ÉCOLE POLYTECHNIQUE FÉDÉRALE DE LAUSANNE

Size: px
Start display at page:

Download "ÉCOLE POLYTECHNIQUE FÉDÉRALE DE LAUSANNE"

Transcription

1 ÉCOLE POLYTECHNIQUE FÉDÉRALE DE LAUSANNE School of Computer ad Commuicatio Scieces Hadout Iformatio Theory ad Sigal Processig Compressio ad Quatizatio November 0, 207 Data compressio Notatio Give a set A we deote by A the set of all fiite sequeces {(a,, a ) : 0, a i A} (icludig the ull sequece λ of legth 0) I particular {0, } = {λ, 0,, 00, 0, 0,, 000, } Cosider the problem of assigig biary sequeces (also called biary strigs) to elemets of a fiite set U Such a assigmet c : U {0, } is called a biary code for the set U The biary strig c(u) is called the codeword for u The collectio {c(u) : u U} is thus the set of codewords Defiitio A code c is called ijective if for all u v we have c(u) c(v) Defiitio 2 A code c is called prefix-free if c(u) is ot a prefix of c(v) for all u v I particular, if c is prefix-free the c is ijective (To be clear: a strig a a m is a prefix of a strig b b if m ad a i = b i for i =,, m Thus, the ull strig is a prefix of ay strig, ad each strig is a prefix of itself) Lemma Suppose c : U {0, } is ijective The, u 2 legth(c(u)) log 2 ( + U ) Proof Without loss of geerality, we ca assume that wheever k = legth(c(u)) for some u, the for every biary strig b of legth i < k there is a v with b = c(v) (Otherwise, there is a b with legth(b) < k which is ot a codeword, ad replacig c(u) with b will preserve the ijectiveess of c ad icrease the left had side of the iequality) For such a code c, with k deotig the legth of the logest codeword, the set of codewords is the uio of k i=0 {0, }i with a o-empty subset of {0, } k With r 2 k deotig the cardiality of this last subset, we have U = 2 k + r ad u 2 legth(c(u)) = k + r2 k As log 2 ( + U ) = k + log 2 ( + r2 k ) ad 0 < r2 k, all we eed to show is x log 2 ( + x) for 0 < x As equality obtais for x = 0 ad x =, the iequality follows from the cocavity of log Lemma 2 Suppose c : U {0, } is prefix-free The, u 2 legth(c(u)) Coversely, if l : U {0,, 2, } with u 2 l(u), the there exists a prefix-free code c : U {0, } with legth(c(u)) = l(u) Proof Give a biary sequece a = a a m, let p(a) = m a i2 i deote the ratioal umber whose biary expasio is 0a a m With this otatio, a biary sequece a = a a m is a prefix of a biary sequece b = b b if ad oly the p(b) lies i the iterval I(a) = [p(a), p(a) + 2 m ) For the first claim, observe that c beig prefix-free thus implies that the itervals I(c(u)) are disjoit As I(c(u)) is of size 2 legth(c(u)) ad all of the itervals are icluded i [0, ), the iequality follows For the secod claim, order the elemets of U as u,, u K such that l := l(u ) l K := l(u K ) Let p k = i<k 2 l i ad set I k = [p k, p k + 2 l k ) Observe that the itervals I,, I K are disjoit, ad I k [0, ) Furthermore, for each k, 2 l k pk is a iteger, thus p k ca be expressed i biary as 0b (k) with b (k) a biary strig of legth l k The code c(u k ) = b (k) ow has the required properties it beig prefix free a cosequece of the disjoitess of the collectio itervals I k

2 Lemma 3 Suppose P Π(U) is a probability distributio o U ad U is radom variable with distributio P The, with H(U) = u P (u) log 2 P (u) deotig the etropy of U, (i) for ay prefix-free c : U {0, }, E[legth(c(U))] H(U); (ii) there exists a prefix-free c : U {0, } with E[legth(c(U))] H(U) + ; (iii) for ay ijective c : U {0, }, E[legth(c(U))] H(U) log 2 log 2 ( + U ), (iv) there exists a ijective c : U {0, } with E[legth(c(U))] H(U) Proof For (i) ad (iii) let Q(u) = 2 legth(c(u)) ad observe that H(U) E[legth(c(U))] = u P (u) log 2 Q(u) P (u) log 2 Q(u), where the iequality is because log is cocave Whe c is prefix-free u Q(u) by Lemma 2, ad whe c is ijective u Q(u) log 2( + U ) by Lemma The iequalities (i) ad (iii) thus follow For (ii) set l(u) = log 2 P (u) As 2 l(u) P (u), we see that u 2 l(u) ad by Lemma 2 there exists a prefix-free code c with legth(c(u)) = l(u) As l(u) < log 2 P (u)+, (ii) follows For (iv) order the elemets of U as u,, u K with P (u ) P (u K ) Let c(u k ) = b k where b k is the kth elemet of the sequece λ, 0,, 00, 0, 0,, 000, 00,, (eg, b = λ, b 2 = 0, b 3 =, b 4 = 00,, b 9 = 00, ) Observe that legth(b k ) = log 2 k log 2 k Also ote that k P (u i) kp (u k ), ad thus log 2 k log 2 P (u k ) Cosequetly, for this c, E[legth(c(U))] k P (u k) log 2 P (u k ) = H(U) Corollary Suppose U, U 2, is a stochastic process The for ay sequece c : U {0, } of ijective codes lim if E[legth(c (U ))] lim if ad there exists a sequece c of prefix-free codes for which lim sup H(U ), E[legth(c (U ))] lim sup H(U ) I particular, if r = lim H(U ) exists, all faithful represetatios of the process U, U 2, with bits will asymptotically require at least r bits per letter, ad there is a represetatio that asymptotically requires exactly as much Proof The first iequality follows from otig that E[legth(c (U ))] H(U ) log 2 log 2 ( + U ) ad observig that lim log 2 log 2 (+ U ) = 0 The secod iequality follows from otig that there exist prefix-free c with ad that lim / = 0 E[legth(c (U ))] H(U ) + u 2

