Supplementary Materials: Proofs and Technical Details for Parsimonious Tensor Response Regression Lexin Li and Xin Zhang

Size: px
Start display at page:

Download "Supplementary Materials: Proofs and Technical Details for Parsimonious Tensor Response Regression Lexin Li and Xin Zhang"

Transcription

1 Suppleentary Materials: Proofs and Tecnical Details for Parsionious Tensor Response Regression Lexin Li and Xin Zang A Soe preliinary results We will apply te following two results repeatedly. For a positive definite atrix B, In addition (Kolda, 2006; Proposition 3.7), B = arg in A>0 {log A + tr(a B)}. A = B C 2 N C N A (n) = C n B (n) (C N C n+ C n C ) T. For te envelope paraeterization of B = Θ; Γ,..., Γ, I p, we ave B (+) = Θ (+) (Γ T Γ T ), and Θ (+) = B (+) (Γ Γ ). As for Y = Z; Γ,..., Γ R r r, we ave te following results by treating Y as (+)- t order tensor wit r + =, and treating vec T (Y) as te ode-( + ) atricization. vec T (Z) = vec T (Y Γ T 2 Γ T ) = vec T (Y)(Γ Γ ), vec(z) = (Γ T Γ T )vec(y). B Proof for Proposition and 2 We first prove Proposition. Ten Proposition 2 follows directly fro te results of Proposition and te definitions of reducing subspace and Tucer decoposition.

2 Under te tensor linear odel Y = B (+) X + ɛ, we can write Y Q = (B (+) X) Q + ɛ Q = (B Q ) (+) X + ɛ Q, wic iplies tat Y Q X Y Q is equivalent to B Q = 0. Siilarly, Y Q Y P X is equivalent to ɛ Q ɛ P, were te independence of two tensor-valued rando variable is defined as independence of teir vectorized fors: vec(ɛ Q ) vec(ɛ P ). Because of te tensor noral distribution, te independence is equivalent to cov{vec(ɛ Q ), vec(ɛ P )} = 0, and ence to Σ Σ + Q Σ P Σ Σ = 0. Terefore, we ave sown tat Y Q Y P X Q Σ P = 0 Σ = P Σ P + Q Σ Q. C Proof for Proposition 3 We first sow tat for subspaces E = E E and covariances Σ = Σ Σ, were E R r and Σ R r r, =,...,, te two conditions are equivalent: () E reduces Σ; (2) E reduces Σ for all =,...,. Te stateent (2) iplies () is straigtforward. We need to sow tat (), E = E E reduces Σ = Σ Σ, iplies (2), E reduces Σ for all =,...,. By definition, () iplies tat P E E (Σ Σ )Q E E = 0, were Q E E = I P E E = I P E P E. Expanding tis, we see tat P E Σ P E Σ P E Σ P E P E Σ P E = 0. If we rigt-ultiply P E P E2 I r to bot sides of te above equation, we obtain P E Σ P E P E2 Σ 2 P E2 (P E Σ P E Σ P E ) = 0, 2

3 wic iplies tat P E Σ P E Σ P E = 0. Since P E Σ = P E Σ (P E + Q E ), we conclude tat P E Σ Q E = 0 and ence E reduces Σ. Using te siilar arguent, we can see tat E reduces Σ, for all =,...,. We next sow tat E = E E contains span(b T ) iplies E (+) contains span(b () ). Let G R r q be sei-ortogonal atrix suc tat span(g ) = span(b () ), ten we ave Tucer decoposition of B as B = η; G,..., G, I p for soe η R q q p. Terefore we can write B T = (G (+) G )η T. Hence E = E (+) E contains span(b T ) (+) iplies tat E contains span(g ) = span(b () ). D Proof for Lea Here we ai to sow tat Σ (0), te Kronecer covariance estiator fro Manceur and Dutilleul (203), is a n-consistent estiator for Σ, for all, if ε i, i =,..., n, are i.i.d. fro tensor noral distribution wit ean 0 and covariances Σ,..., Σ. First we note tat te ode- atricization ε i() R r ( j r j ), i =,..., n, are i.i.d. fro a atrix noral distribution wit ean 0 and covariances Σ (te row covariance atrix) and Σ Σ + Σ Σ (te colun covariance atrix). Ten following Gupta and Nagar (999; Capter 2, Teore 2.3.5), we ave te following second-order oents: E(ε i() ε T i() ) = Σ tr( ), were tr( ) is te trace operator of atrices. In te iterative algorit of obtaining Σ (0), te starting value for Σ(0) is n j r j n e i() e T i(), wic is tus n-consistent for Σ up to a scalar difference due to tr( ). As we discussed in Section 4. regarding te identifiability of Σ s, te scalar difference can be resolved after noralizing Σ (0) = τσ (0) Σ (0) according to te scalar τ = (n j r j ) n vec(e i ){(Σ (0) ) (Σ (0) ) }vec T (e i ) at te end of eac iteration. 3

4 Terefore, after te first iteration, we ave obtained n-consistent estiators Σ (0) for Σ. In te iterations tat follow, te updating equations of Σ (0) for =,...,, are obtained by axiizing te tensor noral lieliood function. Hence it is guaranteed tat te final estiators of Σ (0) fro te algorit are also n-consistent for Σ. E Proof for Teore E. Consistency of te one-step estiator Fro Coo and Zang (206; Proposition 6) we now tat if M and Û are n-consistent estiators for M > 0 and U 0, ten te D algorit for iniizing J n (G) = log G T MG + log G T ( M + Û) G produce n-consistent estiator for te projection onto te envelope E M (U). We use tis result to prove te n-consistency of P os Γ, noting tat f (0) and Algorit 2 are special instances of J n and te D algorit. First Σ (0), te Kronecer covariance estiator fro Manceur and Dutilleul (203), is n- consistent estiator for Σ. Next, by coparing f (0) (G ) to J n (G) we see tat N (0) = (n j r j ) n Y i() Σ (0) YT i() is analogous to ( M + Û) in J n(g), were we ave defined Σ (0) ((Σ(0) ) (Σ (0) + ) (Σ (0) ) (Σ (0) ) ). Terefore we focus on N (0) Σ (0), wic is analogous to Û in J n(g). Recall tat Y i = B (0) (+) X T i +e i and tat Σ (0) is obtained based on e i as Σ (0) = (n j r j ) n e i() Σ (0) et i(). Hence, N (0) Σ (0) = (n j r j ) n (Y i e i ) () Σ (0) (Y i e i ) T (), were we recognize Y i e i = B (0) (+) X T i. Define te following scaled regression coefficient tensor B {} = B; Σ /2,..., Σ /2, I r, Σ /2 +,..., Σ /2, Σ /2 X. Ten we see tat N (0) Σ (0) = ( j r j ) B {} ( B {} ) T for B {} = B; (Σ (0) ) /2,..., (Σ (0) ) /2, I r, (Σ (0) + ) /2,..., (Σ (0) ) /2 /2, Σ X, 4

5 were Σ X = n n X i X T i. We clai tat N(0) Σ (0) is a n-consistent estiator for B {} () (B{} () )T up to an upfront scaling constant, ( j r j ), based on te fact tat te saple atrices Σ (0) j and Σ X are asyptotically independent of te OLS estiator B (0). By Coo and Zang (206; Proposition 6) we now see tat P os Γ is a n-consistent estiator for te projection onto te envelope E Σ (B {} () (B{} () )T ) = E Σ (B {} ). By definition of () B{} and te property of Tucer operator, we ave B {} = B () () (Σ /2 Σ /2 Σ /2 + Σ /2 Σ /2 X ), wic iplies span(b {} () ) = span(b ()), and ence E Σ (B {} () ) = E Σ (B () ). So far we ave sown tat P os Γ is a n-consistent estiator for te projection onto te envelope E Σ (B () ). Te second part of te proposition is based on te n-consistency of B (0) = B OLS and P os Γ, =,...,, and te definition of B os : B os = B (0) ; P os Γ,..., P os Γ, I p. E.2 Consistency of te lieliood-based estiator Te n-consistency of te lieliood-based estiator fro iniizing l(b, Σ) relies on Sapiro s (986) results on te asyptotics of over-paraeterized structural odels. Te proof is parallel to te proof of Proposition 4 in Coo and Zang (205) and is tus oitted. F Proof for Teore 2 (including derivations for te updating equations in Section 4.) Following te discussion in Section 5.2 of te paper, te iterative estiator is not guaranteed to be necessarily te axiu lieliood estiator (MLE), due to te existence of ultiple local inia. However, it is asyptotically equivalent to te MLE. Tis is because, under te tensor noral distribution, te initialization of Algorit is built upon n-consistent estiators, wile eac paraeter in Algorit is iteratively obtained along te partial derivative of 5

