Gradient Estimation Schemes for Noisy Functions Brekelmans, Ruud; Driessen, L.; Hamers, Herbert; den Hertog, Dick

Size: px
Start display at page:

Download "Gradient Estimation Schemes for Noisy Functions Brekelmans, Ruud; Driessen, L.; Hamers, Herbert; den Hertog, Dick"

Transcription

1 Tlburg Uverty Graet Etmato Scheme for Noy Fucto Brekelma, Ruu; Dree, L.; Hamer, Herbert; e Hertog, Dck Publcato ate: 003 Lk to publcato Ctato for publhe vero (APA: Brekelma, R. C. M., Dree, L., Hamer, H. J. M., & e Hertog, D. (003. Graet Etmato Scheme for Noy Fucto. (CetER Dcuo Paper; Vol Tlburg: Operato reearch. Geeral rght Copyrght a moral rght for the publcato mae acceble the publc portal are retae by the author a/or other copyrght ower a t a coto of acceg publcato that uer recoge a abe by the legal requremet aocate wth thee rght.? Uer may owloa a prt oe copy of ay publcato from the publc portal for the purpoe of prvate tuy or reearch? You may ot further trbute the materal or ue t for ay proft-makg actvty or commercal ga? You may freely trbute the URL etfyg the publcato the publc portal Take ow polcy If you beleve that th ocumet breache copyrght, pleae cotact u provg etal, a we wll remove acce to the work mmeately a vetgate your clam. Dowloa ate: 11. me. 016

2 No GRADIENT ESTIMATION SCHEMES FOR NOISY FUNCTIONS By Ruu Brekelma, Loeke Dree, Herbert Hamer a Dck e Hertog February 003 ISSN

3 Graet etmato cheme for oy fucto Ruu Brekelma 1, Loeke Dree, Herbert Hamer 3,DckeHertog 3,4 1 Tlburg Uverty, CetER Apple Reearch Cetre for Quattatve Metho BV, Ehove 3 Tlburg Uverty, Departmet of Ecoometrc a OR, P.O. Box 90153, 5000 LE Tlburg, The Netherla, e-mal: D.eHertog@uvt.l 4 Correpog author. February 7th, 003 Abtract I th paper we aalyze fferet cheme for obtag graet etmate whe the uerlyg fucto oy. Goo graet etmato e.g. mportat for olear programmg olver. A a error crtero we take the orm of the fferece betwee the real a etmate graet. Th error ca be plt up to a etermtc a a tochatc error. For three fte fferece cheme a two Deg of Expermet (DoE cheme we aalyze both the etermtc a the tochatc error. We alo erve optmal tep ze for each cheme, uch that the total error mmze. Some of the cheme have the ce property that th tep ze alo mmze the varace of the error. Bae o thee reult we how that to obta goo graet etmate for oy fucto t worthwhle to ue DoE cheme. We recomme to mplemet uch cheme NLP olver. Key wor: Deg of Expermet, fte fferece, graet etmato, oy fucto 1 Itroucto We are terete a fucto f : R R a more pecfcally t graet f(x. The fucto f ot explctly kow a we caot oberve t exactly. All obervato are the reult of fucto evaluato, whch are ubject to certa perturbato error.

4 Hece, for a fxe x R we oberve a approxmato g(x f(x+ε(x. (1 The error term ε(x repreet a raom compoet. We aume that the error term (1 are... raom error wth E[ε(x] 0 a V [ε(x] σ, hece the error term o ot epe o x. Note that g ca alo be a computer mulato moel. Eve etermtc mulato moel are ofte oy ue to all k of umercal error. I th paper we aalyze both fte fferece cheme a Deg of Expermet (DoE cheme for obtag graet etmato. I all thee cheme the graet etmate by obervg the fucto value everal pot the eghborhoo of x, ug fte tep ze h. We compare the reultg error mae the graet etmato ue to both the preece of oe a the etermtc approxmato error ( lack of ft. It wll appear that DoE cheme are worthwhle alteratve for fte fferece cheme the cae of oy fucto. Moreover, we wll erve effcet tep ze for the fferet cheme, uch that the total error (um of etermtc a tochatc error mmze. We wll compare thee tep ze to thoe whch mmze the varace of the total error. Graet play a mportat role all k of optmzato techque. I mot olear programmg (NLP coe, frt-orer or eve eco-orer ervatve are ue. Sometme thee ervatve ca be calculate ymbolcally: recet year automatc fferetato ha bee evelope; ee e.g. [7] a [3]. Although th becomg more a more popular, there are tll may optmzato techque whch fte fferecg ue to approxmate the ervatve. I almot every NLP coe uch fte fferece cheme are mplemete. Fte fferece cheme have alo bee apple to problem wth tochatc fucto. Kefer a Wolfowtz [8] were the frt to ecrbe the o-calle tochatc (qua graet; ee alo []. Metho bae o tochatc qua graet are tll ubject of much reearch; for a overvew ee [6]. So, although fte fferece cheme orgate from obtag graet etmato for etermtc fucto, they are alo apple to tochatc fucto. Alo the fel of Deg of Expermet (DoE, cheme are avalable for obtag graet etmato. Some popular cheme are full or fractoal factoral cheme, clug Plackett-Burma cheme. Cotrary to fte fferecg, thee cheme take

5 3 oe to accout. The cheme are uch that, for example, the varace of the etmator a mall a poble. However, mot DoE cheme aume a pecal form of the uerlyg moel, e.g. polyomal, a lack of ft uually ot take to accout. I [4] a [5] alo lack of ft take to accout bee the oe. I thoe paper t aalyze what happe whe the potulate lear (rep. quaratc moel mpecfe, ue to the true moel tructure beg of eco (rep. thr orer. I thee two paper ew DoE cheme are erve by mmzg the tegrate mea quare error for ether the prector or the graet. However, we thk that uch etmato are le valuable for optmzato purpoe ce the tegrate mea quare error ot a goo meaure for the graet oe pot. Moreover, the uerlyg aumpto thoe paper tll that the real moel quaratc ( [4] or thr orer ( [5] whch ot ecearly true. The remaer of th paper orgaze a follow. I Secto we aalyze three fte fferece cheme for obtag graet etmato. I Secto 3 we o the ame for two DoE cheme. I Secto 4 we compare the error of all the fve cheme. We e wth ome cocluo Secto 5. Graet etmato ug fte fferecg.1 Forwar fte fferecg Oe clacal approach to etmate the graet of f to apply forwar fte fferecg (FFD to the approxmatg fucto g, efe (1. I th cheme, a etmator of the partal ervatve, f(x x ( 1,...,, obtae by ˆβ FFD (h g(x + he g(x, h > 0, ( h where h the tep ze a e the -th ut vector. Ug (1 a Taylor formula, we ca rewrte the etmator a ˆβ FFD f(x + he f(x+ε(x + he ε(x h f(x x (3 + 1 het f(x + ζhe e + ε(x + he ε(x, (4 h whch 0 ζ 1. We are ow terete how goo th etmator. Note that E[ˆβ FFD ] f(x x + O(h (5

6 4 VAR[ˆβ FFD ] σ. (6 h The etmator error ε(x: ˆβ FFD a ˆβ FFD j Cov[ˆβ FFD FFD, ˆβ j ] E are correlate, becaue both epe o the raom ( (ˆβ FFD E[ˆβ FFD ](ˆβ FFD j E[ˆβ FFD j ] 1 h E ((ε(x + he ε(x(ε(x + he j ε(x 1 h E ( ε(x σ h, j. However, we are ot oly terete the error of the vual ervatve, but more the error mae the reultg etmate graet. A logcal meaure for the qualty of our graet etmato the mea quare error: E ( ˆβFFD f(x. Not oly the expectato mportat, but alo the varace VAR( ˆβFFD f(x, ce hgh varace mea that we ru the rk that the error a real tuato much hgher (or lower tha expecte. Suppoe for example that two mulato cheme have the ame expecte mea quare error, the we prefer the cheme wth the lowet varace. The varace ca alo be ue etermg the optmal tep ze h, a we wll ee Secto 4. By efg the etermtc error error FFD a the tochatc error we get error FFD E f(x+he 1 f(x ḥ. f(x+he f(x h ε(x+he 1 ε(x ḥ. ε(x+he ε(x h f(x ( ˆβFFD f(x ( error FFD + E error FFD.