3 Remark Lemma 2 gives evidece of a strog coectio betwee prefix-free codes ad probability distributios O the oe had, give a prefix-free code c, oe ca costruct a probability distributio Q that assigs the letter u the probability Q(u) = 2 legth(c(u)) By the lemma, u Q(u) ; if equality holds Q is ideed a probability distributio, otherwise, we ca assig u Q(u) as the probability Q(u 0) of a fictitious symbol u 0 U If U is a radom radom variable with distributio P, we the have (by assigig P (u 0 ) = 0 if ecessary), E[legth(c(U))] H(U) = u P (u)[legth(c(u)) + log P (u)] = D(P Q) O the other had, give a distributio Q Π(U), by Lemma 2 we ca costruct a prefix-free code c : U {0, } with legth(c(u)) = log 2 Q(u) As log 2 Q(u) legth(c(u)) < log 2 Q(u) +, we see that E[legth(c(U))] H(U) = u P (u)[legth(c(u)) + log 2 P (u)] is bouded from below by D(P Q), ad from above D(P Q) + These observatios give the divergece D(P Q) a iterpretatio as the expected umber of excess bits (beyod the miimum possible H(U)) a code based o Q requires whe describig a radom variable with distributio P Cosequetly, if we are give S Π ad told that the distributio P of a radom variable U belogs to S, a reasoable strategy to desig a code c is to look for a distributio Q Π such that sup D(P Q) P S is small (eg, by fidig the Q that miimizes this quatity) ad costruct a code c based o Q as above Example To illustrate the remark above, suppose we are told that U, U 2, are biary ad iid radom variables The distributio of U ca be parametrized by θ = Pr(U = ), ad is give by Pr(U = u ) = P θ (u ) = ( θ) 0(u ) θ (u ) where 0 (u ) ad (u ) are the umber of zeros ad oes i the sequece u u With this otatio, S = {Pθ : 0 θ } is the class of distributios that we are told the distributio of U belogs to Cosider ow a sequece of coditioal distributios Q Uk+ U k(u uk ) = u(u k ) + k + 2 where u (u k ) is as above, deotig the umber of u s i u u k Note that Q U (0) = Q U () = /2 Defie Q (u ) = Q Ui U i (u i u i ) Oe ca prove by iductio o, that for ay ad ay u {0, }, Q (u ) ( 0 (u ) + 3 ) 0 (u )( (u ) ) (u )

4 If U,, U are iid with commo distributio P θ, [ D(Pθ Q ) = E log P θ (U ) ] Q (U ) [ log( + ) + E log [ = log( + ) + E 0 (U ) log log( + ) + ( θ) log Pθ (U ) ] ( 0 (U )/) 0(U ) ( (U )/) (U ) ( θ) 0 (U ) + (U ) log ( θ) θ + θ log ( θ) θ θ (U ) ] = log( + ), where the iequality i the last lie is because x x log[/x] is cocave ad E[ 0 (U )] = ( θ), ad E[ (U )] = θ Cosequetly, we see that sup P S D(P Q ) log( + ) If Q were used to costruct a prefix-free code c : {0, } {0, }, by the remark above, c will satisfy E[legth(c (U ))] H(P ) [log( + ) + ] wheever U is iid with distributio P As the right had side vaishes as gets large, we would be justified to call the sequece of codes c asymptotically uiversal for the class of biary iid data I the exercises we will see aother choice of Q which improves the upper boud o D(P Q ) to log 2 Note that, had we chose Q to be a member of S, say Q = Pθ 0 for some θ 0, the D(Pθ Q ) would have grow liearly i for ay θ θ 0 Thus, eve if we kow that the true distributio P is i S, choosig Q outside of S (as we have doe above) may lead to a better code costructio Remark 2 The example above also illustrates a coectio betwee compressio ad predictio (Oe ca also use the term learig istead of predictio) Suppose we have a family S of distributios o U, ad we are give a prefix-free code c : U {0, } performs well, i the sese that sup P S E P [legth(c (U ))] H(U ) is small Costruct the distributio Q associated with the code c, ie, Q(u ) = 2 legth(c(u )) ad factorize it as Q(u ) = Q(u i u i ) As the code c performs well, D(P Q) is small for all P S But D(P Q) = P (u ) log P (u ) Q(u ) u = P (u ) log P (ui ui ) Q(u u i u i ) = P (u i ) log P (ui ui ) Q(u i u i ) u i = P (u i ) P (u i u i ) log P (ui ui ) Q(u u i u i u i ) i = P (u i )D(P ( u i ) Q( u i )), u i 4