6 te log-lieliood. Fro te classical teory of point estiation, we now tat one Newton- Rapson step fro te starting value provides an estiator tat is asyptotically equivalent to te MLE even in te presence of ultiple local inia (Leann and Casella, 998, p. 454). Consequently, to prove Teore 2, we only focus on te asyptotic properties of te teoretical MLE. Specifically, we need to derive all te objective functions and updating equations in Section 4. fro te negative noral log-lieliood function: l(b, Σ) = log Σ + n n {vec(y i ) B T (+) X i}σ {vec(y i ) B T (+) X i}, wic is to be optiized over Σ = Σ({Γ, Ω, Ω 0 } = ) and B = B({Γ } =, Θ) for seiortogonal Γ R r u, positive definite and syetric Ω S u and Ω0 S r u, and Θ R u u p. Since it is ipossible to obtain explicit fors of MLEs, we sow tat te series of equations used in Algorit coe fro partially iniizing l wit specified paraeters fixed. We suarize our findings in te following stateents and give detailed derivations iediately after. Note tat te updating equations in Section 4. is obtained fro te following equations by superscripting (t) on te left and sides and superscripting (t+) on te rigt and sides of equations. Under te tensor noral assuption, te MLEs satisfy te following equations, Θ = Z (+) {(XX T ) X}, B = Y P Γ 2 P Γ (+) {(XX T ) X}, = B OLS ; P Γ,, P Γ, I p Ω = n s i() {Ω Ω + Ω Ω }s T i() n j r j Ω 0 = n n j r j Γ T 0Y i() {Σ + Σ Σ }Y T i() Γ 0, were te data tensors Z i and Z are defined according to Z = Y; Γ T,..., Γ T, and te residual tensor s i = Z i Θ (+) X i. Under te tensor noral assuption, te MLE of {Γ } = can be 6

7 obtained as iniizer of te following objective function were M, N are defined as M = (n j N = (n j f (Γ ) = log Γ T M Γ + log Γ T N Γ, r j ) n δ i() (Σ + Σ Σ )δ T i(), r j ) n Y i() (Σ + Σ Σ )Y T i(), were δ i() is te -t atricization of te residual δ i, δ i = Y i Y P Γ 2 ( ) P Γ (+) P Γ+ +2 P Γ (+) X T i{(xx T ) X} = Y i B OLS ; P Γ,..., P Γ, I r, P Γ+,..., P Γ, I p (+) X i. F. Estiation of oter paraeters given {Γ } = We first decopose te log Σ ter as log Σ = log Σ Σ = = {( r ) log Σ } = j = {( r )(log Ω + log Ω 0 )}, wic is essentially 2 additive ters of log Ω and log Ω 0. Te conditional log-lieliood can be separated into two independent parts regarding P(Y) X and Q(Y) Q(Y) X. Studying te regression of P(Y) on X is essentially studying tat of Z = Y; Γ T,..., Γ T on X. Following te discussion in te paper, we ave te MLEs for Θ, B and {Ω } = given {Γ } =. We next derive te MLE equations for {Ω 0 } = wit given {Γ } =. Witout loss of generality, we write down our derivations wit respect to Ω 0. We decopose Σ = + 0, according to Ω and Ω 0 in te decoposition of Σ = Γ Ω Γ T + Γ 0 Ω 0Γ T 0. = Σ 2 (Γ Ω Γ T ) = (Γ Ω Γ T + Γ 0 Ω 0Γ T 0) (Γ 2 Ω 2 Γ T 2 + Γ 02 Ω 02Γ T 02) (Γ Ω Γ T ), 0 = Σ 2 (Γ 0 Ω 0Γ T 0) = (Γ Ω Γ T + Γ 0 Ω 0Γ T 0) (Γ 2 Ω 2 Γ T 2 + Γ 02 Ω 02Γ T 02) (Γ 0 Ω 0Γ T 0). j 7

8 Hence, we can write te negative partial log-lieliood for solving Ω 0 as l(ω 0 {Γ } = ) ( r j ) log Ω 0 + n n {vec(y i ) B T (+) X i} T 0 {vec(y i ) B T (+) X i}. j> Recall tat B (+) = η (+) (Γ T Γ T ), ence 0 B T (+) = 0 as a result of Γ 0Ω 0Γ T 0 Γ = 0. Terefore, l(ω 0 {Γ } = ) ( r j ) log Ω 0 + n n vec T (Y i ) 0 vec(y i ). j> Te quadratic for vec T (Y i ) 0 vec(y i ) equals te squared nor of vec(y i Ω /2 0 Γ T 0 2 Σ /2 2 Σ /2 ) vec(v i ). By definition of tensor nor, V i 2 = vec(v i ) 2 = V i() 2 F = tr(v T i() V i()), were V i() is te ode- atricization of V i : Hence V i() = (Ω /2 0 Γ T 0)Y i() (Σ /2 /2 2 ). tr(v T i() V i()) = tr(v i() V T i() ) = tr {Ω 0Γ T 0Y i() (Σ 2 ) Y T i() Γ 0}. Te partial conditional log-lieliood becoes l(ω 0 {Γ } = ) ( r j ) log Ω 0 + n n tr(v T i() V i()) j> = ( j> r j ) log Ω 0 + n tr [Ω 0 n wic lead to te following equations for iteratively solving Ω 0. Ω 0 = (n j> {Γ T 0Y i() (Σ 2 ) Y T i() Γ 0}], r j ) n Γ T 0Y i() (Σ 2 )Y T i() Γ 0, were Σ = Γ Ω Γ T + Γ 0 Ω 0 Γ T 0 for =,...,. It is ten easy to obtain te following result, for any, Ω 0 = (n j r j ) n Γ T 0Y i() (Σ + Σ Σ )Y T i() Γ 0. 8

9 F.2 Estiation of Γ given {Γ, Ω, Ω 0 } =2 Treating te oter paraeters {Γ, Ω, Ω 0 } =2 as fixed constants, we write B = B(Γ ) and Σ = Σ(Γ, Ω (Γ ), Ω 0 (Γ )) as functions of Γ. We ten plug te into te lieliood l(b, Σ) to partially optiize over Ω and Ω 0 analytically, and ten te objective function for optiizing over Γ R r u as follows. First, ignoring all te fixed constants, te log-deterinant ter in l(b, Σ) becoes log Σ ( r j ){log Ω (Γ ) + log Ω 0 (Γ ) }. j=2 Siilar to te previous section, we can decopose Σ = Σ Σ into two parts according to te decoposition Σ = Γ Ω Γ T + Γ 0 Ω 0Γ T 0. Ten we can write = n {vec(y i ) B T (+) X i}σ {vec(y i ) B T (+) X i} n tr {Ω 0Γ T 0Y i() (Σ 2 )Y T i() Γ 0} + n tr {Ω Γ T e i() (Σ 2 )e T i() Γ }, were e i e i (Γ ) = Y i B(Γ ) (+) X T i and B(Γ ) = Y P Γ 2 P Γ (+) {(XX T ) X}. Ten, Γ T e i() = (Y i Γ T Y Γ T P Γ 2 P Γ (+) X T i{(xx T ) X}) () = (Y i Γ T Y Γ T 2 P Γ (+) X T i{(xx T ) X}) () = {(Y i Y I r 2 P Γ (+) X T i{(xx T ) X}) Γ T } () = Γ T δ i(), were δ i = (Y i Y I r 2 P Γ (+) X T i {(XXT ) X}) does not involve Γ. Te partially axiized negative log-lieliood now becoes 9