7 5 From(3weealyervethat error FFD 1 4 h D, whch D the maxmal eco orer ervatve of f(x. Let u ow aalyze the tochatc error. The frt part of the followg theorem well-kow the lterature; ee ([10]. Theorem 1 For FFD we have ( E error FFD σ ( h VAR error FFD VAR( ˆβFFD f(x VAR h 4 [ (M4 σ 4 +M 4 +3σ 4] ( error FFD +σ D whch M 4 the fourth momet of ε(x (1,.e. M 4 E(ε(x 4. Proof. By efg ε ε(x + he, 1,...,, a ε 0 ε(x, we have for forwar fte fferecg E( error FFD 1 ( h E (ε ε 0 1 ( h E (ε + ε 0 ε ε 0 (7 σ h, (8 whch prove the frt part of the theorem. Coerg the eco part, we have VAR( error FFD E( error FFD 4 E ( error FFD. (9 Let u ow cocetrate o the frt term of the rght-ha e of (9: E( [ error FFD 4 1 h E (ε 4 + ε 0 ε ε 0 ] (ε j + ε 0 ε j ε 0 j 1 h [E ( ( ( ε 4 ε j + E ε ε 0 E ε εj ε 0 +E ( ( ( ε 0 ε j + E ε 0 ε 0 E ε 0 εj ε 0 E ( ( ε ε 0 ε j E ε ε 0 ε 0 +4E ( ε ε 0 εj ε 0 ] 1 h 4 [(M 4 + ( 1σ 4 + σ σ 4 + M σ 4 ] 1 h 4 [ (M 4 +3σ 4 +(M 4 +3σ 4 ] Subttutg th reult a the quare of (8 to (9, we have the eco part of the theorem. To prove the thr part, frt oberve that

8 6 ( ˆβFFD VAR f(x VAR( error FFD Further ote that a (error FFD E( error FFD error FFD + error FFD + error FFD 4 + E( error FFD 4 E ( error FFD + error FFD 4 + error FFD E( error FFD +4E((error FFD T error FFD ( +4E error FFD (error FFD T error FFD +4 error FFD E( ( error FFD T error FFD error FFD 4 error FFD E( error FFD E ( error FFD VAR( error FFD +4E((error FFD T error FFD + ( error FFD 4E (error FFD T error FFD VAR( error FFD +4 (error FFD E(error FFD + 4 (error FFD E(error FFD 3. (10 ( 1 E(error FFD 4 h D ( σ 1 σ D (11 h (error FFD E(error FFD 3 0, (1 ce E(error FFD 3 1 h E(ε 3 ε ( h E (ε 3 3ε ε 0 +3ε ε 0 ε (13 Fally, ubttutg (11 a (1 to (10 reult to the thr part of the theorem. Cetral fte fferecg A varat of the forwar fte fferecg (FFD the cetral fte fferecg (CFD approach. I th cheme, a etmato of the partal ervatve, f(x x ( 1,...,, obtae by ˆβ CFD (h g(x + he g(x he, h > 0, (14 h

9 7 where h the tep ze a e the -th ut vector. Ug (1 we ca rewrte the etmate a ˆβ CFD f(x + he f(x he +ε(x + he ε(x he h f(x x (15 + h 1 3 f(x + ζ 1 he [e,e,e ]+ h 1 3 f(x + ζ he [e,e,e ] (16 + ε(x + he ε(x he, (17 h where the lat equalty follow from Taylor formula f(x + he f(x+h f(x x whch 0 ζ 1 1, a + 1 h e T f(xe + h3 6 3 f(x + ζ 1 he [e,e,e ] f(x he f(x h f(x + 1 x h e T f(xe h3 6 3 f(x + ζ he [e,e,e ] whch 0 ζ 1. Let u frt aalyze the vual ervatve: a E[ˆβ CFD ] f(x + O(h (18 x VAR[ˆβ CFD ] σ h. (19 Cotrary to the FFD etmato, the etmato ˆβ CFD a ˆβ CFD j are ot correlate: ( Cov[ˆβ CFD CFD, ˆβ j ] E (ˆβ CFD E[ˆβ CFD ](ˆβ CFD j E[ˆβ CFD j ] 1 h E[(ε(x + he ε(x he (ε(x + he j ε(x he j ] 0, j. We ow aalyze the mea quare error crtero E ( ˆβCFD f(x, a t varace VAR( ˆβCFD f(x.

10 8 By efg error CFD f(x+he 1 f(x he 1 ḥ. f(x+he f(x he h f(x a error CFD ε(x+he 1 ε(x he 1 ḥ. ε(x+he ε(x he h we get ( ˆβCFD E f(x ( error CFD error CFD + E From (15 t eay to verfy that error CFD 1 36 h4 D3, whch D 3 the maxmal thr orer ervatve of f(x. Let u ow aalyze the tochatc error. The frt part of the followg theorem well-kow the lterature; ee ([10]. Theorem For CFD we have: VAR ( error CFD E σ ( h error CFD VAR [ M4 + σ 4] ( 8h 4 ˆβCFD f(x VAR( error CFD h σ D3. Proof. By efg ε ε(x + he, ε ε(x he,1,...,, a ε 0 ε(x we have for CFD E( error CFD 1 ( 4h E (ε ε 1 ( 4h E (ε + ε ε ε (0 σ h, (1 whch prove the frt part of the theorem. For the varace we have: VAR( error CFD E( error CFD 4 E ( error CFD (

11 9 Let u ow cocetrate o the frt term of the rght-ha e (: E( error CFD h E (ε 4 + ε ε ε (ε j + ε j ε j ε j j 1 16h [E ( ( ( ε 4 ε j + E ε ε j E ε εj ε j +E ( ( ( ε ε j + E ε ε j E ε εj ε j E ( ( ε ε ε j E ε ε ε j +4E ( ε ε εj ε j ] 1 16h [(M ( 1σ 4 + σ σ 4 +(M 4 + ( 1σ σ 4 ] 1 ( M4 + ( +1σ 4 8h 4 Subttuto of th reult a the quare of (1 to formula ( prove the eco part of the theorem. The lat part of the theorem follow mlar a the proof of the lat part of the prevou theorem: ( ˆβCFD VAR f(x VAR( error CFD +4 (error CFD E(error CFD + Th coclue the proof. 4 (error CFD VAR( error CFD VAR( error CFD E(error CFD 3 +4( 1 36 h4 D3( σ h h σ D3. The reult of th theorem ca be mply checke for a pecal cae. Suppoe that all ε(x are taar ormal trbute. The by ormalzg the tochatc error through the varace (ee (19, we kow that h error CFD σ (3 the um of quare tochatc ormally trbute varable, ce ( error CFD ε ε ormally trbute. Hece (3 χ ( trbute, wth expectato a varace. So,weget E( error CFD σ h, whch exactly the reult of the theorem. Furthermore, VAR( error CFD σ4 4h σ4 4 h, 4 whch alo agree wth the reult of the theorem, ce for a ormal trbuto we have M 4 3σ 4.

12 10.3 Replcate cetral fte fferecg To ecreae the tochatc error oe ca repeat cetral fte fferecg K tme. We call th replcate cetral fte fferecg (RCFD. Of coure the etermtc error wll ot chage by og replcato. The ext theorem how the expectato a varace of the reultg tochatc error. Theorem 3 For RCFD we have: VAR E ( error RCF D ( error RCF D σ h K [ M4 +(4K 3σ 4] 8h 4 K 3 VAR ( ˆβRCF D f(x VAR( error RCF D K h σ D 3. Proof. By efg ε k ε k (x + he, ε,k ε k (x he,1,...,, k 1,..., K a ε 0k ε k (x, where k eote the k-th replcate, we have for RCFD E( ( ( error RCF D 1 4h K E (ε k ε,k (4 k ( 1 4h K E (ε k ε l ε,k ε,l + ε,k ε,l ε,k ε,l (5,k,l σ h K, (6 whch prove the frt part of the theorem. For the varace we have: VAR( error RCF D E( error RCF D 4 E ( error RCF D (7

13 11 Let u ow cocetrate o the frt term of the rght-ha e (7: E( error RCF D 4 ( 1 16h 4 K E (ε 4 k ε l ε,k ε,l + ε,k ε,l ε,k ε,l,k,l ( ( 1 16h 4 K [E ε 4 k ε l + E,k,l ( ( +E ε,k ε,l + E,k,l ( +E ε,k ε,l ε k ε l + E,k,l,k,l ε k ε l ε,k ε,l,k,l,k,l ε,k ε,l ε,k ε,l,k,l,k,l ( ε,k ε,l,k,l +E ( ε,k ε,l ε,k ε,l + E,k,l,k,l ( ε,k ε,l ],k,l 1 16h 4 K 4 [[KM 4 + K ( 1σ 4 +3K(K 1σ 4 ]+[K σ 4 ] +[K σ 4 ]+[K σ 4 ] +[K σ 4 ]+[KM 4 + K ( 1σ 4 +3K(K 1σ 4 ] +[K σ 4 ]+[K σ 4 ]] 1 16h 4 K [KM 4 4 +(K ( 1 + 6K(K 1 + K +4K σ 4 ] Subttuto of th reult a the quare of formula (6 to formula (7 prove the eco part of the theorem. Fally, the thr part ca be erve almot etcal a the proof of the prevou theorem. 3 Graet etmato ug DoE 3.1 Plackett-Burma We ow aalyze Deg of Expermet (DOE cheme for etmatg the graet. Let u tart wth the Plackett-Burma cheme. Suppoe that we have a et of vector k R (k 1,...,N wth k 1a that we oberve g(x + h k for fxe x R a h>0. Defe the matrx 1 h T 1 X :... (8 1 h T N

14 1 Now uppoe that N, wth +1 N +4, a multple of four. The the Plackett- Burma cheme ca be wrtte a 1 p T 1 H.., 1 p T N where p k { 1, 1}. Th o-calle Haamar matrx ha the property H T H NI, where I the etty matrx. For more formato, ee [1] or [8] a for a example ee the Appex. Now let the vector k (8 be efe by k p k, k 1,...,N. It the follow that ( X T X ag N, h N,..., h N. The vector cotag the fucto value of f at x a the graet ca be etmate by the OLS etmator (ˆβPB g(x + h 1 0 (X T X 1 X T ˆβ PB. g(x + h N f(x + h 1 ε(x + h 1 (X T X 1 X T. +(X T X 1 X T.. f(x + h N ε(x + h N Frt ote that E[ˆβ PB 0 ]f(x+o(h, V[ˆβ PB 0 ] 1 +1 σ E[ˆβ PB ] f(x x + O( h, V[ˆβ PB ] σ, (+1h 1,...,. (9 Furthermore, ce the colum of X are orthogoal, we have Cov[ˆβ PB PB, ˆβ j ]0, j.

