0`0`0`0`0`0`0`(00 0` 000Y000000~0~000~0 0 0 0 00000000~00~0w00n 0n 0n 0n 0n  0n 00000c#0X#0q$0X#0$00&0&0&0&0G'0&0P*0&0-0&0i.00//0//0@@0@@0@@0 0!!rrru*6"&56 !#6F2X !,LT[]bmou!8@@0(  B S  ?2 _Ref7324915022w{EHNR" % R[ HLX[z~26]acl ! !$!+! "("""""""()* *#*+*~**L+P+112.2]2b2m2p222+, !!""""##()U-]-112-2[2b2l2p22233333333333Hans-Martin Adorf8C:\gavo\doc\projects\avo\SED classifier requirements.docHans-Martin AdorftC:\Documents and Settings\adorf\Application Data\Microsoft\Word\AutoRecovery save of SED classifier requirements.asd|\}B Z~ P kxfp.jVPWN,hJjYJlA3|]\2b.a:zq l}aTs%I{l^`.^`.^`.^`.^`OJQJ^Jo(^`OJQJ^Jo(^`OJQJ^Jo(^`OJQJ^Jo(hh^h`.hh^h`OJQJ^Jo(h^`OJQJ^Jo(h^`OJQJ^Jo(ohpp^p`OJQJ^Jo(h@@ @@ ^@@ `OJQJ^Jo(h^`OJQJ^Jo(oh^`OJQJ^Jo(h^`OJQJ^Jo(h^`OJQJ^Jo(ohPP^P`OJQJ^Jo(h^`.h^`.hpLp^p`L.h@@ @@ ^@@ `.h^`.hL^`L.h^`.h^`.hPLP^P`L.h^`.h^`.hpLp^p`L.h@@ @@ ^@@ `.h^`.hL^`L.h^`.h^`.hPLP^P`L.h^`.h^`.hpLp^p`L.h@@ @@ ^@@ `.h^`.hL^`L.h^`.h^`.hPLP^P`L.h^`OJQJ^Jo(h^`OJQJ^Jo(ohpp^p`OJQJ^Jo(h@@ @@ ^@@ `OJQJ^Jo(h^`OJQJ^Jo(oh^`OJQJ^Jo(h^`OJQJ^Jo(h^`OJQJ^Jo(ohPP^P`OJQJ^Jo(h  ^ `OJQJ^Jo(h^`OJQJ^Jo(oh^`OJQJ^Jo(h| | ^| `OJQJ^Jo(hLL^L`OJQJ^Jo(oh^`OJQJ^Jo(h^`OJQJ^Jo(h^`OJQJ^Jo(oh^`OJQJ^Jo(h ^`OJQJo(h ^`OJQJo(oh pp^p`OJQJo(h @@ @@ ^@@ `OJQJo(h ^`OJQJo(oh ^`OJQJo(h ^`OJQJo(h ^`OJQJo(oh PP^P`OJQJo(h^`OJQJ^Jo(h^`.hpLp^p`L.h@@ @@ ^@@ `.h^`.hL^`L.h^`.h^`.hPLP^P`L.zq~}|VjYJ%I{bNA3|]aTs                           @@2P@@UnknownGz Times New Roman5Symbol3& z Arial71Courier?5 z Courier New;Wingdings"A˅F&˅F&˅F=F)X!2022QGAVO-AVO collaboration proposalHans-Martin AdorfHans-Martin AdorfOh+'0 8D ` l x  GAVO-AVO collaboration proposalAVOHans-Martin Adorftiansans Normal.dot Hans-Martin Adorfti15sMicrosoft Word 9.0i@@캃@@4YDB@@@@B@@4YDB=F)՜.+,D՜.+,X hp  INTERVAL LOGICX2  GAVO-AVO collaboration proposal TitleX 8@@ _PID_HLINKSA ".,http://vizier.u-strasbg.fr/viz-bin/VizieR-4Y|%figures\deterministic classifier.jpg  !"#$&'()*+,-./0123456789:;<=>?@@ABCDEFGHIJKLMNOPRSTUVWXYZ[\]^_`abcdefghiklmnopqstuvwxy|Root Entry FMjDB~Data %V1TableQ1WordDocument2HSummaryInformation(jDocumentSummaryInformation8rCompObjjObjectPoolMjDBMjDB  FMicrosoft Word Document MSWordDocWord.Document.89q@