10 l(b, Σ) = log Σ + n n {vec(y i ) B T (+) X i}σ {vec(y i ) B T (+) X i} ( r j ){log Ω (Γ ) + log Ω 0 (Γ ) } j=2 + n n tr {Ω 0Γ T 0Y i() (Σ 2 )Y T i() Γ 0} + n n tr {Ω Γ T δ i() (Σ 2 )δ T i()γ }, wic leads to partial MLE of Ω (Γ ) and Ω 0 (Γ ) as Ω (Γ ) = (n j=2 Ω 0 (Γ ) = (n j=2 r j ) n Γ T δ i() (Σ 2 )δ T i()γ r j ) n Γ T 0Y i() (Σ 2 )Y T i() Γ 0. Ten, substitute tese bac to l(b, Σ) to get te lieliood-based objective function for Γ : F n (Γ ) = log (n j=2 + log (n j=2 r j ) n Γ T δ i() (Σ 2 )δ T i()γ r j ) n Γ T 0Y i() (Σ 2 )Y T i() Γ 0 log Γ T M Γ + log Γ T N Γ, were M = (n j=2 r j ) n δ i() (Σ Σ 2 )δ T i() and N = (n j=2 r j ) n Y i() (Σ Σ 2 )Y T i(). Te last step of te above equations coe fro te fact tat log ΓT 0AΓ 0 log Γ A Γ for any positive definite syetric atrix A. G Proof for Teore 3 Fro te proof of Teore 2, we ave seen tat B Γ = B OLS ; P Γ,..., P Γ, I p, were P Γ, =,...,, are all true projections onto te envelopes (i.e. population values). Ten vec( B Γ ) = (I p P Γ P Γ )vec( B OLS ) = P t vec( B OLS ). Te OLS estiator vec( B OLS ) is n-consistent and asyptotically noral wit ean zero covariance equals to U OLS = 0

11 Σ X Σ. Since P t = I p P Γ P Γ is fixed as population trut, we ave proven tat te envelope estiator B Γ is n-consistent and asyptotically noral wit ean zero covariance U Γ = P t U OLS P t = Σ X P Γ Σ P Γ P Γ Σ P Γ. H Proof for Teore 4 Recall tat te paraeter vectors involved in tis Teore are: = ( 2 ) = ( vec(b) vec(σ) ), φ = φ φ 2 φ + = vec(b) vec(σ ) vec(σ ), ξ = ξ ξ 3+ were ξ = vec(θ), {ξ j } + j=2 = {vec(γ )} =, {ξ j} 2+ j=+2 = {vec(ω )} =, {ξ j} 3+ j=2+2 = {vec(ω 0 )} =. Te lengt of vectors are onotonically decreasing fro to φ and ten to ξ, because tey corresponding to tree nested odel assuptions: (M) corresponding to te unrestricted, vectorized linear regression odel, or te OLS odel; (M2) φ is fro te Kronecer covariance assuption, so letting = (φ) is essentially iposing te Kronecer atrix structure tat Σ = Σ Σ ; (M3) ξ is based on envelope assuptions of E Σ (B () ), =,...,. Fro Teore 2, we now tat ( ) = (φ ( ) ) = (ξ ( ) ) is te MLE under te envelope assuption, i.e. odel assuption (M3). Also, (0) = (φ (0) ) contains te OLS estiator and te Kronecer covariance estiator fro Manceur and Dutilleul (203). Hence it is te MLE under tensor noral assuption, i.e. odel assuption (M2). Since = (φ) = (ξ) is overparaeterized, fro Sapiro s (986) we see te following results: n( (0) True ) N(0, V 0 ) and n( ( ) True ) N(0, V ), were V 0 = H(H T J H) H T, V = K(K T J K) K T, and te gradient atrices H = (φ)/ φ and K = (ξ)/ ξ. Moreover, te Fiser inforation atrix J as te sae for as in te usual vector-response linear regression odel, J = ( Σ X Σ ET d (Σ Σ )E d ),

12 were te expansion atrix E d R d2 d(d+)/2 as te corresponding diension d = = r. Ten, H(H T J H) H T = J /2 J /2 H(HT J /2 J/2 H) H T J /2 J /2 = J /2 and siilarly, K(K T J K) K T = J /2 P /2 J KJ /2. By cain rule, we can write K = (ξ)/ ξ = (φ)/ φ φ(ξ)/ ξ = H φ(ξ)/ ξ. P /2 J HJ /2, Terefore span(k) span(h) and span(j /2 K) span(j/2 H), wic iplies tat P J /2 P /2 J HP = P J /2 K J /2 KP J /2 H. Finally, we ave arrived at V 0 V = H(H T J H) H T K(K T J K) K T = J /2 P /2 J HJ /2 J /2 P /2 J KJ /2 = J /2 (P P /2 J H J /2 = J /2 (P P /2 J H J /2 HP J /2 = J /2 P /2 J HQ J /2 KJ /2 0. K) J /2 J /2 K) K = So far, we ave proved te ain part of te Proposition 6. We next provide details about te gradient atrices H and K. First of all, since = φ = vec(b), we ave H = ( I p = r 0 0 vec(σ) 0 vec(σ ) vec(σ) vec(σ ) ). (H) For a syetric atrix A R a a, vec(a) = C a vec(a) and vec(a) = E a vec(a), were C a R a(a+)/2 a2 is te contraction atrix and E a R a2 a(a+)/2 is te extraction atrix. Terefore, vec(σ) vec(σ ) = C vec(σ) j= r j vec(σ ) E r, =,...,. We ten use te Kronecer structure Σ = Σ Σ to calculate vec(σ) vec(σ ). For = and =, we can use te forulas fro Facler (2005) for te derivatives of vec(a B) over vec(a) and over vec(b). For 2, tere is no copact way of writing down te atrix for but eleentwise derivatives, wic is straigtforward but non-trivial. 2

13 Next we write K as K = H φ(ξ) ξ = HR, R = ( ( vec(b) ξ ) T ( vec(σ ) ξ ) T ( vec(σ) ξ ) T ) T, (H2) were we calculate eac blocs of R in te following. Recall tat ξ = (ξ T,..., ξ T 3+) T and ξ = vec(θ), {ξ j } + j=2 = {vec(γ )} =, {ξ j} 2+ j=+2 = {vec(ω )} =, {ξ j} 3+ j=2+2 = {vec(ω 0)} =. Te first bloc in R is vec(b) ξ = ( vec(b) vec(θ) vec(b) vec(γ ) vec(b) vec(γ ) 0 0 ), (H3) were te zeros are because of B does not depend on Ω or Ω 0. We re-write vec(b) as vec(b) = vec(b T (+) ) = vec ((Γ Γ )Θ T (+) ) = (I p Γ Γ )vec(θ T (+)) = (I p Γ Γ )vec(θ). Tus, vec(b) vec(θ) = (I p Γ Γ ). We next introduce te notation of re-arranging te vectorizations of a ode-n tensor T R d d N : squared constant atrixπ T n satisfies: vec(t) = Π T n vec(t (n) ). Terefore, vec(b) = Π B vec(b () ) and ence vec(b) = Π B vec (Γ Θ () (Γ T Γ T + Γ T Γ T )) = Π B ((Γ Γ + Γ Γ )Θ T () I r ) vec(γ ), (H4) vec(b) vec(γ ) = ΠB ((Γ Γ + Γ Γ )Θ T () I r ). Finally for Σ = Γ Ω Γ T + Γ 0 Ω 0 Γ T 0, we ave (H5) vec(σ ) ξ = ( 0 0 vec(σ ) vec(γ ) 0 0 vec(σ ) vec(ω ) 0 0 vec(σ ) vec(ω 0 ) 0 0 ), (H6) 3