15 13 Now efg D a the X matrx exclug the frt colum, a f(x + h 1 error PB h N DT. f(x f(x + h N ε(x + h 1 error PB h N DT. ε(x + h N we have ( ˆβPB E f(x ( error PB error PB + E. Let u ow cocetrate o the etermtce error. Ug Taylor formula f(x + h k f(x+h T k f(x+ h T k f(x + ζh k k. whch 0 ζ 1, t eay to erve that error PB h D 4 whch D a overall upper bou for the eco orer ervatve. Cocerg the expectato a the varace of the tochatc error we have the followg theorem. Theorem 4 For Plackett-Burma eg we have: VAR E ( error PB σ Nh ( error PB VAR 4 N 3 h ( 4 ˆβPB f(x VAR( error PB Proof. For the Plackett-Burma cheme we have: error PB h N DT ν hn P T ν whch P the H matrx exclug the frt colum, a ε(x + h 1 ν.. ε(x + h N (M 4 +( N 3σ4 + 3 σ D N

16 14 We ca ow erve the followg: E( error PB h N E( P T ν (30 h N Nσ σ (31 Nh whch prove the frt part of the theorem. For the varace we have: VAR( error PB E( error PB 4 E ( error PB Let u ow cocetrate o the frt term of the rght-ha e of (3:. (3 E( error PB 4 h 4 N 4 E h 4 N 4 E h 4 N 4 E h 4 N [E 4 ( +E +E ( ( (P j ε j k ( P j P kj ε ε k,j,k ( (P kj ε k P j P kj P r P tr ε ε k ε ε t,j,k,r,,t ( P j P kj P r P kr ε ε k,j,k,r P j P j P r P r ε ε,j,,r ( P j P j P r P r ε 4,j,r ], where the lat equalty hol ce the expectato of the term ε ε k ε ε t, ε 3 ε k a ε ε k ε are zero. We ow cocetrate o the three term the lat equalty. For the frt term we have ( ( ( E P j P kj P r P kr ε ε k P j P r P kj P kr σ 4 + P j P j P kj P kj,j,k,r,j,r j k,,j k, P j P r ( P j P r +(N 1σ 4,j,r j ( 1N + (N 1σ 4 N(N σ 4. Moreover, for the eco term t hol ( E P j P j P r P r ε ε N(N 1σ 4.,j,,r σ 4

17 15 Forthethrtermwehave: ( E P j P j P r P r ε 4 NM 4.,j,r Subttutg thee reult a the quare of (31 to (3, we have prove the eco part of the theorem. The thr part of the theorem follow mlar a the proof of the lat part of Theorem 1: ( ˆβPB VAR f(x VAR( error BP +4 (error PB E(error PB + Th coclue the proof. 4 (error PB VAR( error PB E(error PB 3 +4( 1 4 h D( σ Nh +0 VAR( error PB + 3 σ D N. 3. Factoral eg Factoral eg are bae o the ame prcple a Plackett-Burma cheme, but ow N for full factoral eg a N p,p, for fractoal factoral eg; for more formato ee [1] or [8], a for a example ee the Appex. For the etermtc error we ca erve a better bou tha for Plackett-Burma cheme. Aga we have error FD f(x + h 1 h N DT. f(x. f(x + h N Now ug Taylor formula f(x + h k f(x+h T k f(x+ h T k f(x k + h3 6 3 f(x + ζh k [ k, k, k ], (33 whch 0 ζ 1, a ug the fact that factoral eg for each vector k there ext exactly oe other vector j the factoral eg cheme uch that k j,we obta by ag thee two vector: f(x + h k f(x + h j h T k f(x+ h3 3 D 3

18 16 whch D 3 a overall upper bou for the thr orer ervatve. Combg all N/ par of vector we get error FD f(x + h 1 h N DT. f(x f(x + h N h 4 D Cocerg the tochatg error we ca erve the followg reult. Theorem 5 For factoral eg we have: VAR E ( error FD σ Nh ( VAR error FD 4 N 3 h ( 4 ˆβFD f(x VAR( error FD (M 4 +( N 3σ N 3 h σ D3. Proof. Cocerg the frt a eco part we ca erve the ame reult a for Plackett-Burma eg the ame way. We therefore omt the proof of thee part. The thr part of the theorem follow mlar a the proof of the lat part of Theorem 1: ( ˆβFD VAR f(x VAR( error FD +4 (error FD E(error FD + 4 (error FD E(error FD 3. VAR( error FD +4( 1 36 h4 D3( σ Nh +0. VAR( error FD + 1 9N 3 h σ D3. 4 Comparo of the fve cheme I the prevou ecto we have erve both the etermtc a the tochatc etmato error for everal cheme; ee Table 1. The etermtc error are creag the tep ze h, whle the tochatc error are ecreag h. The expreo for the total error are covex fucto h. It traghtforwar to calculate the optmal tep

19 17 ze for each cheme uch that the total error mmze. The reult are metoe the lat colum of Table 1. Of coure, uually we o ot kow the value for σ,d a D 3. However, for a practcal problem we mght etmate thee value by amplg. Moreover, thee optmal tep ze gve ome cato; e.g., the tep ze are creag σ a ecreag N,D, a D 3, whch agree wth our tuto. #eval error E( error opt. h e Forwar FD h D Cetral FD 1 36 h4 D 3 Replcate CFD K 1 36 h4 D 3 σ h 4 σ h 6 σ h K 8 σ D 9 σ D σ KD 3 Plackett-Burma +1 N h D σ Nh 4 4 σ ND Factoral N p 1 36 h 4 D 3 σ Nh 18 6 σ ND 3 Table 1: Overvew of the umber of evaluato a the error for both fte fferece a DoE cheme, a the optmal tep ze uch that the total error mmze. Forwar FD Cetral FD Replcate CFD VAR( error VAR( error +error opt. h v h 4 [(M 4 σ4 +M 4 +3σ 4 ] 8h 4 (M 4 +σ4 8h 4 K 3 [M 4+(4K 3σ 4 ] h 4 [(M 4 σ4 +M 4 +3σ 4 ]+σ D 8h 4 (M 4 +σ h σ D 3 6 9(M 4 +σ 4 σ D 3 8h 4 K 3 [M 4+(4K 3σ 4 ]+ 18K 1 h σ D3 6 9(M 4 +(4K 3σ 4 K σ D 3 Plackett-Burma 4 N 3 h 4 (M 4 +( N 3σ4 4 N 3 h 4 (M 4 +( N 3σ4 + 3 σ D N Factoral 4 N 3 h 4 (M 4 +( N 3σ4 4 N 3 h 4 (M 4 +( N 3σ N 3 h σ D (M 4 +( N 3σ4 N σ D 3 Table : Overvew of the varace of the error vector for both fte fferece a DoE cheme a the optmal tep ze to mmze the varace. From the lterature we kow that CFD gve a much lower etermtc error tha FFD. Cocerg the tochatc error we ee from the table that the CFD cheme four tme better tha FFD. However, the umber of evaluato two tme more. To ave evaluato, we ca ue a Plackett-Burma eg: t umber of evaluato mlar to the FFD cheme, but the tochatc error two tme lower; the etermtc error, however, tme hgher. Full or fractoal factoral eg have a much

20 18 lower etermtc error tha Plackett-Burma cheme. The tochatc error mlar, but ce the umber of evaluato hgher tha for a Plackett-Burma cheme the tochatc error ca be mae much lower by creag N. However th reult more evaluato. Oberve alo that the etermtc error for Plackett-Burma a factoral cheme are epeet of the umber of evaluato, N. For the factoral cheme th alo mea that we ca ecreae the tochatc error by creag N, wthout affectg the etermtc error. Cocerg the varace of the tochatc error t appear that CFD, Plackett-Burma a factoral cheme are much better tha FFD. Whe comparg RCFD a factoral cheme t appear that the reult are mlar, ceforagoocomparowehavetotaken K. Note, however, that the cae of umercal oe, e.g. may etermtc mulato, RCFD ot applcable, ce replcate wll lea to the ame outcome. For uch cae factoral cheme are ueful. I Table we have lte the varace of the tochatc error a the total error. Note that the calculato for the optmal tep ze h e Table 1 the varace of the error are ot take to accout. Oe ca alo eterme a fferet tep ze by e.g. mmzg the expecte error plu a certa umber tme the taar evato. It ca ealy be verfe that th wll creae the optmal tep ze h. I the lat colum of Table we have calculate the optmal tep ze uch that the total varace mmze. Th calculato ot poble for FFD a Plackett-Burma ce thoe varace are ecreag fucto h. The optmal tepze h v for the other cheme reemble the correpog h e. Suppoe for example that all ε(x are taar ormal trbute, the t ca ealy be verfe that h v 6 h e 1.1h e, ce the M 4 3σ 4. Th mea that the tep ze h e whch mmze the total error equal approxmately the tep ze whch mmze the upper bou for the varace of the error. Th property a avatage of the cheme CFD, RCFD a FD above CFD a PB. I th paper we focu o the etmato of graet. However, ote that CFD, Plackett- Burma, a factoral cheme alo elver better etmato for the fucto value. Thee better etmato ca alo be valuable for NLP olver. Cocerg the amout of work eee to calculate the graet etmato, we emphaze that the etmato bae o the DoE cheme ee N ato/ubtracto a multplcato, whle FFD a CFD ee ato/ubtracto a multplcato a RCFD ee K ato/ubtracto a multplcato. So, the extra amout of work eee DOE cheme lmte