5 so we coclude that for a large fractio of i s i,,, ad for a set of u i s with large P probability, the quatity D(P ( u i ) Q( u i )) is small Which is to say, o matter what P from S is the true distributio of the data, if after observig u i we predicted the distributio of the ext symbol u i to be Q( u i ), our predictio will be close to the true distributio P ( u i ) for most i s ad for a high probability set of u i s 2 Uiversal data compressio with the Lempel-Ziv algorithm I the example i the previous sectio we saw a compressio method that was uiversal over the class of biary iid processes We will ow see a much more powerful method that is uiversal over all statioary processes The method was iveted by Ziv ad Lempel i 977, the versio we preset here is a variat due to Welch from 984 Give a alphabet U = {a,, a K }, the method ecodes a ifiite sequece u u 2 from this alphabet to biary as follows: Set a dictioary D = U Deote the dictioary etries as d(0) = a,, d(s ) = a K, with s = K beig the size of the dictioary Set i = 0 (the umber of iput letters read so far) 2 Fid the largest l such that w = u i+ u i+l is i D 3 With 0 j < s deotig the idex of w i D, output the log 2 s bit biary represetatio of j 4 Add the word wu i+l+ to D, ie, set d(s) = wu i+l+, ad icremet s by Icremet i by l Goto step 2 For example, with U = {a, b}, the iput strig abbbbaaab will lead to the executio steps D at 2 w output at 3 added-word at 4 a b a 0 ab a b ab b 0 bb a b ab bb bb bbb a b ab bb bbb b 00 ba a b ab bb bbb ba a 000 aa a b ab bb bbb ba aa aa 0 aab The first questio we eed to aswer is if we ca recover the iput sequece u u 2 from the output of the algorithm The questio is aswered i the affirmative i Lemma 4 below Note that the algorithm parses the sequece u u 2 ito a sequece of words w, w 2, foud at step 2 of the algorithm So, recovery of u u 2 is equivalet to the recovery of these words Let j, j 2, the dictioary idices that appear at step 3, ad d, d 2, the words added to the dictioary i step 4 As the dictioary size s icreases by each time a dictioary word is parsed, the bitstream that is output by the algorithm ca be parsed ito the idices j, j 2 Lemma 4 From j,, j i we ca determie w,, w i I other words, we ca recover the iput u u 2 from the output of the algorithm To be cocrete, if D(P Q) is less tha ɛ, the, except for a ɛ/3 fractio of the i s, we have u P i (ui )D(P ( u i ) Q( u i )) < ɛ 2/3, ad except for a set of P probability ɛ /3 of u i s, we have D(P ( u i ) Q( u i )) < ɛ /3 5