14 were te tree nonzero eleents are, (analogous to Coo et al. (200)), vec(σ ) vec(γ ) vec(σ ) vec(ω ) vec(σ ) vec(ω 0 ) = 2C r (Γ Ω I r Γ r Γ 0 Ω 0 Γ T 0), = C r (Γ Γ )E u, = C r (Γ 0 Γ 0 )E r u. Finally te explicit gradient atrices H is obtained by plugging eac blocs (H3) (H6) into (H); siilarly, te explicit gradient atrices K is obtained by plugging eac blocs (H3) (H6) into (H2). We ave tus copleted te proof of tis teore. Additional References Coo, R. D., Li, B. and Ciaroonte, F. (200), Envelope odels for parsionious and efficient ultivariate linear regression, Statistica Sinica, 20(3), Coo, R. D. and Zang, X. (205), Siultaneous envelope for ultivariate linear regression, Tecnoetrics, 57(), 25. Coo, R. D. and Zang, X. (206), Algorits for envelope estiation, Journal of Coputational and Grapical Statistics, In press. Facler, P. L. (2005). Notes on atrix calculus. ttp://www4.ncsu.edu/~pfacler/ MatCalc.pdf, Tecnical Report. Gupta, A. and Nagar, D. (999), Matrix variate distributions, CRC Press. Kolda, T. G. (2006). Multilinear operators for iger-order decopositions. United States, Departent of Energy, Tecnical Report. Leann, E. L. and Casella, G. (998). Business Media. Teory of point estiation, Springer Science & 4

15 Manceur, A. M. and Dutilleul, P. (203), Maxiu lieliood estiation for te tensor noral distribution: Algorit, iniu saple size, and epirical bias and dispersion, Journal of Coputational and Applied Mateatics, 239, Sapiro, A. (986), Asyptotic teory of overparaeterized structural odels, Journal of te Aerican Statistical Association, 8,

Block designs and statistics

Block designs and statistics Bloc designs and statistics Notes for Math 447 May 3, 2011 The ain paraeters of a bloc design are nuber of varieties v, bloc size, nuber of blocs b. A design is built on a set of v eleents. Each eleent

More information

Supplementary Material for Fast and Provable Algorithms for Spectrally Sparse Signal Reconstruction via Low-Rank Hankel Matrix Completion

Supplementary Material for Fast and Provable Algorithms for Spectrally Sparse Signal Reconstruction via Low-Rank Hankel Matrix Completion Suppleentary Material for Fast and Provable Algoriths for Spectrally Sparse Signal Reconstruction via Low-Ran Hanel Matrix Copletion Jian-Feng Cai Tianing Wang Ke Wei March 1, 017 Abstract We establish

More information

lecture 35: Linear Multistep Mehods: Truncation Error

lecture 35: Linear Multistep Mehods: Truncation Error 88 lecture 5: Linear Multistep Meods: Truncation Error 5.5 Linear ultistep etods One-step etods construct an approxiate solution x k+ x(t k+ ) using only one previous approxiation, x k. Tis approac enoys

More information

Supplementary to Learning Discriminative Bayesian Networks from High-dimensional Continuous Neuroimaging Data

Supplementary to Learning Discriminative Bayesian Networks from High-dimensional Continuous Neuroimaging Data Suppleentary to Learning Discriinative Bayesian Networks fro High-diensional Continuous Neuroiaging Data Luping Zhou, Lei Wang, Lingqiao Liu, Philip Ogunbona, and Dinggang Shen Proposition. Given a sparse

More information

Derivative at a point

Derivative at a point Roberto s Notes on Differential Calculus Capter : Definition of derivative Section Derivative at a point Wat you need to know already: Te concept of liit and basic etods for coputing liits. Wat you can

More information

5.1 The derivative or the gradient of a curve. Definition and finding the gradient from first principles

5.1 The derivative or the gradient of a curve. Definition and finding the gradient from first principles Capter 5: Dierentiation In tis capter, we will study: 51 e derivative or te gradient o a curve Deinition and inding te gradient ro irst principles 5 Forulas or derivatives 5 e equation o te tangent line

More information

CS Lecture 13. More Maximum Likelihood

CS Lecture 13. More Maximum Likelihood CS 6347 Lecture 13 More Maxiu Likelihood Recap Last tie: Introduction to axiu likelihood estiation MLE for Bayesian networks Optial CPTs correspond to epirical counts Today: MLE for CRFs 2 Maxiu Likelihood

More information

Stationary Gaussian Markov processes as limits of stationary autoregressive time series

Stationary Gaussian Markov processes as limits of stationary autoregressive time series Stationary Gaussian Markov processes as liits of stationary autoregressive tie series Pilip A. rnst 1,, Lawrence D. Brown 2,, Larry Sepp 3,, Robert L. Wolpert 4, Abstract We consider te class, C p, of

More information

Neural Networks Trained with the EEM Algorithm: Tuning the Smoothing Parameter

Neural Networks Trained with the EEM Algorithm: Tuning the Smoothing Parameter eural etworks Trained wit te EEM Algorit: Tuning te Sooting Paraeter JORGE M. SATOS,2, JOAQUIM MARQUES DE SÁ AD LUÍS A. ALEXADRE 3 Intituto de Engenaria Bioédica, Porto, Portugal 2 Instituto Superior de

More information

A KERNEL APPROACH TO ESTIMATING THE DENSITY OF A CONDITIONAL EXPECTATION. Samuel G. Steckley Shane G. Henderson

A KERNEL APPROACH TO ESTIMATING THE DENSITY OF A CONDITIONAL EXPECTATION. Samuel G. Steckley Shane G. Henderson Proceedings of te 3 Winter Siulation Conference S Cick P J Sáncez D Ferrin and D J Morrice eds A KERNEL APPROACH TO ESTIMATING THE DENSITY OF A CONDITIONAL EXPECTATION Sauel G Steckley Sane G Henderson

More information

Estimation for the Parameters of the Exponentiated Exponential Distribution Using a Median Ranked Set Sampling

Estimation for the Parameters of the Exponentiated Exponential Distribution Using a Median Ranked Set Sampling Journal of Modern Applied Statistical Metods Volue 14 Issue 1 Article 19 5-1-015 Estiation for te Paraeters of te Exponentiated Exponential Distribution Using a Median Ranked Set Sapling Monjed H. Sau

More information

A NOTE ON BOOLEAN LATTICES AND FAREY SEQUENCES. Andrey O. Matveev Data-Center Co., RU , Ekaterinburg, P.O. Box 5, Russian Federation

A NOTE ON BOOLEAN LATTICES AND FAREY SEQUENCES. Andrey O. Matveev Data-Center Co., RU , Ekaterinburg, P.O. Box 5, Russian Federation INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 7 27 #A2 A NOTE ON BOOLEAN LATTICES AND FAREY SEQUENCES Anrey O Matveev Data-Center Co RU-6234 Eaterinburg PO Box 5 Russian Feeration aoatveev@cru

More information

c hc h c h. Chapter Since E n L 2 in Eq. 39-4, we see that if L is doubled, then E 1 becomes (2.6 ev)(2) 2 = 0.65 ev.

c hc h c h. Chapter Since E n L 2 in Eq. 39-4, we see that if L is doubled, then E 1 becomes (2.6 ev)(2) 2 = 0.65 ev. Capter 39 Since n L in q 39-4, we see tat if L is doubled, ten becoes (6 ev)() = 065 ev We first note tat since = 666 0 34 J s and c = 998 0 8 /s, 34 8 c6 66 0 J sc 998 0 / s c 40eV n 9 9 60 0 J / ev 0

More information

1 Proving the Fundamental Theorem of Statistical Learning

1 Proving the Fundamental Theorem of Statistical Learning THEORETICAL MACHINE LEARNING COS 5 LECTURE #7 APRIL 5, 6 LECTURER: ELAD HAZAN NAME: FERMI MA ANDDANIEL SUO oving te Fundaental Teore of Statistical Learning In tis section, we prove te following: Teore.