21 19 5 Cocluo I the prevou ecto we have cue everal metho for etmatg the graet of a fucto that ubject to... raom error. The error that we make whe etmatg the graet ca be plt to two part: a etermtc error a a tochatc error. The etermtc error are becaue we o ot oberve the fucto exactly at x, but the eghborhoo of x ug fte tep ze h. The tochatc error are becaue of the oe. We have erve upper bou for both the etermtc a tochatc error. Bae o thee upper bou we have cue the avatage a avatage of three fte fferece cheme a two DoE cheme. The cocluo that whe the uerlyg fucto ee oy the (fractoal or full factoral DoE cheme are ueful to reuce the tochatc error. Such cheme o ot vary the varable oe at a tme, but vary all varable multaeouly. The error for factoral cheme are exactly the ame a for replcate cetral fte fferece, but cae of umercal oe we ca ue factoral cheme whle replcate are meagle. Plackett-Burma cheme are ueful whe the evaluato are expeve. The tochatc error of thee cheme are two tme lower tha FFD, but the etermtc error hgher. Moreover, our error aaly cate how to chooe the tep ze h. Italo how that for CFD, RCFD a FD-cheme the tep ze whch mmze the total error, alo mmze the varace of the error. The DoE cheme ca be ealy clue the NLP olver to etmate graet. Ackowlegemet. We woul lke to thak our colleague Jack Kleje for h ueful remark o a earler vero of th paper, a Gul Gurka for provg u wth relevat lterature. Referece [1] Box, G.E.P., W.G. Huter, a J.S. Huter (1987, Stattc for Expermeter, Wley, New York. [] Blum, J.R. (1954, Multmeoal Stochatc Approxmato Metho, Aal of Mathematcal Stattc 5, [3] Dxo, L.C.W. (1994, O Automatc Dfferetato a Cotuou Optmzato, NATO Avace Stuy Ittute Sere 434,

22 0 [4] Doohue, J.M., E.C. Houck a R.H. Myer (1993, Smulato Deg for Cotrollg Seco-Orer Ba Frt-Orer Repoe Surface, Operato Reearch 41, [5] Doohue, J.M., E.C. Houck a R.H. Myer (1995, Smulato Deg for the Etmato of Quaratc Repoe Surface Graet the Preece of Moel Mpecfcato, Maagemet Scece 41 (, [6] Ermolev, Y. (1980 Stochatc Quagraet Metho, : Y. Ermolev a R.J- B. Wet, e., Numercal Techque for Stochatc Optmzato, Sprger Verlag, Chapter 6. [7] Grewak, A. (1989 O Automatc Dfferetato, : M. Ir a K. Taabe, e., Mathematcal Programmg, KTK Scetfc Publher, Tokyo, [8] Kefer, J. a J. Wolfowtz (195, Stochatc Etmato of a Regreo Fucto, Aal of Mathematcal Stattc 3, [9] Motgomery, D.C. (1984, Deg a Aaly of Expermet, eto, Wley, New York. [10] Zaza, M.A. a R. Sur (1993 Covergece Rate of Fte-Dfferece Setvty Etmate for Stochatc Sytem, Operato reearch 41 (4, Appex: DoE cheme I Table 3, four evaluato cheme are gve for 4. Note that for Plackett-Burma we have N 8, whch mea that 8 evaluato are eee. I th cae the umber of evalato for Plackett-Burma the ame a for CFD; geeral, however, the umber of evaluato eee by CFD more. Moreover, t eay to verfy that the orthogoalty property hol for th pecfc full factoral a Plackett-Burma cheme. I fact, Plackett-Burma cheme were evelope to reuce the umber of evaluato, but uch that the orthogoalty property tll hol. There o ee for tabulatg the DoE cheme, ce there a mple proceure for geeratg uch cheme.

23 1 FFD CFD Plackett-Burma Full factoral x 1 x x 3 x 4 x 1 x x 3 x 4 x 1 x x 3 x 4 x 1 x x 3 x Table 3: Evaluato cheme for 4factor.

Simple Linear Regression Analysis

Simple Linear Regression Analysis LINEAR REGREION ANALYSIS MODULE II Lecture - 5 Smple Lear Regreo Aaly Dr Shalabh Departmet of Mathematc Stattc Ida Ittute of Techology Kapur Jot cofdece rego for A jot cofdece rego for ca alo be foud Such

More information

1. Linear second-order circuits

1. Linear second-order circuits ear eco-orer crcut Sere R crcut Dyamc crcut cotag two capactor or two uctor or oe uctor a oe capactor are calle the eco orer crcut At frt we coer a pecal cla of the eco-orer crcut, amely a ere coecto of

More information

Linear Approximating to Integer Addition

Linear Approximating to Integer Addition Lear Approxmatg to Iteger Addto L A-Pg Bejg 00085, P.R. Cha apl000@a.com Abtract The teger addto ofte appled cpher a a cryptographc mea. I th paper we wll preet ome reult about the lear approxmatg for

More information

Reaction Time VS. Drug Percentage Subject Amount of Drug Times % Reaction Time in Seconds 1 Mary John Carl Sara William 5 4

Reaction Time VS. Drug Percentage Subject Amount of Drug Times % Reaction Time in Seconds 1 Mary John Carl Sara William 5 4 CHAPTER Smple Lear Regreo EXAMPLE A expermet volvg fve ubject coducted to determe the relatohp betwee the percetage of a certa drug the bloodtream ad the legth of tme t take the ubject to react to a tmulu.

More information

UNIT 7 RANK CORRELATION

UNIT 7 RANK CORRELATION UNIT 7 RANK CORRELATION Rak Correlato Structure 7. Itroucto Objectves 7. Cocept of Rak Correlato 7.3 Dervato of Rak Correlato Coeffcet Formula 7.4 Te or Repeate Raks 7.5 Cocurret Devato 7.6 Summar 7.7

More information

Solutions to Odd-Numbered End-of-Chapter Exercises: Chapter 17

Solutions to Odd-Numbered End-of-Chapter Exercises: Chapter 17 Itroucto to Ecoometrcs (3 r Upate Eto) by James H. Stock a Mark W. Watso Solutos to O-Numbere E-of-Chapter Exercses: Chapter 7 (Ths erso August 7, 04) 05 Pearso Eucato, Ic. Stock/Watso - Itroucto to Ecoometrcs

More information

C.11 Bang-bang Control

C.11 Bang-bang Control Itroucto to Cotrol heory Iclug Optmal Cotrol Nguye a e -.5 C. Bag-bag Cotrol. Itroucto hs chapter eals wth the cotrol wth restrctos: s boue a mght well be possble to have scotutes. o llustrate some of

More information

M2S1 - EXERCISES 8: SOLUTIONS

M2S1 - EXERCISES 8: SOLUTIONS MS - EXERCISES 8: SOLUTIONS. As X,..., X P ossoλ, a gve that T ˉX, the usg elemetary propertes of expectatos, we have E ft [T E fx [X λ λ, so that T s a ubase estmator of λ. T X X X Furthermore X X X From

More information

= 2. Statistic - function that doesn't depend on any of the known parameters; examples:

= 2. Statistic - function that doesn't depend on any of the known parameters; examples: of Samplg Theory amples - uemploymet househol cosumpto survey Raom sample - set of rv's... ; 's have ot strbuto [ ] f f s vector of parameters e.g. Statstc - fucto that oes't epe o ay of the ow parameters;

More information

Generalized Linear Regression with Regularization

Generalized Linear Regression with Regularization Geeralze Lear Regresso wth Regularzato Zoya Bylsk March 3, 05 BASIC REGRESSION PROBLEM Note: I the followg otes I wll make explct what s a vector a what s a scalar usg vec t or otato, to avo cofuso betwee

More information

8 The independence problem

8 The independence problem Noparam Stat 46/55 Jame Kwo 8 The depedece problem 8.. Example (Tua qualty) ## Hollader & Wolfe (973), p. 87f. ## Aemet of tua qualty. We compare the Huter L meaure of ## lghte to the average of coumer

More information

ANOVA with Summary Statistics: A STATA Macro

ANOVA with Summary Statistics: A STATA Macro ANOVA wth Summary Stattc: A STATA Macro Nadeem Shafque Butt Departmet of Socal ad Prevetve Pedatrc Kg Edward Medcal College, Lahore, Pata Shahd Kamal Ittute of Stattc, Uverty of the Puab Lahore, Pata Muhammad