6 Proof From j, we ca determie w, ad so the claim is true for i = We proceed by iductio Suppose ow we observe j,, j i+ By the iductio hypothesis, j,, j i determies w,, w i Sice d k is the cocateatio of w k with the first letter of w k+, we kow d,, d i, ad, except for its last letter, d i We eed to show that we ca recostruct w i+ from the additioal iformatio obtaied by j i+ Note that j i+ refers to a word i the dictioary formed by augmetig U with the words d,, d i If j i+ refers to ay word other tha d i, we already determie w i+ Otherwise j i+ refers to d i But i this case the last letter of d i equals the first letter of d i which is already kow, ad we ca agai determie w i+ Next, we will obtai a upper boud o the umber of bits per letter the algorithm uses to describe a sequece u u 2 The aswer will be give i Theorem below Lemma 5 A word w ca appear i the sequece w, w 2, at most U times Proof As the algorithm always looks for the logest word w i the dictioary that matches the start of the as-yet-uprocessed segmet of the iput, wu i+l+ is ot i the dictioary before its additio to the dictioary i step 4 Thus the words added to the dictioary are distict For each occurrece of a word w i the parsig a word of the form wu with u U is added to the dictioary Sice these are distict, w caot appear more tha U times i the parsig Lemma 6 Suppose u = u u is parsed ito m(u ) words w w m by the algorithm The lim m(u )/ = 0 Proof There are U i words of legth i, ad by the previous lemma, each ca appear at most U times i the list w,, w m As the algorithm does ot parse the ull strig, at most F (k) = U k U i words i the list are of legth k or less, ad each of the remaiig words i the list has legth k or more Thus k[m F (k)] Cosequetly, lim sup m(u ) As k is arbitrary, the lemma follows lim sup /k + F (k) = /k Theorem Let l(u ) deote the umber of bits produced by the algorithm after readig u The, lim sup l(u )/ lim sup m(u ) log m(u ) Proof As the dictioary size icreases by at each iteratio of the algorithm, with m = m(u ), l(u ) = m i=0 log 2( U + i) Thus l(u ) m log 2 ( U + m ) + m = m log 2 m + m log 2 ( + ( U )/m) + m, ad the lemma follows from lim m(u ) = 0 We ow have a upper boud to the umber of bits per letter the LZW algorithm requires to describe a sequece u u 2 Next, we will derive a lower boud to the umber of bits per letter produced by ay iformatio loss fiite state machie that maps u u 2 to a sequece of bits (the terms fiite state machie ad iformatio lossless will be defied formally i the paragraphs that follow) The lower boud will eve apply to machies that may have bee desiged with prior kowledge of the sequece u u 2, ad most remarkably, will match the upper boud just derived above That is to say, for ay u u 2, LZW competes well agaist ay iformatio lossless fiite state machie I particular, each prefix-free ecoder c that appears i Corollary ca be implemeted by a fiite state iformatio lossless machie Cosequetly, oce we obtai the lower boud, we will have proved: 6

7 Theorem 2 If U, U 2 is a statioary ad ergodic stochastic process, the the umber of bits per letter emitted by LZW whe its iput is U U 2 approaches lim H(U )/ with probability oe 2 Fiite state iformatio lossless ecoders For our purposes, a fiite state machie is a device that reads the iput sequece oe symbol at a time Each symbol of the iput sequece belogs to a fiite alphabet U with U symbols The machie is i oe of a fiite umber s of states before it reads a symbol, ad goes to a ew state determied by the old state ad the symbol read We will assume that the machie is i a fixed, kow state z before it reads the first iput symbol The machie also produces a fiite strig of biary digits (possibly the ull strig) after each iput This output strig is agai a fuctio of the old state ad the iput symbol That is, whe the ifiite sequece u = u u 2 is give as the iput, the ecoder produces y = y y 2, while visitig a ifiite sequece of states z = z z 2, give by y k = f(z k, u k ), k z k+ = g(z k, u k ), k where the fuctio f takes values o the set {0, } of fiite biary strigs, so that each y k is a (perhaps ull) biary strig A fiite segmet x k x k+ x j of a sequece x = x x 2 will be deoted by x j k, ad by a abuse of the otatio, the fuctios f ad g will be exteded to idicate the output sequece ad the fial state Thus, f(z k, u j k ) will deote yj k ad g(z k, u j k ) will deote z j+ Without loss of geerality we will assume that ay state z is reachable from the iitial state z ie, that some iput sequece will take the machie from state z to z To make the questio of compressibility meaigful oe has to require some sort of a ivertibility coditio o the fiite state ecoders Give the descriptio of the fiite state machie that ecoded the strig, ad the startig state z, but (of course) without the kowledge of the iput strig, it should be possible to recostruct the iput strig u from the output of the ecoder y A weaker requiremet tha this is the followig: for ay state z ad two distict iput sequeces v m ad ṽ, either f(z, v m ) f(z, ṽ ) or g(z, v m ) g(z, ṽ ) A ecoder satisfyig this coditio will be called iformatio lossless (IL) It is clear that if a ecoder is ot IL, the there is o hope to recover the iput from the output, ad thus every ivertible ecoder is IL 2 We will eed to the followig fact: Lemma 7 Suppose v,, v m are biary strigs, with o strig occurrig more tha k times The, writig m = j i=0 k2i + r with 0 r < k2 j, we have m legth(v i) k j i=0 i2i + rj Proof The set of biary strigs ordered i icreasig legth cosists of: strig of legth 0, 2 strigs of legth,, 2 i strigs of legth i, The shortest total legth for the v i s will be attaied if the v i s are chose by traversig the set of all biary strigs i icreasig legth, each strig repeated k times, util all strigs of legth j or less are repeated k times, ad we are left to fid 0 r < k2 j strigs, which are chose from the set of strigs 2 However, as illustrated i Figure, a IL ecoder is ot ecessarily uiquely decodable Startig from state S, two distict iput sequeces will leave the ecoder i distict states if they have differet first symbols, otherwise they will lead to differet output sequeces Thus, the above ecoder is IL Nevertheless, o decoder ca distiguish betwee the iput sequeces aaaa ad bbbb by observig the output 000 7