More information

Estimating the Density of a Conditional Expectation

Estimating the Density of a Conditional Expectation Estiating te Density of a Conditional Expectation Sauel G. Steckley Sane G. Henderson David Ruppert Ran Yang Daniel W. Apley Jerey Stau Abstract In tis paper, we analyze etods for estiating te density

More information

Submanifold density estimation

Submanifold density estimation Subanifold density estiation Arkadas Ozakin Georgia Tec Researc Institute Georgia Insitute of Tecnology arkadas.ozakin@gtri.gatec.edu Alexander Gray College of Coputing Georgia Institute of Tecnology agray@cc.gatec.edu

More information

Support recovery in compressed sensing: An estimation theoretic approach

Support recovery in compressed sensing: An estimation theoretic approach Support recovery in copressed sensing: An estiation theoretic approach Ain Karbasi, Ali Horati, Soheil Mohajer, Martin Vetterli School of Coputer and Counication Sciences École Polytechnique Fédérale de

More information

Physics 215 Winter The Density Matrix

Physics 215 Winter The Density Matrix Physics 215 Winter 2018 The Density Matrix The quantu space of states is a Hilbert space H. Any state vector ψ H is a pure state. Since any linear cobination of eleents of H are also an eleent of H, it

More information

A KERNEL APPROACH TO ESTIMATING THE DENSITY OF A CONDITIONAL EXPECTATION. Samuel G. Steckley Shane G. Henderson

A KERNEL APPROACH TO ESTIMATING THE DENSITY OF A CONDITIONAL EXPECTATION. Samuel G. Steckley Shane G. Henderson Proceedings of te 3 Winter Siulation Conference S Cick P J Sáncez D Ferrin and D J Morrice eds A KERNEL APPROACH TO ESTIMATING THE DENSITY OF A CONDITIONAL EXPECTATION Sauel G Steckley Sane G Henderson

More information

Sharp Time Data Tradeoffs for Linear Inverse Problems

Sharp Time Data Tradeoffs for Linear Inverse Problems Sharp Tie Data Tradeoffs for Linear Inverse Probles Saet Oyak Benjain Recht Mahdi Soltanolkotabi January 016 Abstract In this paper we characterize sharp tie-data tradeoffs for optiization probles used

More information

The proofs of Theorem 1-3 are along the lines of Wied and Galeano (2013).

The proofs of Theorem 1-3 are along the lines of Wied and Galeano (2013). A Appendix: Proofs The proofs of Theore 1-3 are along the lines of Wied and Galeano (2013) Proof of Theore 1 Let D[d 1, d 2 ] be the space of càdlàg functions on the interval [d 1, d 2 ] equipped with

More information

Feature Extraction Techniques

Feature Extraction Techniques Feature Extraction Techniques Unsupervised Learning II Feature Extraction Unsupervised ethods can also be used to find features which can be useful for categorization. There are unsupervised ethods that

More information

The Distribution of the Covariance Matrix for a Subset of Elliptical Distributions with Extension to Two Kurtosis Parameters

The Distribution of the Covariance Matrix for a Subset of Elliptical Distributions with Extension to Two Kurtosis Parameters journal of ultivariate analysis 58, 96106 (1996) article no. 0041 The Distribution of the Covariance Matrix for a Subset of Elliptical Distributions with Extension to Two Kurtosis Paraeters H. S. Steyn

More information

Consistent Multiclass Algorithms for Complex Performance Measures. Supplementary Material

Consistent Multiclass Algorithms for Complex Performance Measures. Supplementary Material Consistent Multiclass Algoriths for Coplex Perforance Measures Suppleentary Material Notations. Let λ be the base easure over n given by the unifor rando variable (say U over n. Hence, for all easurable

More information

EN40: Dynamics and Vibrations. Midterm Examination Tuesday March

EN40: Dynamics and Vibrations. Midterm Examination Tuesday March EN4: Dynaics and ibrations Midter Exaination Tuesday Marc 4 14 Scool of Engineering Brown University NAME: General Instructions No collaboration of any kind is peritted on tis exaination. You ay bring

More information

Lecture 20 November 7, 2013

Lecture 20 November 7, 2013 CS 229r: Algoriths for Big Data Fall 2013 Prof. Jelani Nelson Lecture 20 Noveber 7, 2013 Scribe: Yun Willia Yu 1 Introduction Today we re going to go through the analysis of atrix copletion. First though,

More information

Parsimonious Tensor Response Regression

Parsimonious Tensor Response Regression Parsimonious Tensor Response Regression Lexin Li and Xin Zhang University of California at Bereley; and Florida State University Abstract Aiming at abundant scientific and engineering data with not only

More information

3.3 Variational Characterization of Singular Values

3.3 Variational Characterization of Singular Values 3.3. Variational Characterization of Singular Values 61 3.3 Variational Characterization of Singular Values Since the singular values are square roots of the eigenvalues of the Heritian atrices A A and

More information

Using EM To Estimate A Probablity Density With A Mixture Of Gaussians

Using EM To Estimate A Probablity Density With A Mixture Of Gaussians Using EM To Estiate A Probablity Density With A Mixture Of Gaussians Aaron A. D Souza adsouza@usc.edu Introduction The proble we are trying to address in this note is siple. Given a set of data points

More information

arxiv: v2 [stat.me] 28 Aug 2016

arxiv: v2 [stat.me] 28 Aug 2016 arxiv:509.04704v [stat.me] 8 Aug 06 Central liit teores for network driven sapling Xiao Li Scool of Mateatical Sciences Peking University Karl Roe Departent of Statistics University of Wisconsin-Madison

More information

An unbalanced Optimal Transport splitting scheme for general advection-reaction-diffusion problems

An unbalanced Optimal Transport splitting scheme for general advection-reaction-diffusion problems An unbalanced Optial Transport splitting scee for general advection-reaction-diffusion probles T.O. Gallouët, M. Laborde, L. Monsaingeon May 2, 27 Abstract In tis paper, we sow tat unbalanced optial transport

More information

A SHORT INTRODUCTION TO BANACH LATTICES AND

A SHORT INTRODUCTION TO BANACH LATTICES AND CHAPTER A SHORT INTRODUCTION TO BANACH LATTICES AND POSITIVE OPERATORS In tis capter we give a brief introduction to Banac lattices and positive operators. Most results of tis capter can be found, e.g.,

More information

Web Appendix for Joint Variable Selection for Fixed and Random Effects in Linear Mixed-Effects Models

Web Appendix for Joint Variable Selection for Fixed and Random Effects in Linear Mixed-Effects Models Web Appendix for Joint Variable Selection for Fixed and Rando Effects in Linear Mixed-Effects Models Howard D. Bondell, Arun Krishna, and Sujit K. Ghosh APPENDIX A A. Regularity Conditions Assue that the

More information

Testing equality of variances for multiple univariate normal populations

Testing equality of variances for multiple univariate normal populations University of Wollongong Research Online Centre for Statistical & Survey Methodology Working Paper Series Faculty of Engineering and Inforation Sciences 0 esting equality of variances for ultiple univariate

More information

Testing the lag length of vector autoregressive models: A power comparison between portmanteau and Lagrange multiplier tests

Testing the lag length of vector autoregressive models: A power comparison between portmanteau and Lagrange multiplier tests Working Papers 2017-03 Testing the lag length of vector autoregressive odels: A power coparison between portanteau and Lagrange ultiplier tests Raja Ben Hajria National Engineering School, University of

More information

arxiv: v2 [math.st] 11 Dec 2018

arxiv: v2 [math.st] 11 Dec 2018 esting for high-diensional network paraeters in auto-regressive odels arxiv:803659v [aths] Dec 08 Lili Zheng and Garvesh Raskutti Abstract High-diensional auto-regressive odels provide a natural way to

More information

232 Calculus and Structures

232 Calculus and Structures 3 Calculus and Structures CHAPTER 17 JUSTIFICATION OF THE AREA AND SLOPE METHODS FOR EVALUATING BEAMS Calculus and Structures 33 Copyrigt Capter 17 JUSTIFICATION OF THE AREA AND SLOPE METHODS 17.1 THE