More information

Quiz 1- Linear Regression Analysis (Based on Lectures 1-14)

Quiz 1- Linear Regression Analysis (Based on Lectures 1-14) Quz - Lear Regreo Aaly (Baed o Lecture -4). I the mple lear regreo model y = β + βx + ε, wth Tme: Hour Ε ε = Ε ε = ( ) 3, ( ), =,,...,, the ubaed drect leat quare etmator ˆβ ad ˆβ of β ad β repectvely,

More information

A Result of Convergence about Weighted Sum for Exchangeable Random Variable Sequence in the Errors-in-Variables Model

A Result of Convergence about Weighted Sum for Exchangeable Random Variable Sequence in the Errors-in-Variables Model AMSE JOURNALS-AMSE IIETA publcato-17-sere: Advace A; Vol. 54; N ; pp 3-33 Submtted Mar. 31, 17; Reved Ju. 11, 17, Accepted Ju. 18, 17 A Reult of Covergece about Weghted Sum for Exchageable Radom Varable

More information

REVIEW OF SIMPLE LINEAR REGRESSION SIMPLE LINEAR REGRESSION

REVIEW OF SIMPLE LINEAR REGRESSION SIMPLE LINEAR REGRESSION REVIEW OF SIMPLE LINEAR REGRESSION SIMPLE LINEAR REGRESSION I lear regreo, we coder the frequecy dtrbuto of oe varable (Y) at each of everal level of a ecod varable (X). Y kow a the depedet varable. The

More information

On a Truncated Erlang Queuing System. with Bulk Arrivals, Balking and Reneging

On a Truncated Erlang Queuing System. with Bulk Arrivals, Balking and Reneging Appled Mathematcal Scece Vol. 3 9 o. 3 3-3 O a Trucated Erlag Queug Sytem wth Bul Arrval Balg ad Reegg M. S. El-aoumy ad M. M. Imal Departmet of Stattc Faculty Of ommerce Al- Azhar Uverty. Grl Brach Egypt

More information

European Journal of Mathematics and Computer Science Vol. 5 No. 2, 2018 ISSN

European Journal of Mathematics and Computer Science Vol. 5 No. 2, 2018 ISSN Europea Joural of Mathematc ad Computer Scece Vol. 5 o., 018 ISS 059-9951 APPLICATIO OF ASYMPTOTIC DISTRIBUTIO OF MA-HITEY STATISTIC TO DETERMIE THE DIFFERECE BETEE THE SYSTOLIC BLOOD PRESSURE OF ME AD

More information

STK4011 and STK9011 Autumn 2016

STK4011 and STK9011 Autumn 2016 STK4 ad STK9 Autum 6 Pot estmato Covers (most of the followg materal from chapter 7: Secto 7.: pages 3-3 Secto 7..: pages 3-33 Secto 7..: pages 35-3 Secto 7..3: pages 34-35 Secto 7.3.: pages 33-33 Secto

More information

European Journal of Mathematics and Computer Science Vol. 5 No. 2, 2018 ISSN

European Journal of Mathematics and Computer Science Vol. 5 No. 2, 2018 ISSN Europea Joural of Mathematc ad Computer Scece Vol. 5 o., 018 ISS 059-9951 APPLICATIO OF ASYMPTOTIC DISTRIBUTIO OF MA-HITEY STATISTIC TO DETERMIE THE DIFFERECE BETEE THE SYSTOLIC BLOOD PRESSURE OF ME AD

More information

Study on Bayes Semiparametric Regression

Study on Bayes Semiparametric Regression Joural of Avace Apple Mathematc, Vol., No. 4, October 7 http://x.o.org/.66/jaam.7.4 97 Stuy o Baye Semparametrc Regreo Abulhue Saber AL-Mouel a Ameera Jaber Mohae Mathematc Departmet College of Eucato

More information

CS473-Algorithms I. Lecture 12b. Dynamic Tables. CS 473 Lecture X 1

CS473-Algorithms I. Lecture 12b. Dynamic Tables. CS 473 Lecture X 1 CS473-Algorthm I Lecture b Dyamc Table CS 473 Lecture X Why Dyamc Table? I ome applcato: We do't kow how may object wll be tored a table. We may allocate pace for a table But, later we may fd out that

More information

X ε ) = 0, or equivalently, lim

X ε ) = 0, or equivalently, lim Revew for the prevous lecture Cocepts: order statstcs Theorems: Dstrbutos of order statstcs Examples: How to get the dstrbuto of order statstcs Chapter 5 Propertes of a Radom Sample Secto 55 Covergece

More information

Maximum Walk Entropy Implies Walk Regularity

Maximum Walk Entropy Implies Walk Regularity Maxmum Walk Etropy Imples Walk Regularty Eresto Estraa, a José. e la Peña Departmet of Mathematcs a Statstcs, Uversty of Strathclye, Glasgow G XH, U.K., CIMT, Guaajuato, Mexco BSTRCT: The oto of walk etropy

More information

Trignometric Inequations and Fuzzy Information Theory

Trignometric Inequations and Fuzzy Information Theory Iteratoal Joural of Scetfc ad Iovatve Mathematcal Reearch (IJSIMR) Volume, Iue, Jauary - 0, PP 00-07 ISSN 7-07X (Prt) & ISSN 7- (Ole) www.arcjoural.org Trgometrc Iequato ad Fuzzy Iformato Theory P.K. Sharma,

More information

Econometric Methods. Review of Estimation

Econometric Methods. Review of Estimation Ecoometrc Methods Revew of Estmato Estmatg the populato mea Radom samplg Pot ad terval estmators Lear estmators Ubased estmators Lear Ubased Estmators (LUEs) Effcecy (mmum varace) ad Best Lear Ubased Estmators

More information

ROOT-LOCUS ANALYSIS. Lecture 11: Root Locus Plot. Consider a general feedback control system with a variable gain K. Y ( s ) ( ) K

ROOT-LOCUS ANALYSIS. Lecture 11: Root Locus Plot. Consider a general feedback control system with a variable gain K. Y ( s ) ( ) K ROOT-LOCUS ANALYSIS Coder a geeral feedback cotrol yte wth a varable ga. R( Y( G( + H( Root-Locu a plot of the loc of the pole of the cloed-loop trafer fucto whe oe of the yte paraeter ( vared. Root locu

More information

LINEARLY CONSTRAINED MINIMIZATION BY USING NEWTON S METHOD

LINEARLY CONSTRAINED MINIMIZATION BY USING NEWTON S METHOD Jural Karya Asl Loreka Ahl Matematk Vol 8 o 205 Page 084-088 Jural Karya Asl Loreka Ahl Matematk LIEARLY COSTRAIED MIIMIZATIO BY USIG EWTO S METHOD Yosza B Dasrl, a Ismal B Moh 2 Faculty Electrocs a Computer

More information

r y Simple Linear Regression How To Study Relation Between Two Quantitative Variables? Scatter Plot Pearson s Sample Correlation Correlation

r y Simple Linear Regression How To Study Relation Between Two Quantitative Variables? Scatter Plot Pearson s Sample Correlation Correlation Maatee Klled Correlato & Regreo How To Study Relato Betwee Two Quattatve Varable? Smple Lear Regreo 6.11 A Smple Regreo Problem 1 I there relato betwee umber of power boat the area ad umber of maatee klled?

More information

Bounds on the expected entropy and KL-divergence of sampled multinomial distributions. Brandon C. Roy

Bounds on the expected entropy and KL-divergence of sampled multinomial distributions. Brandon C. Roy Bouds o the expected etropy ad KL-dvergece of sampled multomal dstrbutos Brado C. Roy bcroy@meda.mt.edu Orgal: May 18, 2011 Revsed: Jue 6, 2011 Abstract Iformato theoretc quattes calculated from a sampled

More information

A note on testing the covariance matrix for large dimension

A note on testing the covariance matrix for large dimension A ote o tetg the covarace matrx for large dmeo Melae Brke Ruhr-Uvertät Bochum Fakultät für Mathematk 44780 Bochum, Germay e-mal: melae.brke@ruhr-u-bochum.de Holger ette Ruhr-Uvertät Bochum Fakultät für

More information

Chapter 8: Statistical Analysis of Simulated Data

Chapter 8: Statistical Analysis of Simulated Data Marquette Uversty MSCS600 Chapter 8: Statstcal Aalyss of Smulated Data Dael B. Rowe, Ph.D. Departmet of Mathematcs, Statstcs, ad Computer Scece Copyrght 08 by Marquette Uversty MSCS600 Ageda 8. The Sample

More information

MA 524 Homework 6 Solutions

MA 524 Homework 6 Solutions MA 524 Homework 6 Solutos. Sce S(, s the umber of ways to partto [] to k oempty blocks, ad c(, s the umber of ways to partto to k oempty blocks ad also the arrage each block to a cycle, we must have S(,

More information

Chapter 5 Properties of a Random Sample

Chapter 5 Properties of a Random Sample Lecture 6 o BST 63: Statstcal Theory I Ku Zhag, /0/008 Revew for the prevous lecture Cocepts: t-dstrbuto, F-dstrbuto Theorems: Dstrbutos of sample mea ad sample varace, relatoshp betwee sample mea ad sample

More information

Simple Linear Regression. How To Study Relation Between Two Quantitative Variables? Scatter Plot. Pearson s Sample Correlation.