8 a / 0 b / a / λ A S b / λ B b / 0 a / A fiite state machie with three states S, A ad B The otatio i /output meas that the machie produces output i respose to the iput i λ deotes the ull output Figure : A IL ecoder which is ot uiquely decodable of legth j The lower boud i the lemma is precisely the total legth of this optimal collectio Lemma 8 Suppose v,, v m are biary strigs, with o strig occurrig more tha k times The, m m legth(v i ) m log 2 8k () Proof Notig that j i=0 2i = 2 j ad j i=0 i2i = (j 2)2 j + 2, the previous lemma states: writig m = k2 j k + r with 0 r < k2 j, the total legth of the v i s is lower bouded by k((j 2)2 j + 2) + rj = (j 2)m + kj + 2r (j 2)m As r < k2 j, we have m < k(2 j+ ) Rearragig, we get 2 j+ > + m/k > m/k, ad thus j 2 > log m 8k Now we ca state the followig Lemma 9 For ay IL-ecoder with s states, legth(y ) m(u ) log 2 m(u ) 8 U s 2 (2) where, as before, m(u ) is the umber of words i the parsig of u by LZW Proof Let u = w w m be the parsig of the iput by LZW Let z i be the state the IL machie is i at just before it reads w i, ad z i+ be the state just after it has read w i Let t i be the biary strig output by the IL machie while it reads w i, so that y = t m No biary strig t ca occur i t,, t m more tha k = s 2 U times If it did, there will be a state-pair (z, z ) that occurs amog (z i, z i+ ) more tha U times with t i = t By Lemma 5 the there will be w i w j with t i = t j = t ad (z i, z i+ ) = (z j, z j+ ) = (z, z ) But this 8

9 cotradicts the IL property of the machie Thus the output of the IL machie t t m is a cocateatio of m biary strigs with o strig occurrig more tha k = s 2 U times, so their total legth is at least m log m 8k Usig the lemma just proved, ad by Lemma 6 we have Theorem 3 For ay fiite state iformatio lossless ecoder, the umber of output bits Let l(u ) produced by the ecoder after readig u satisfies 3 Quatizatio lim sup l(u m(u ) )/ lim sup log m(u ) Ofte it is ot ecessary to represet data exactly; a approximate represetatio suffices, eg, we may be cotet if a sequece u of bits areare reproduced as a sequece v of bits as log as v ad u differ oly at a few idices We formulate this state of affairs as follows: our data is a sequece of radom variables U, each U i takig values i a alphabet U We are give a represetatio alphabet V ad also a distortio measure d : U V R that gives the distortio caused by represetig a data letter u by v For defie d (u, v ) = d(u i, v i ) A quatizer cosists of two maps: f : U {,, M} ad g : {,, M} V The data sequece u is mapped by the quatizer to f (u ), which is a log M bit represetatio of the data, ad recostructed from this represetatio as v = g (f (u )) We measure the quality of the the quatizer by two criteria: its rate R = (log M)/ (the umber of bits per data letter), ad its expected distortio = E[d (U, V )] where V = g (f (U )) The followig lemma says that oe caot have too small rate for a give level of distortio Lemma 0 Suppose d is a distortio measure, U, U 2, are iid with distributio P U, ad we have a quatizer with rate R ad expected distortio The, there exists a distributio P UV such that R I(U; V ) ad = E[d(U, V )] Proof Let P Ui V i deote the joit distributio of (U i, V i ) ad set P UV = P U i V i We will show that P UV satisfies the coditios claimed i the lemma First ote that the U- margial of each P Ui V i equals P U, so P UV ideed has U margial P U Also, P V = i P V i Now, by the data processig theorem for mutual iformatio, R I(U ; V ) As the U i s are idepedet, H(U ) = i H(U i) Also by the chai rule ad removig terms from the coditioig H(U V ) i H(U i V i ) Thus I(U ; V ) = H(U ) H(U V ) ad cosequetly R I(U i ; V i ) = H(U i ) H(U i V i ) = I(U i ; V i ), D(P Ui V i P U P Vi ) D(P UV P U P V ) = I(U; V ) where the last iequality is due to the covexity of D( ) Furthermore, E[d(U, V )] = u,v P UV (u, v)d(u, v) = P Ui V i (u, v)d(u, v) = E[d (U, V )] = 9