More information

Least Squares Fitting of Data

Least Squares Fitting of Data Least Squares Fitting of Data David Eberly, Geoetric Tools, Redond WA 98052 https://www.geoetrictools.co/ This work is licensed under the Creative Coons Attribution 4.0 International License. To view a

More information

Tail Estimation of the Spectral Density under Fixed-Domain Asymptotics

Tail Estimation of the Spectral Density under Fixed-Domain Asymptotics Tail Estiation of the Spectral Density under Fixed-Doain Asyptotics Wei-Ying Wu, Chae Young Li and Yiin Xiao Wei-Ying Wu, Departent of Statistics & Probability Michigan State University, East Lansing,

More information

LAB #3: ELECTROSTATIC FIELD COMPUTATION

LAB #3: ELECTROSTATIC FIELD COMPUTATION ECE 306 Revised: 1-6-00 LAB #3: ELECTROSTATIC FIELD COMPUTATION Purpose During tis lab you will investigate te ways in wic te electrostatic field can be teoretically predicted. Bot analytic and nuerical

More information

which together show that the Lax-Milgram lemma can be applied. (c) We have the basic Galerkin orthogonality

which together show that the Lax-Milgram lemma can be applied. (c) We have the basic Galerkin orthogonality UPPSALA UNIVERSITY Departent of Inforation Technology Division of Scientific Coputing Solutions to exa in Finite eleent ethods II 14-6-5 Stefan Engblo, Daniel Elfverson Question 1 Note: a inus sign in

More information

Differentiation in higher dimensions

Differentiation in higher dimensions Capter 2 Differentiation in iger dimensions 2.1 Te Total Derivative Recall tat if f : R R is a 1-variable function, and a R, we say tat f is differentiable at x = a if and only if te ratio f(a+) f(a) tends

More information

In this chapter, we consider several graph-theoretic and probabilistic models

In this chapter, we consider several graph-theoretic and probabilistic models THREE ONE GRAPH-THEORETIC AND STATISTICAL MODELS 3.1 INTRODUCTION In this chapter, we consider several graph-theoretic and probabilistic odels for a social network, which we do under different assuptions

More information

Distributed Subgradient Methods for Multi-agent Optimization

Distributed Subgradient Methods for Multi-agent Optimization 1 Distributed Subgradient Methods for Multi-agent Optiization Angelia Nedić and Asuan Ozdaglar October 29, 2007 Abstract We study a distributed coputation odel for optiizing a su of convex objective functions

More information

arxiv: v1 [stat.me] 30 Jan 2015

arxiv: v1 [stat.me] 30 Jan 2015 Parsimonious Tensor Response Regression Lexin Li and Xin Zhang arxiv:1501.07815v1 [stat.me] 30 Jan 2015 University of California, Bereley; and Florida State University Abstract Aiming at abundant scientific

More information

arxiv: v1 [math.na] 10 Oct 2016

arxiv: v1 [math.na] 10 Oct 2016 GREEDY GAUSS-NEWTON ALGORITHM FOR FINDING SPARSE SOLUTIONS TO NONLINEAR UNDERDETERMINED SYSTEMS OF EQUATIONS MÅRTEN GULLIKSSON AND ANNA OLEYNIK arxiv:6.395v [ath.na] Oct 26 Abstract. We consider the proble

More information

Boosting with log-loss

Boosting with log-loss Boosting with log-loss Marco Cusuano-Towner Septeber 2, 202 The proble Suppose we have data exaples {x i, y i ) i =... } for a two-class proble with y i {, }. Let F x) be the predictor function with the

More information

Continuity and Differentiability of the Trigonometric Functions

Continuity and Differentiability of the Trigonometric Functions [Te basis for te following work will be te definition of te trigonometric functions as ratios of te sides of a triangle inscribed in a circle; in particular, te sine of an angle will be defined to be te

More information

Combining functions: algebraic methods

Combining functions: algebraic methods Combining functions: algebraic metods Functions can be added, subtracted, multiplied, divided, and raised to a power, just like numbers or algebra expressions. If f(x) = x 2 and g(x) = x + 2, clearly f(x)

More information

Optimal Estimator for a Sample Set with Response Error. Ed Stanek

Optimal Estimator for a Sample Set with Response Error. Ed Stanek Optial Estiator for a Saple Set wit Respose Error Ed Staek Itroductio We develop a optial estiator siilar to te FP estiator wit respose error tat was cosidered i c08ed63doc Te first 6 pages of tis docuet

More information

1. Questions (a) through (e) refer to the graph of the function f given below. (A) 0 (B) 1 (C) 2 (D) 4 (E) does not exist

1. Questions (a) through (e) refer to the graph of the function f given below. (A) 0 (B) 1 (C) 2 (D) 4 (E) does not exist Mat 1120 Calculus Test 2. October 18, 2001 Your name Te multiple coice problems count 4 points eac. In te multiple coice section, circle te correct coice (or coices). You must sow your work on te oter

More information

Lower Bounds for Quantized Matrix Completion

Lower Bounds for Quantized Matrix Completion Lower Bounds for Quantized Matrix Copletion Mary Wootters and Yaniv Plan Departent of Matheatics University of Michigan Ann Arbor, MI Eail: wootters, yplan}@uich.edu Mark A. Davenport School of Elec. &

More information

Chapter 6 1-D Continuous Groups

Chapter 6 1-D Continuous Groups Chapter 6 1-D Continuous Groups Continuous groups consist of group eleents labelled by one or ore continuous variables, say a 1, a 2,, a r, where each variable has a well- defined range. This chapter explores:

More information

Estimating Parameters for a Gaussian pdf

Estimating Parameters for a Gaussian pdf Pattern Recognition and achine Learning Jaes L. Crowley ENSIAG 3 IS First Seester 00/0 Lesson 5 7 Noveber 00 Contents Estiating Paraeters for a Gaussian pdf Notation... The Pattern Recognition Proble...3

More information

Intelligent Systems: Reasoning and Recognition. Artificial Neural Networks

Intelligent Systems: Reasoning and Recognition. Artificial Neural Networks Intelligent Systes: Reasoning and Recognition Jaes L. Crowley MOSIG M1 Winter Seester 2018 Lesson 7 1 March 2018 Outline Artificial Neural Networks Notation...2 Introduction...3 Key Equations... 3 Artificial

More information

Semicircle law for generalized Curie-Weiss matrix ensembles at subcritical temperature

Semicircle law for generalized Curie-Weiss matrix ensembles at subcritical temperature Seicircle law for generalized Curie-Weiss atrix ensebles at subcritical teperature Werner Kirsch Fakultät für Matheatik und Inforatik FernUniversität in Hagen, Gerany Thoas Kriecherbauer Matheatisches

More information

The Fading Number of Memoryless Multiple-Input Multiple-Output Fading Channels 2 R n 2 =0 = 1 = 1.

The Fading Number of Memoryless Multiple-Input Multiple-Output Fading Channels 2 R n 2 =0 = 1 = 1. 65 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 53, NO. 7, JULY 007 @ R n () @ =0 = n n0 Cn(ix i +(n 0 i)y 0 n) i=0 = n n0 Cn((x i 0 y)i +ny 0 n) i=0 =(x 0 y) n Cni i +(ny 0 n) n n + (x 0 y)(ny 0 n) Cni:

More information

3.1 Extreme Values of a Function

3.1 Extreme Values of a Function .1 Etreme Values of a Function Section.1 Notes Page 1 One application of te derivative is finding minimum and maimum values off a grap. In precalculus we were only able to do tis wit quadratics by find

More information

ADVANCES ON THE BESSIS- MOUSSA-VILLANI TRACE CONJECTURE

ADVANCES ON THE BESSIS- MOUSSA-VILLANI TRACE CONJECTURE ADVANCES ON THE BESSIS- MOUSSA-VILLANI TRACE CONJECTURE CHRISTOPHER J. HILLAR Abstract. A long-standing conjecture asserts that the polynoial p(t = Tr(A + tb ] has nonnegative coefficients whenever is