Simple Linear Regression. How To Study Relation Between Two Quantitative Variables? Scatter Plot. Pearson s Sample Correlation. Correlato & Regreo How To Study Relato Betwee Two Quattatve Varable? Smple Lear Regreo 6. A Smple Regreo Problem I there relato betwee umber of power boat the area ad umber of maatee klled? Year NPB( )

More information

1. a. Houston Chronicle, Des Moines Register, Chicago Tribune, Washington Post

1. a. Houston Chronicle, Des Moines Register, Chicago Tribune, Washington Post Homework Soluto. Houto Chrocle, De Moe Regter, Chcago Trbue, Wahgto Pot b. Captal Oe, Campbell Soup, Merrll Lych, Pultzer c. Bll Japer, Kay Reke, Hele Ford, Davd Meedez d..78,.44, 3.5, 3.04 5. No, the

More information

INEQUALITIES USING CONVEX COMBINATION CENTERS AND SET BARYCENTERS

INEQUALITIES USING CONVEX COMBINATION CENTERS AND SET BARYCENTERS Joural of Mathematcal Scece: Advace ad Alcato Volume 24, 23, Page 29-46 INEQUALITIES USING CONVEX COMBINATION CENTERS AND SET BARYCENTERS ZLATKO PAVIĆ Mechacal Egeerg Faculty Slavok Brod Uverty of Ojek

More information

Summary of the lecture in Biostatistics

Summary of the lecture in Biostatistics Summary of the lecture Bostatstcs Probablty Desty Fucto For a cotuos radom varable, a probablty desty fucto s a fucto such that: 0 dx a b) b a dx A probablty desty fucto provdes a smple descrpto of the

More information

Ordinary Least Squares Regression. Simple Regression. Algebra and Assumptions.

Ordinary Least Squares Regression. Simple Regression. Algebra and Assumptions. Ordary Least Squares egresso. Smple egresso. Algebra ad Assumptos. I ths part of the course we are gog to study a techque for aalysg the lear relatoshp betwee two varables Y ad X. We have pars of observatos

More information

An Unbiased Class of Ratio Type Estimator for Population Mean Using an Attribute and a Variable

An Unbiased Class of Ratio Type Estimator for Population Mean Using an Attribute and a Variable Advace Comutatoal Scece ad Techology ISS 973-67 Volume, umber 7). 39-46 Reearch Ida Publcato htt://www.rublcato.com A Ubaed Cla of Rato Tye Etmator for Poulato Mea Ug a Attrbute ad a Varable Shah Bhuha,

More information

A nonsmooth Levenberg-Marquardt method for generalized complementarity problem

A nonsmooth Levenberg-Marquardt method for generalized complementarity problem ISSN 746-7659 Egla UK Joural of Iformato a Computg Scece Vol. 7 No. 4 0 pp. 67-7 A osmooth Leveberg-Marquart metho for geeralze complemetarty problem Shou-qag Du College of Mathematcs Qgao Uversty Qgao

More information

PENALIZED CHI SQUARE DISTANCE FUNCTION IN SURVEY SAMPLING

PENALIZED CHI SQUARE DISTANCE FUNCTION IN SURVEY SAMPLING Jot Stattcal Meetg - Secto o Surve Reearch Metho PEAIZED CHI SQUARE DISTACE FUCTIO I SURE SAMPIG Patrck J. Farrell a Sarer Sgh School of Mathematc a Stattc, Carleto Uvert, 5 Coloel B Drve, Ottaa, Otaro

More information

Chapter 11 Systematic Sampling

Chapter 11 Systematic Sampling Chapter stematc amplg The sstematc samplg techue s operatoall more coveet tha the smple radom samplg. It also esures at the same tme that each ut has eual probablt of cluso the sample. I ths method of

More information

NP!= P. By Liu Ran. Table of Contents. The P versus NP problem is a major unsolved problem in computer

NP!= P. By Liu Ran. Table of Contents. The P versus NP problem is a major unsolved problem in computer NP!= P By Lu Ra Table of Cotets. Itroduce 2. Prelmary theorem 3. Proof 4. Expla 5. Cocluso. Itroduce The P versus NP problem s a major usolved problem computer scece. Iformally, t asks whether a computer

More information

CHAPTER VI Statistical Analysis of Experimental Data

CHAPTER VI Statistical Analysis of Experimental Data Chapter VI Statstcal Aalyss of Expermetal Data CHAPTER VI Statstcal Aalyss of Expermetal Data Measuremets do ot lead to a uque value. Ths s a result of the multtude of errors (maly radom errors) that ca

More information

Evaluating Polynomials

Evaluating Polynomials Uverst of Nebraska - Lcol DgtalCommos@Uverst of Nebraska - Lcol MAT Exam Expostor Papers Math the Mddle Isttute Partershp 7-7 Evaluatg Polomals Thomas J. Harrgto Uverst of Nebraska-Lcol Follow ths ad addtoal

More information

Multivariate Transformation of Variables and Maximum Likelihood Estimation

Multivariate Transformation of Variables and Maximum Likelihood Estimation Marquette Uversty Multvarate Trasformato of Varables ad Maxmum Lkelhood Estmato Dael B. Rowe, Ph.D. Assocate Professor Departmet of Mathematcs, Statstcs, ad Computer Scece Copyrght 03 by Marquette Uversty

More information

Hamilton s principle for non-holonomic systems

Hamilton s principle for non-holonomic systems Das Hamltosche Przp be chtholoome Systeme, Math. A. (935), pp. 94-97. Hamlto s prcple for o-holoomc systems by Georg Hamel Berl Traslate by: D. H. Delphech I the paper Le prcpe e Hamlto et l holoomsme,

More information

Lecture 4 Sep 9, 2015

Lecture 4 Sep 9, 2015 CS 388R: Radomzed Algorthms Fall 205 Prof. Erc Prce Lecture 4 Sep 9, 205 Scrbe: Xagru Huag & Chad Voegele Overvew I prevous lectures, we troduced some basc probablty, the Cheroff boud, the coupo collector

More information

Chapter 4 Multiple Random Variables

Chapter 4 Multiple Random Variables Revew for the prevous lecture: Theorems ad Examples: How to obta the pmf (pdf) of U = g (, Y) ad V = g (, Y) Chapter 4 Multple Radom Varables Chapter 44 Herarchcal Models ad Mxture Dstrbutos Examples:

More information

System Reliability-Based Design Optimization Using the MPP-Based Dimension Reduction Method

System Reliability-Based Design Optimization Using the MPP-Based Dimension Reduction Method Sytem Relablty-Baed Deg Optmzato Ug the M-Baed Dmeo Reducto Method I Lee ad KK Cho Departmet of Mechacal & Idutral Egeerg College of Egeerg, The Uverty of Iowa Iowa Cty, IA 54 ad Davd Gorch 3 US Army RDECOM/TARDEC,

More information

Mean is only appropriate for interval or ratio scales, not ordinal or nominal.

Mean is only appropriate for interval or ratio scales, not ordinal or nominal. Mea Same as ordary average Sum all the data values ad dvde by the sample sze. x = ( x + x +... + x Usg summato otato, we wrte ths as x = x = x = = ) x Mea s oly approprate for terval or rato scales, ot

More information

Chapter 14 Logistic Regression Models

Chapter 14 Logistic Regression Models Chapter 4 Logstc Regresso Models I the lear regresso model X β + ε, there are two types of varables explaatory varables X, X,, X k ad study varable y These varables ca be measured o a cotuous scale as

More information

Assignment 5/MATH 247/Winter Due: Friday, February 19 in class (!) (answers will be posted right after class)

Assignment 5/MATH 247/Winter Due: Friday, February 19 in class (!) (answers will be posted right after class) Assgmet 5/MATH 7/Wter 00 Due: Frday, February 9 class (!) (aswers wll be posted rght after class) As usual, there are peces of text, before the questos [], [], themselves. Recall: For the quadratc form

More information

Lecture 7. Confidence Intervals and Hypothesis Tests in the Simple CLR Model

Lecture 7. Confidence Intervals and Hypothesis Tests in the Simple CLR Model Lecture 7. Cofdece Itervals ad Hypothess Tests the Smple CLR Model I lecture 6 we troduced the Classcal Lear Regresso (CLR) model that s the radom expermet of whch the data Y,,, K, are the outcomes. The

More information

Multiple Choice Test. Chapter Adequacy of Models for Regression

Multiple Choice Test. Chapter Adequacy of Models for Regression Multple Choce Test Chapter 06.0 Adequac of Models for Regresso. For a lear regresso model to be cosdered adequate, the percetage of scaled resduals that eed to be the rage [-,] s greater tha or equal to

More information

22 Nonparametric Methods.

22 Nonparametric Methods. 22 oparametrc Methods. I parametrc models oe assumes apror that the dstrbutos have a specfc form wth oe or more ukow parameters ad oe tres to fd the best or atleast reasoably effcet procedures that aswer

More information

Median as a Weighted Arithmetic Mean of All Sample Observations

Median as a Weighted Arithmetic Mean of All Sample Observations Meda as a Weghted Arthmetc Mea of All Sample Observatos SK Mshra Dept. of Ecoomcs NEHU, Shllog (Ida). Itroducto: Iumerably may textbooks Statstcs explctly meto that oe of the weakesses (or propertes) of