10 The lemma says that the possible rate distortio pairs whe quatizig a iid source with distributio P lie i the followig regio of R 2 : { } (R, ) : R I(U; V ), = E[d(U, V )] P UV :P U =P We will ow show that there are quatizers whose performace closely approximates the boud give i the lemma Theorem 4 Give P U, a distortio measure d ad a joit distributio P UV, for ay ɛ > 0, there is a quatizer (f, g ) with rate R < I(U; V ) + ɛ ad expected distortio satisfyig E[d(U, V )] < ɛ whe it is used to quatize iid data U with distributio P U To prove the theorem we will make use of the followig lemma Lemma For all δ > 0, for all large eough, for every joit type P UV Π (U V), there is a subset S := S(P UV ) T (P V ) with S 2 (I(U;V )+δ) so that for every u T (P U ) there is a v S for which (u, v ) T (P UV ) Proof Let M = 2 (I(U;V )+δ), ad choose V (), V (M) idepedetly ad uiformly from T (P V ) Let S be the (radom) set {V (),, V (M)} For each u T (P U ), let Z(u ) = M {(u, V (m)) T (P UV )} m= Note that the S has the claimed property if ad oly if Z = u T (P U ) Z(u ) = 0 We will prove the existece of the set S with the property, by showig that E[Z] < To that ed, ote that for a give u T (P U ), the M radom variables {(u, V (m)) T (P UV )} are idepedet, with expectatio T (P UV ) T (P U ) T (P V ) Cosequetly, ( ) + U V 2 [H(UV ) H(U) H(V )] 2 [I(U;V )+δ/2] U V E[Z(u )] ( 2 (I(U;V )+δ/2) ) M exp( M2 (I(U;V )+δ/2) ) exp( 2 δ/2 ) As this expectatio is doubly expoetially small i, ad sice T (P U ) is oly expoetially large, we coclude that for large eough we will have E[Z] < Proof of the theorem Give P UV, set R 0 = I(U; V ) ad 0 = E[d(U, V )] Fix δ > 0, ad let Ω the set of all joit distributios Q o U V such that I(U; V ) < R 0 + δ ad E[d(U, V )] 0 < δ whe (U, V ) has joit distributio Q The distributio PUV belogs to Ω, thus Ω is a o-empty ope set of joit distributios Let Ω 0 be the set of U margials of the distributios i Ω Note that Ω 0 is also ope ad o-empty as it cotais P U As a cosequece D := if P Ω0 D(P P U ) = D(P P U ) for some P Ω 0, ad thus D > 0 sice P ca t equal P U Ω 0 Let S(Q) be as i the lemma above ad set S = S(Q) where the uio is over all Q Ω Π (U V) For large eough M = S 2 (R0+3δ), sice each S(Q) has S(Q) 2 (R0+2δ) ad the uio cotais oly Π (which is a polyomial i ) such sets Assig to each v i S a uique idex i {,, M}, ad for i =,, M, let g (i) be the elemet of S with idex i 0

11 By the costructio of the set S, for every P Ω 0 Π (U) ad every u T (P ) we ca fid a v i S such that the type of (u, v ) belogs to Ω, ad thus d (u, v ) 0 < δ For u P Ω 0 T (P ) defie f (u ) to be the idex of such a v i S For u ot i this uio defie f (u ) i ay arbitrary fashio, eg, by settig f (u ) = For those u, with v = g (f (u )), we do ot ecessarily have a small value for d (u, v ) 0, but evertheless we ca say d (u, v ) 0 < 2 max u,v d(u, v) Whe U,, U are iid with distributio P U, the probability that the type of U does ot belog to Ω 0 decays expoetially to zero as 2 D Cosequetly, with V = g (f (U )) η = Pr ( d (U, V ) 0 + δ ) decays to zero like 2 D Thus the quatizer (f, g ) has rate log M R 0 + 3δ, ad distortio = E[d (U, V )] satisfyig 0 E [ d(u, V ) 0 ] < δ + η 2 max d(u, v) < 2δ u,v for large eough As δ > 0 is arbitrary, the theorem follows I the light of the Theorem above ad Lemma 0 that precedes it, it makes sese to defie Defiitio 3 Give source ad reproductio alphabets U ad V, a distortio fuctio d : U V R, ad a distributio P U o U, defie the rate distortio fuctio R( ) ad distortio rate fuctio (R) as R( ) = if{i(u; V ) : E[d(U, V )] }, (R) = if{e[d(u, V ) : I(U; V ) R} where the ifima are over all joit distributios P UV with U margial P U Lemma 0 says that ay quatizer with rate at most R must have expected distortio at least (R), ad ay quatizer with expected distortio at most must have rate at least R( ) The theorem says that these bouds are achievable arbitrarily closely As a example cosider U = V = {0, }, d(u, v) = {u v}, ad P U () = p /2 It is easy to show that { 0 p R( ) = h 2 (p) h 2 ( ) 0 < p} where h 2 (q) = q log 2 q ( q) log 2 ( q) is the biary etropy fuctio I particular, for p = /2 ad = 0, R( ) 053, so it is possible to compress biary data by almost a factor half ad still be 90 percet accurate The aive approach for achievig the same accuracy would have retaied 80 percet of the data bits (the remaig 20 percet ca the be guessed i ay fashio; the guesses would be right with probability /2, resultig i 90 percet overall accuracy)