More information

Logarithmic functions

Logarithmic functions Roberto s Notes on Differential Calculus Capter 5: Derivatives of transcendental functions Section Derivatives of Logaritmic functions Wat ou need to know alread: Definition of derivative and all basic

More information

Machine Learning Basics: Estimators, Bias and Variance

Machine Learning Basics: Estimators, Bias and Variance Machine Learning Basics: Estiators, Bias and Variance Sargur N. srihari@cedar.buffalo.edu This is part of lecture slides on Deep Learning: http://www.cedar.buffalo.edu/~srihari/cse676 1 Topics in Basics

More information

Robust Principal Component Analysis: Exact Recovery of Corrupted Low-Rank Matrices via Convex Optimization

Robust Principal Component Analysis: Exact Recovery of Corrupted Low-Rank Matrices via Convex Optimization Robust Principal Coponent Analysis: Exact Recovery of Corrupted Low-Rank Matrices via Convex Optiization John Wright, Arvind Ganesh, Shankar Rao, and Yi Ma Departent of Electrical Engineering University

More information

Chaotic Coupled Map Lattices

Chaotic Coupled Map Lattices Chaotic Coupled Map Lattices Author: Dustin Keys Advisors: Dr. Robert Indik, Dr. Kevin Lin 1 Introduction When a syste of chaotic aps is coupled in a way that allows the to share inforation about each

More information

SUPPLEMENT TO GEMINI: GRAPH ESTIMATION WITH MATRIX VARIATE NORMAL INSTANCES. By Shuheng Zhou

SUPPLEMENT TO GEMINI: GRAPH ESTIMATION WITH MATRIX VARIATE NORMAL INSTANCES. By Shuheng Zhou Subitted to the Annals of Statistics SUPPLEMENT TO GEMINI: GRAPH ESTIMATION WITH MATRIX VARIATE NORMAL INSTANCES By Shuheng Zhou 9. Notation and an outline. Suppose that we have n i.i.d. rando atrices

More information

Continuity and Differentiability Worksheet

Continuity and Differentiability Worksheet Continuity and Differentiability Workseet (Be sure tat you can also do te grapical eercises from te tet- Tese were not included below! Typical problems are like problems -3, p. 6; -3, p. 7; 33-34, p. 7;

More information

Chapter 5 FINITE DIFFERENCE METHOD (FDM)

Chapter 5 FINITE DIFFERENCE METHOD (FDM) MEE7 Computer Modeling Tecniques in Engineering Capter 5 FINITE DIFFERENCE METHOD (FDM) 5. Introduction to FDM Te finite difference tecniques are based upon approximations wic permit replacing differential

More information

13.2 Fully Polynomial Randomized Approximation Scheme for Permanent of Random 0-1 Matrices

13.2 Fully Polynomial Randomized Approximation Scheme for Permanent of Random 0-1 Matrices CS71 Randoness & Coputation Spring 018 Instructor: Alistair Sinclair Lecture 13: February 7 Disclaier: These notes have not been subjected to the usual scrutiny accorded to foral publications. They ay

More information

Curious Bounds for Floor Function Sums

Curious Bounds for Floor Function Sums 1 47 6 11 Journal of Integer Sequences, Vol. 1 (018), Article 18.1.8 Curious Bounds for Floor Function Sus Thotsaporn Thanatipanonda and Elaine Wong 1 Science Division Mahidol University International

More information

Characterization of the Line Complexity of Cellular Automata Generated by Polynomial Transition Rules. Bertrand Stone

Characterization of the Line Complexity of Cellular Automata Generated by Polynomial Transition Rules. Bertrand Stone Characterization of the Line Coplexity of Cellular Autoata Generated by Polynoial Transition Rules Bertrand Stone Abstract Cellular autoata are discrete dynaical systes which consist of changing patterns

More information

paper prepared for the 1996 PTRC Conference, September 2-6, Brunel University, UK ON THE CALIBRATION OF THE GRAVITY MODEL

paper prepared for the 1996 PTRC Conference, September 2-6, Brunel University, UK ON THE CALIBRATION OF THE GRAVITY MODEL paper prepared for the 1996 PTRC Conference, Septeber 2-6, Brunel University, UK ON THE CALIBRATION OF THE GRAVITY MODEL Nanne J. van der Zijpp 1 Transportation and Traffic Engineering Section Delft University

More information

A Simple Regression Problem

A Simple Regression Problem A Siple Regression Proble R. M. Castro March 23, 2 In this brief note a siple regression proble will be introduced, illustrating clearly the bias-variance tradeoff. Let Y i f(x i ) + W i, i,..., n, where

More information

DELFT UNIVERSITY OF TECHNOLOGY Faculty of Electrical Engineering, Mathematics and Computer Science

DELFT UNIVERSITY OF TECHNOLOGY Faculty of Electrical Engineering, Mathematics and Computer Science DELFT UNIVERSITY OF TECHNOLOGY Faculty of Electrical Engineering, Matematics and Computer Science. ANSWERS OF THE TEST NUMERICAL METHODS FOR DIFFERENTIAL EQUATIONS (WI3097 TU) Tuesday January 9 008, 9:00-:00

More information

INNER CONSTRAINTS FOR A 3-D SURVEY NETWORK

INNER CONSTRAINTS FOR A 3-D SURVEY NETWORK eospatial Science INNER CONSRAINS FOR A 3-D SURVEY NEWORK hese notes follow closely the developent of inner constraint equations by Dr Willie an, Departent of Building, School of Design and Environent,

More information

Numerical Differentiation

Numerical Differentiation Numerical Differentiation Finite Difference Formulas for te first derivative (Using Taylor Expansion tecnique) (section 8.3.) Suppose tat f() = g() is a function of te variable, and tat as 0 te function

More information

Constrained Consensus and Optimization in Multi-Agent Networks arxiv: v2 [math.oc] 17 Dec 2008

Constrained Consensus and Optimization in Multi-Agent Networks arxiv: v2 [math.oc] 17 Dec 2008 LIDS Report 2779 1 Constrained Consensus and Optiization in Multi-Agent Networks arxiv:0802.3922v2 [ath.oc] 17 Dec 2008 Angelia Nedić, Asuan Ozdaglar, and Pablo A. Parrilo February 15, 2013 Abstract We

More information

ESTIMATING AND FORMING CONFIDENCE INTERVALS FOR EXTREMA OF RANDOM POLYNOMIALS. A Thesis. Presented to. The Faculty of the Department of Mathematics

ESTIMATING AND FORMING CONFIDENCE INTERVALS FOR EXTREMA OF RANDOM POLYNOMIALS. A Thesis. Presented to. The Faculty of the Department of Mathematics ESTIMATING AND FORMING CONFIDENCE INTERVALS FOR EXTREMA OF RANDOM POLYNOMIALS A Thesis Presented to The Faculty of the Departent of Matheatics San Jose State University In Partial Fulfillent of the Requireents

More information

M ath. Res. Lett. 15 (2008), no. 2, c International Press 2008 SUM-PRODUCT ESTIMATES VIA DIRECTED EXPANDERS. Van H. Vu. 1.