More information

Beam Warming Second-Order Upwind Method

Beam Warming Second-Order Upwind Method Beam Warmg Secod-Order Upwd Method Petr Valeta Jauary 6, 015 Ths documet s a part of the assessmet work for the subject 1DRP Dfferetal Equatos o Computer lectured o FNSPE CTU Prague. Abstract Ths documet

More information

THE ROYAL STATISTICAL SOCIETY HIGHER CERTIFICATE

THE ROYAL STATISTICAL SOCIETY HIGHER CERTIFICATE THE ROYAL STATISTICAL SOCIETY 00 EXAMINATIONS SOLUTIONS HIGHER CERTIFICATE PAPER I STATISTICAL THEORY The Socety provdes these solutos to assst caddates preparg for the examatos future years ad for the

More information

Part 4b Asymptotic Results for MRR2 using PRESS. Recall that the PRESS statistic is a special type of cross validation procedure (see Allen (1971))

Part 4b Asymptotic Results for MRR2 using PRESS. Recall that the PRESS statistic is a special type of cross validation procedure (see Allen (1971)) art 4b Asymptotc Results for MRR usg RESS Recall that the RESS statstc s a specal type of cross valdato procedure (see Alle (97)) partcular to the regresso problem ad volves fdg Y $,, the estmate at the

More information

Linear Regression. Can height information be used to predict weight of an individual? How long should you wait till next eruption?

Linear Regression. Can height information be used to predict weight of an individual? How long should you wait till next eruption? Iter-erupto Tme Weght Correlato & Regreo 1 1 Lear Regreo 0 80 70 80 Heght 1 Ca heght formato be ued to predct weght of a dvdual? How log hould ou wat tll et erupto? Weght: Repoe varable (Outcome, Depedet)

More information

A Primer on Summation Notation George H Olson, Ph. D. Doctoral Program in Educational Leadership Appalachian State University Spring 2010

A Primer on Summation Notation George H Olson, Ph. D. Doctoral Program in Educational Leadership Appalachian State University Spring 2010 Summato Operator A Prmer o Summato otato George H Olso Ph D Doctoral Program Educatoal Leadershp Appalacha State Uversty Sprg 00 The summato operator ( ) {Greek letter captal sgma} s a structo to sum over

More information

Chapter 3. Differentiation 3.3 Differentiation Rules

Chapter 3. Differentiation 3.3 Differentiation Rules 3.3 Dfferetato Rules 1 Capter 3. Dfferetato 3.3 Dfferetato Rules Dervatve of a Costat Fucto. If f as te costat value f(x) = c, te f x = [c] = 0. x Proof. From te efto: f (x) f(x + ) f(x) o c c 0 = 0. QED

More information

Collapsing to Sample and Remainder Means. Ed Stanek. In order to collapse the expanded random variables to weighted sample and remainder

Collapsing to Sample and Remainder Means. Ed Stanek. In order to collapse the expanded random variables to weighted sample and remainder Collapg to Saple ad Reader Mea Ed Staek Collapg to Saple ad Reader Average order to collape the expaded rado varable to weghted aple ad reader average, we pre-ultpled by ( M C C ( ( M C ( M M M ( M M M,

More information

Chapter 2: Descriptive Statistics

Chapter 2: Descriptive Statistics Chapte : Decptve Stattc Peequte: Chapte. Revew of Uvaate Stattc The cetal teecy of a oe o le yetc tbuto of a et of teval, o hghe, cale coe, ofte uaze by the athetc ea, whch efe a We ca ue the ea to ceate

More information

NP!= P. By Liu Ran. Table of Contents. The P vs. NP problem is a major unsolved problem in computer

NP!= P. By Liu Ran. Table of Contents. The P vs. NP problem is a major unsolved problem in computer NP!= P By Lu Ra Table of Cotets. Itroduce 2. Strategy 3. Prelmary theorem 4. Proof 5. Expla 6. Cocluso. Itroduce The P vs. NP problem s a major usolved problem computer scece. Iformally, t asks whether

More information

On the energy of complement of regular line graphs

On the energy of complement of regular line graphs MATCH Coucato Matheatcal ad Coputer Chetry MATCH Cou Math Coput Che 60 008) 47-434 ISSN 0340-653 O the eergy of copleet of regular le graph Fateeh Alaghpour a, Baha Ahad b a Uverty of Tehra, Tehra, Ira

More information

Some distances and sequences in a weighted graph

Some distances and sequences in a weighted graph IOSR Joural of Mathematc (IOSR-JM) e-issn: 78-578 p-issn: 19 765X PP 7-15 wwworjouralorg Some dtace ad equece a weghted graph Jll K Mathew 1, Sul Mathew Departmet of Mathematc Federal Ittute of Scece ad

More information

Taylor s Series and Interpolation. Interpolation & Curve-fitting. CIS Interpolation. Basic Scenario. Taylor Series interpolates at a specific

Taylor s Series and Interpolation. Interpolation & Curve-fitting. CIS Interpolation. Basic Scenario. Taylor Series interpolates at a specific CIS 54 - Iterpolato Roger Crawfs Basc Scearo We are able to prod some fucto, but do ot kow what t really s. Ths gves us a lst of data pots: [x,f ] f(x) f f + x x + August 2, 25 OSU/CIS 54 3 Taylor s Seres

More information

Consensus Tracking of Multi-Agent Systems with Constrained Time-delay by Iterative Learning Control

Consensus Tracking of Multi-Agent Systems with Constrained Time-delay by Iterative Learning Control Coeu Trackg of Mult-Aget Sytem wth Cotrae Tme-elay by Iteratve Learg Cotrol Yogl R Yog Fag, Hogwe Yu School of Commucato a Iformato Egeerg, Shagha Uverty, Shagha 00444, Cha E-mal: ryogl@163.com yfag@taff.hu.eu.c

More information

Solving Constrained Flow-Shop Scheduling. Problems with Three Machines

Solving Constrained Flow-Shop Scheduling. Problems with Three Machines It J Cotemp Math Sceces, Vol 5, 2010, o 19, 921-929 Solvg Costraed Flow-Shop Schedulg Problems wth Three Maches P Pada ad P Rajedra Departmet of Mathematcs, School of Advaced Sceces, VIT Uversty, Vellore-632

More information

(2014) ISSN

(2014) ISSN Estraa, Eresto a e la Pea, Jose too (4) Maxmum walk etropy mples walk regularty. Lear lgebra a ts pplcatos, 458. pp. 54-547. ISSN 4-3795, http://x.o.org/.6/j.laa.4.6.3 Ths verso s avalable at https://strathprts.strath.ac.uk/5879/

More information

ECON 5360 Class Notes GMM

ECON 5360 Class Notes GMM ECON 560 Class Notes GMM Geeralzed Method of Momets (GMM) I beg by outlg the classcal method of momets techque (Fsher, 95) ad the proceed to geeralzed method of momets (Hase, 98).. radtoal Method of Momets

More information

2006 Jamie Trahan, Autar Kaw, Kevin Martin University of South Florida United States of America

2006 Jamie Trahan, Autar Kaw, Kevin Martin University of South Florida United States of America SOLUTION OF SYSTEMS OF SIMULTANEOUS LINEAR EQUATIONS Gauss-Sedel Method 006 Jame Traha, Autar Kaw, Kev Mart Uversty of South Florda Uted States of Amerca kaw@eg.usf.edu Itroducto Ths worksheet demostrates

More information

PROJECTION PROBLEM FOR REGULAR POLYGONS

PROJECTION PROBLEM FOR REGULAR POLYGONS Joural of Mathematcal Sceces: Advaces ad Applcatos Volume, Number, 008, Pages 95-50 PROJECTION PROBLEM FOR REGULAR POLYGONS College of Scece Bejg Forestry Uversty Bejg 0008 P. R. Cha e-mal: sl@bjfu.edu.c

More information

to the estimation of total sensitivity indices

to the estimation of total sensitivity indices Applcato of the cotrol o varate ate techque to the estmato of total sestvty dces S KUCHERENKO B DELPUECH Imperal College Lodo (UK) skuchereko@mperalacuk B IOOSS Electrcté de Frace (Frace) S TARANTOLA Jot

More information

( ) = ( ) ( ) Chapter 13 Asymptotic Theory and Stochastic Regressors. Stochastic regressors model

( ) = ( ) ( ) Chapter 13 Asymptotic Theory and Stochastic Regressors. Stochastic regressors model Chapter 3 Asmptotc Theor ad Stochastc Regressors The ature of eplaator varable s assumed to be o-stochastc or fed repeated samples a regresso aalss Such a assumpto s approprate for those epermets whch

More information

International Journal of Pure and Applied Sciences and Technology

International Journal of Pure and Applied Sciences and Technology It J Pure Appl Sc Techol, () (00), pp 79-86 Iteratoal Joural of Pure ad Appled Scece ad Techology ISSN 9-607 Avalable ole at wwwjopaaat Reearch Paper Some Stroger Chaotc Feature of the Geeralzed Shft Map

More information

f X (x i ;θ) = n ( n logx i = 0 = θml = n/ n logx i 1 θ +1 n n 2 < 0 for all θ (θ +1) n logx i 1 ESTIMATOR: = logx i θ n for all θ θ 2 < 0 2θ 2 x 3

f X (x i ;θ) = n ( n logx i = 0 = θml = n/ n logx i 1 θ +1 n n 2 < 0 for all θ (θ +1) n logx i 1 ESTIMATOR: = logx i θ n for all θ θ 2 < 0 2θ 2 x 3 MATH 557 - EXERCISES SOLUTIONS 1 The lkelhoo the orgal parameterzato s ( ( 1 L (θ 1,θ x 1,x = θ x 1 1 x (1 θ 1 1 x 1 θ x 1 x (1 θ x If φ = θ /(1 θ, the θ = φ/(1+φ θ 1 = (φψ/(1+φψ. Ether by wrtg out the

More information

On the convergence of derivatives of Bernstein approximation

On the convergence of derivatives of Bernstein approximation O the covergece of dervatves of Berste approxmato Mchael S. Floater Abstract: By dfferetatg a remader formula of Stacu, we derve both a error boud ad a asymptotc formula for the dervatves of Berste approxmato.