Information Theory and Statistics Lecture 4: Lempel-Ziv code

Information Theory and Statistics Lecture 4: Lempel-Ziv code Iformatio Theory ad Statistics Lecture 4: Lempel-Ziv code Łukasz Dębowski ldebowsk@ipipa.waw.pl Ph. D. Programme 203/204 Etropy rate is the limitig compressio rate Theorem For a statioary process (X i)

More information

Lecture 14: Graph Entropy

Lecture 14: Graph Entropy 15-859: Iformatio Theory ad Applicatios i TCS Sprig 2013 Lecture 14: Graph Etropy March 19, 2013 Lecturer: Mahdi Cheraghchi Scribe: Euiwoog Lee 1 Recap Bergma s boud o the permaet Shearer s Lemma Number

More information

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014.

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014. Product measures, Toelli s ad Fubii s theorems For use i MAT3400/4400, autum 2014 Nadia S. Larse Versio of 13 October 2014. 1. Costructio of the product measure The purpose of these otes is to preset the

More information

Math 155 (Lecture 3)

Math 155 (Lecture 3) Math 55 (Lecture 3) September 8, I this lecture, we ll cosider the aswer to oe of the most basic coutig problems i combiatorics Questio How may ways are there to choose a -elemet subset of the set {,,,

More information

(A sequence also can be thought of as the list of function values attained for a function f :ℵ X, where f (n) = x n for n 1.) x 1 x N +k x N +4 x 3

(A sequence also can be thought of as the list of function values attained for a function f :ℵ X, where f (n) = x n for n 1.) x 1 x N +k x N +4 x 3 MATH 337 Sequeces Dr. Neal, WKU Let X be a metric space with distace fuctio d. We shall defie the geeral cocept of sequece ad limit i a metric space, the apply the results i particular to some special

More information

Axioms of Measure Theory

Axioms of Measure Theory MATH 532 Axioms of Measure Theory Dr. Neal, WKU I. The Space Throughout the course, we shall let X deote a geeric o-empty set. I geeral, we shall ot assume that ay algebraic structure exists o X so that

More information

Lecture 27. Capacity of additive Gaussian noise channel and the sphere packing bound

Lecture 27. Capacity of additive Gaussian noise channel and the sphere packing bound Lecture 7 Ageda for the lecture Gaussia chael with average power costraits Capacity of additive Gaussia oise chael ad the sphere packig boud 7. Additive Gaussia oise chael Up to this poit, we have bee

More information

It is always the case that unions, intersections, complements, and set differences are preserved by the inverse image of a function.

It is always the case that unions, intersections, complements, and set differences are preserved by the inverse image of a function. MATH 532 Measurable Fuctios Dr. Neal, WKU Throughout, let ( X, F, µ) be a measure space ad let (!, F, P ) deote the special case of a probability space. We shall ow begi to study real-valued fuctios defied

More information

Problem Set 2 Solutions

Problem Set 2 Solutions CS271 Radomess & Computatio, Sprig 2018 Problem Set 2 Solutios Poit totals are i the margi; the maximum total umber of poits was 52. 1. Probabilistic method for domiatig sets 6pts Pick a radom subset S

More information

An Introduction to Randomized Algorithms

An Introduction to Randomized Algorithms A Itroductio to Radomized Algorithms The focus of this lecture is to study a radomized algorithm for quick sort, aalyze it usig probabilistic recurrece relatios, ad also provide more geeral tools for aalysis

More information

Infinite Sequences and Series

Infinite Sequences and Series Chapter 6 Ifiite Sequeces ad Series 6.1 Ifiite Sequeces 6.1.1 Elemetary Cocepts Simply speakig, a sequece is a ordered list of umbers writte: {a 1, a 2, a 3,...a, a +1,...} where the elemets a i represet

More information

HOMEWORK 2 SOLUTIONS

HOMEWORK 2 SOLUTIONS HOMEWORK SOLUTIONS CSE 55 RANDOMIZED AND APPROXIMATION ALGORITHMS 1. Questio 1. a) The larger the value of k is, the smaller the expected umber of days util we get all the coupos we eed. I fact if = k

More information

Chapter 6 Infinite Series

Chapter 6 Infinite Series Chapter 6 Ifiite Series I the previous chapter we cosidered itegrals which were improper i the sese that the iterval of itegratio was ubouded. I this chapter we are goig to discuss a topic which is somewhat

More information

The Boolean Ring of Intervals

The Boolean Ring of Intervals MATH 532 Lebesgue Measure Dr. Neal, WKU We ow shall apply the results obtaied about outer measure to the legth measure o the real lie. Throughout, our space X will be the set of real umbers R. Whe ecessary,