M ath. Res. Lett. 15 (2008), no. 2, c International Press 2008 SUM-PRODUCT ESTIMATES VIA DIRECTED EXPANDERS. Van H. Vu. 1. M ath. Res. Lett. 15 (2008), no. 2, 375 388 c International Press 2008 SUM-PRODUCT ESTIMATES VIA DIRECTED EXPANDERS Van H. Vu Abstract. Let F q be a finite field of order q and P be a polynoial in F q[x

More information

On Conditions for Linearity of Optimal Estimation

On Conditions for Linearity of Optimal Estimation On Conditions for Linearity of Optial Estiation Erah Akyol, Kuar Viswanatha and Kenneth Rose {eakyol, kuar, rose}@ece.ucsb.edu Departent of Electrical and Coputer Engineering University of California at

More information

Support Vector Machine Classification of Uncertain and Imbalanced data using Robust Optimization

Support Vector Machine Classification of Uncertain and Imbalanced data using Robust Optimization Recent Researches in Coputer Science Support Vector Machine Classification of Uncertain and Ibalanced data using Robust Optiization RAGHAV PAT, THEODORE B. TRAFALIS, KASH BARKER School of Industrial Engineering

More information

Topic 5a Introduction to Curve Fitting & Linear Regression

Topic 5a Introduction to Curve Fitting & Linear Regression /7/08 Course Instructor Dr. Rayond C. Rup Oice: A 337 Phone: (95) 747 6958 E ail: rcrup@utep.edu opic 5a Introduction to Curve Fitting & Linear Regression EE 4386/530 Coputational ethods in EE Outline

More information

The degree of a typical vertex in generalized random intersection graph models

The degree of a typical vertex in generalized random intersection graph models Discrete Matheatics 306 006 15 165 www.elsevier.co/locate/disc The degree of a typical vertex in generalized rando intersection graph odels Jerzy Jaworski a, Michał Karoński a, Dudley Stark b a Departent

More information

CONVERGENCE OF A MIXED METHOD FOR A SEMI-STATIONARY COMPRESSIBLE STOKES SYSTEM

CONVERGENCE OF A MIXED METHOD FOR A SEMI-STATIONARY COMPRESSIBLE STOKES SYSTEM CONVERGENCE OF A MIXED METHOD FOR A SEMI-STATIONARY COMPRESSIBLE STOKES SYSTEM KENNETH H. KARLSEN AND TRYGVE K. KARPER Abstract. We propose and analyze a finite eleent etod for a sei stationary Stoes syste

More information

W-BASED VS LATENT VARIABLES SPATIAL AUTOREGRESSIVE MODELS: EVIDENCE FROM MONTE CARLO SIMULATIONS

W-BASED VS LATENT VARIABLES SPATIAL AUTOREGRESSIVE MODELS: EVIDENCE FROM MONTE CARLO SIMULATIONS W-BASED VS LATENT VARIABLES SPATIAL AUTOREGRESSIVE MODELS: EVIDENCE FROM MONTE CARLO SIMULATIONS. Introduction When it coes to applying econoetric odels to analyze georeferenced data, researchers are well

More information

Detection and Estimation Theory

Detection and Estimation Theory ESE 54 Detection and Estiation Theory Joseph A. O Sullivan Sauel C. Sachs Professor Electronic Systes and Signals Research Laboratory Electrical and Systes Engineering Washington University 11 Urbauer

More information

HOMEWORK HELP 2 FOR MATH 151

HOMEWORK HELP 2 FOR MATH 151 HOMEWORK HELP 2 FOR MATH 151 Here we go; te second round of omework elp. If tere are oters you would like to see, let me know! 2.4, 43 and 44 At wat points are te functions f(x) and g(x) = xf(x)continuous,

More information

Supplement to: Subsampling Methods for Persistent Homology

Supplement to: Subsampling Methods for Persistent Homology Suppleent to: Subsapling Methods for Persistent Hoology A. Technical results In this section, we present soe technical results that will be used to prove the ain theores. First, we expand the notation

More information

Finding and Using Derivative The shortcuts

Finding and Using Derivative The shortcuts Calculus 1 Lia Vas Finding and Using Derivative Te sortcuts We ave seen tat te formula f f(x+) f(x) (x) = lim 0 is manageable for relatively simple functions like a linear or quadratic. For more complex

More information

Lecture 21 Nov 18, 2015

Lecture 21 Nov 18, 2015 CS 388R: Randoized Algoriths Fall 05 Prof. Eric Price Lecture Nov 8, 05 Scribe: Chad Voegele, Arun Sai Overview In the last class, we defined the ters cut sparsifier and spectral sparsifier and introduced

More information

Lecture 9 November 23, 2015

Lecture 9 November 23, 2015 CSC244: Discrepancy Theory in Coputer Science Fall 25 Aleksandar Nikolov Lecture 9 Noveber 23, 25 Scribe: Nick Spooner Properties of γ 2 Recall that γ 2 (A) is defined for A R n as follows: γ 2 (A) = in{r(u)

More information

Pattern Recognition and Machine Learning. Artificial Neural networks

Pattern Recognition and Machine Learning. Artificial Neural networks Pattern Recognition and Machine Learning Jaes L. Crowley ENSIMAG 3 - MMIS Fall Seester 2017 Lessons 7 20 Dec 2017 Outline Artificial Neural networks Notation...2 Introduction...3 Key Equations... 3 Artificial

More information

Computable Shell Decomposition Bounds

Computable Shell Decomposition Bounds Coputable Shell Decoposition Bounds John Langford TTI-Chicago jcl@cs.cu.edu David McAllester TTI-Chicago dac@autoreason.co Editor: Leslie Pack Kaelbling and David Cohn Abstract Haussler, Kearns, Seung

More information

Keywords: Estimator, Bias, Mean-squared error, normality, generalized Pareto distribution

Keywords: Estimator, Bias, Mean-squared error, normality, generalized Pareto distribution Testing approxiate norality of an estiator using the estiated MSE and bias with an application to the shape paraeter of the generalized Pareto distribution J. Martin van Zyl Abstract In this work the norality

More information

PROXSCAL. Notation. W n n matrix with weights for source k. E n s matrix with raw independent variables F n p matrix with fixed coordinates

PROXSCAL. Notation. W n n matrix with weights for source k. E n s matrix with raw independent variables F n p matrix with fixed coordinates PROXSCAL PROXSCAL perfors ultidiensional scaling of proxiity data to find a leastsquares representation of the obects in a low-diensional space. Individual differences odels can be specified for ultiple

More information

Analytic Functions. Differentiable Functions of a Complex Variable

Analytic Functions. Differentiable Functions of a Complex Variable Analytic Functions Differentiable Functions of a Complex Variable In tis capter, we sall generalize te ideas for polynomials power series of a complex variable we developed in te previous capter to general

More information

Ch 12: Variations on Backpropagation

Ch 12: Variations on Backpropagation Ch 2: Variations on Backpropagation The basic backpropagation algorith is too slow for ost practical applications. It ay take days or weeks of coputer tie. We deonstrate why the backpropagation algorith

More information

Inference in the Presence of Likelihood Monotonicity for Polytomous and Logistic Regression

Inference in the Presence of Likelihood Monotonicity for Polytomous and Logistic Regression Advances in Pure Matheatics, 206, 6, 33-34 Published Online April 206 in SciRes. http://www.scirp.org/journal/ap http://dx.doi.org/0.4236/ap.206.65024 Inference in the Presence of Likelihood Monotonicity

More information

Biostatistics Department Technical Report

Biostatistics Department Technical Report Biostatistics Departent Technical Report BST006-00 Estiation of Prevalence by Pool Screening With Equal Sized Pools and a egative Binoial Sapling Model Charles R. Katholi, Ph.D. Eeritus Professor Departent

More information

A note on the multiplication of sparse matrices

A note on the multiplication of sparse matrices Cent. Eur. J. Cop. Sci. 41) 2014 1-11 DOI: 10.2478/s13537-014-0201-x Central European Journal of Coputer Science A note on the ultiplication of sparse atrices Research Article Keivan Borna 12, Sohrab Aboozarkhani

More information

AN OPTIMAL SHRINKAGE FACTOR IN PREDICTION OF ORDERED RANDOM EFFECTS

AN OPTIMAL SHRINKAGE FACTOR IN PREDICTION OF ORDERED RANDOM EFFECTS Statistica Sinica 6 016, 1709-178 doi:http://dx.doi.org/10.5705/ss.0014.0034 AN OPTIMAL SHRINKAGE FACTOR IN PREDICTION OF ORDERED RANDOM EFFECTS Nilabja Guha 1, Anindya Roy, Yaakov Malinovsky and Gauri

More information

A Bernstein-Markov Theorem for Normed Spaces

A Bernstein-Markov Theorem for Normed Spaces A Bernstein-Markov Theore for Nored Spaces Lawrence A. Harris Departent of Matheatics, University of Kentucky Lexington, Kentucky 40506-0027 Abstract Let X and Y be real nored linear spaces and let φ :

More information