More information

1 Solution to Problem 6.40

1 Solution to Problem 6.40 1 Soluto to Problem 6.40 (a We wll wrte T τ (X 1,...,X where the X s are..d. wth PDF f(x µ, σ 1 ( x µ σ g, σ where the locato parameter µ s ay real umber ad the scale parameter σ s > 0. Lettg Z X µ σ we

More information

Robust Stability Analysis of Discrete Uncertain Singularly Perturbed Time-Delay Systems

Robust Stability Analysis of Discrete Uncertain Singularly Perturbed Time-Delay Systems Robut Stablty Aaly o Dcrete Ucerta Sgularly Perturbe Tme-Delay Sytem Shg-Ta Pa a Chg-Fa Che Departmet o Computer Scece a Iormato Egeerg, Shu-Te Uverty, Kaohug, Tawa 8, R.O.C. Departmet o Electroc Egeerg,

More information

hp calculators HP 30S Statistics Averages and Standard Deviations Average and Standard Deviation Practice Finding Averages and Standard Deviations

hp calculators HP 30S Statistics Averages and Standard Deviations Average and Standard Deviation Practice Finding Averages and Standard Deviations HP 30S Statstcs Averages ad Stadard Devatos Average ad Stadard Devato Practce Fdg Averages ad Stadard Devatos HP 30S Statstcs Averages ad Stadard Devatos Average ad stadard devato The HP 30S provdes several

More information

Simulation Output Analysis

Simulation Output Analysis Smulato Output Aalyss Summary Examples Parameter Estmato Sample Mea ad Varace Pot ad Iterval Estmato ermatg ad o-ermatg Smulato Mea Square Errors Example: Sgle Server Queueg System x(t) S 4 S 4 S 3 S 5

More information

Data Fitting Report. Jurriaan Heuberger, Wouter Pasman. 8/4/2005 BRDF Fitting of Paintings TU Delft

Data Fitting Report. Jurriaan Heuberger, Wouter Pasman. 8/4/2005 BRDF Fitting of Paintings TU Delft Data Fttg Report Jurraa Heuberger, Wouter Pama 8/4/2005 BRDF Fttg of Patg TU Delft Itroucto... 1 The ata fttg problem... 1 The Leveberg-Marquart Algorthm... 1 Other poblte... 2 The ata et... 3 Geerate

More information

PTAS for Bin-Packing

PTAS for Bin-Packing CS 663: Patter Matchg Algorthms Scrbe: Che Jag /9/00. Itroducto PTAS for B-Packg The B-Packg problem s NP-hard. If we use approxmato algorthms, the B-Packg problem could be solved polyomal tme. For example,

More information

TESTS BASED ON MAXIMUM LIKELIHOOD

TESTS BASED ON MAXIMUM LIKELIHOOD ESE 5 Toy E. Smth. The Basc Example. TESTS BASED ON MAXIMUM LIKELIHOOD To llustrate the propertes of maxmum lkelhood estmates ad tests, we cosder the smplest possble case of estmatg the mea of the ormal

More information

A Remark on the Uniform Convergence of Some Sequences of Functions

A Remark on the Uniform Convergence of Some Sequences of Functions Advaces Pure Mathematcs 05 5 57-533 Publshed Ole July 05 ScRes. http://www.scrp.org/joural/apm http://dx.do.org/0.436/apm.05.59048 A Remark o the Uform Covergece of Some Sequeces of Fuctos Guy Degla Isttut

More information

Discrete Mathematics and Probability Theory Fall 2016 Seshia and Walrand DIS 10b

Discrete Mathematics and Probability Theory Fall 2016 Seshia and Walrand DIS 10b CS 70 Dscrete Mathematcs ad Probablty Theory Fall 206 Sesha ad Walrad DIS 0b. Wll I Get My Package? Seaky delvery guy of some compay s out delverg packages to customers. Not oly does he had a radom package

More information

Predicting the eruption time after observed an eruption of 4 minutes in duration.

Predicting the eruption time after observed an eruption of 4 minutes in duration. Lear Regreo ad Correlato 00 Predctg the erupto tme after oberved a erupto of 4 mute durato. 90 80 70 Iter-erupto Tme.5.0.5 3.0 3.5 4.0 4.5 5.0 5.5 Durato A ample of tererupto tme wa take durg Augut -8,

More information

Arithmetic Mean and Geometric Mean

Arithmetic Mean and Geometric Mean Acta Mathematca Ntresa Vol, No, p 43 48 ISSN 453-6083 Arthmetc Mea ad Geometrc Mea Mare Varga a * Peter Mchalča b a Departmet of Mathematcs, Faculty of Natural Sceces, Costate the Phlosopher Uversty Ntra,

More information

MA/CSSE 473 Day 27. Dynamic programming

MA/CSSE 473 Day 27. Dynamic programming MA/CSSE 473 Day 7 Dyamc Programmg Bomal Coeffcets Warshall's algorthm (Optmal BSTs) Studet questos? Dyamc programmg Used for problems wth recursve solutos ad overlappg subproblems Typcally, we save (memoze)

More information

Scheduling Jobs with a Common Due Date via Cooperative Game Theory

Scheduling Jobs with a Common Due Date via Cooperative Game Theory Amerca Joural of Operato Reearch, 203, 3, 439-443 http://dx.do.org/0.4236/ajor.203.35042 Publhed Ole eptember 203 (http://www.crp.org/joural/ajor) chedulg Job wth a Commo Due Date va Cooperatve Game Theory

More information

Maximum Likelihood Estimation

Maximum Likelihood Estimation Marquette Uverst Maxmum Lkelhood Estmato Dael B. Rowe, Ph.D. Professor Departmet of Mathematcs, Statstcs, ad Computer Scece Coprght 08 b Marquette Uverst Maxmum Lkelhood Estmato We have bee sag that ~

More information

CHAPTER 3 POSTERIOR DISTRIBUTIONS

CHAPTER 3 POSTERIOR DISTRIBUTIONS CHAPTER 3 POSTERIOR DISTRIBUTIONS If scece caot measure the degree of probablt volved, so much the worse for scece. The practcal ma wll stck to hs apprecatve methods utl t does, or wll accept the results

More information

best estimate (mean) for X uncertainty or error in the measurement (systematic, random or statistical) best

best estimate (mean) for X uncertainty or error in the measurement (systematic, random or statistical) best Error Aalyss Preamble Wheever a measuremet s made, the result followg from that measuremet s always subject to ucertaty The ucertaty ca be reduced by makg several measuremets of the same quatty or by mprovg

More information

ρ < 1 be five real numbers. The

ρ < 1 be five real numbers. The Lecture o BST 63: Statstcal Theory I Ku Zhag, /0/006 Revew for the prevous lecture Deftos: covarace, correlato Examples: How to calculate covarace ad correlato Theorems: propertes of correlato ad covarace

More information

ON THE CHROMATIC NUMBER OF GENERALIZED STABLE KNESER GRAPHS

ON THE CHROMATIC NUMBER OF GENERALIZED STABLE KNESER GRAPHS ON THE CHROMATIC NUMBER OF GENERALIZED STABLE KNESER GRAPHS JAKOB JONSSON Abstract. For each teger trple (, k, s) such that k 2, s 2, a ks, efe a graph the followg maer. The vertex set cossts of all k-subsets

More information

Overview of the weighting constants and the points where we evaluate the function for The Gaussian quadrature Project two

Overview of the weighting constants and the points where we evaluate the function for The Gaussian quadrature Project two Overvew of the weghtg costats ad the pots where we evaluate the fucto for The Gaussa quadrature Project two By Ashraf Marzouk ChE 505 Fall 005 Departmet of Mechacal Egeerg Uversty of Teessee Koxvlle, TN

More information

Summarizing Bivariate Data. Correlation. Scatter Plot. Pearson s Sample Correlation. Summarizing Bivariate Data SBD - 1

Summarizing Bivariate Data. Correlation. Scatter Plot. Pearson s Sample Correlation. Summarizing Bivariate Data SBD - 1 Summarzg Bvarate Data Summarzg Bvarate Data - Eamg relato betwee two quattatve varable I there relato betwee umber of hadgu regtered the area ad umber of people klled? Ct NGR ) Nkll ) 447 3 4 3 48 4 4

More information