More information

Shannon s noiseless coding theorem

Shannon s noiseless coding theorem 18.310 lecture otes May 4, 2015 Shao s oiseless codig theorem Lecturer: Michel Goemas I these otes we discuss Shao s oiseless codig theorem, which is oe of the foudig results of the field of iformatio

More information

UC Berkeley CS 170: Efficient Algorithms and Intractable Problems Handout 17 Lecturer: David Wagner April 3, Notes 17 for CS 170

UC Berkeley CS 170: Efficient Algorithms and Intractable Problems Handout 17 Lecturer: David Wagner April 3, Notes 17 for CS 170 UC Berkeley CS 170: Efficiet Algorithms ad Itractable Problems Hadout 17 Lecturer: David Wager April 3, 2003 Notes 17 for CS 170 1 The Lempel-Ziv algorithm There is a sese i which the Huffma codig was

More information

Randomized Algorithms I, Spring 2018, Department of Computer Science, University of Helsinki Homework 1: Solutions (Discussed January 25, 2018)

Randomized Algorithms I, Spring 2018, Department of Computer Science, University of Helsinki Homework 1: Solutions (Discussed January 25, 2018) Radomized Algorithms I, Sprig 08, Departmet of Computer Sciece, Uiversity of Helsiki Homework : Solutios Discussed Jauary 5, 08). Exercise.: Cosider the followig balls-ad-bi game. We start with oe black

More information

Convergence of random variables. (telegram style notes) P.J.C. Spreij

Convergence of random variables. (telegram style notes) P.J.C. Spreij Covergece of radom variables (telegram style otes).j.c. Spreij this versio: September 6, 2005 Itroductio As we kow, radom variables are by defiitio measurable fuctios o some uderlyig measurable space

More information

Advanced Stochastic Processes.

Advanced Stochastic Processes. Advaced Stochastic Processes. David Gamarik LECTURE 2 Radom variables ad measurable fuctios. Strog Law of Large Numbers (SLLN). Scary stuff cotiued... Outlie of Lecture Radom variables ad measurable fuctios.

More information

62. Power series Definition 16. (Power series) Given a sequence {c n }, the series. c n x n = c 0 + c 1 x + c 2 x 2 + c 3 x 3 +

62. Power series Definition 16. (Power series) Given a sequence {c n }, the series. c n x n = c 0 + c 1 x + c 2 x 2 + c 3 x 3 + 62. Power series Defiitio 16. (Power series) Give a sequece {c }, the series c x = c 0 + c 1 x + c 2 x 2 + c 3 x 3 + is called a power series i the variable x. The umbers c are called the coefficiets of

More information

6.3 Testing Series With Positive Terms

6.3 Testing Series With Positive Terms 6.3. TESTING SERIES WITH POSITIVE TERMS 307 6.3 Testig Series With Positive Terms 6.3. Review of what is kow up to ow I theory, testig a series a i for covergece amouts to fidig the i= sequece of partial

More information

Lecture 7: October 18, 2017

Lecture 7: October 18, 2017 Iformatio ad Codig Theory Autum 207 Lecturer: Madhur Tulsiai Lecture 7: October 8, 207 Biary hypothesis testig I this lecture, we apply the tools developed i the past few lectures to uderstad the problem

More information

Lecture 11: Channel Coding Theorem: Converse Part

Lecture 11: Channel Coding Theorem: Converse Part EE376A/STATS376A Iformatio Theory Lecture - 02/3/208 Lecture : Chael Codig Theorem: Coverse Part Lecturer: Tsachy Weissma Scribe: Erdem Bıyık I this lecture, we will cotiue our discussio o chael codig

More information

Chapter 3. Strong convergence. 3.1 Definition of almost sure convergence

Chapter 3. Strong convergence. 3.1 Definition of almost sure convergence Chapter 3 Strog covergece As poited out i the Chapter 2, there are multiple ways to defie the otio of covergece of a sequece of radom variables. That chapter defied covergece i probability, covergece i

More information

Lecture 11: Pseudorandom functions

Lecture 11: Pseudorandom functions COM S 6830 Cryptography Oct 1, 2009 Istructor: Rafael Pass 1 Recap Lecture 11: Pseudoradom fuctios Scribe: Stefao Ermo Defiitio 1 (Ge, Ec, Dec) is a sigle message secure ecryptio scheme if for all uppt

More information

Sequences and Series of Functions

Sequences and Series of Functions Chapter 6 Sequeces ad Series of Fuctios 6.1. Covergece of a Sequece of Fuctios Poitwise Covergece. Defiitio 6.1. Let, for each N, fuctio f : A R be defied. If, for each x A, the sequece (f (x)) coverges

More information

Chapter 0. Review of set theory. 0.1 Sets

Chapter 0. Review of set theory. 0.1 Sets Chapter 0 Review of set theory Set theory plays a cetral role i the theory of probability. Thus, we will ope this course with a quick review of those otios of set theory which will be used repeatedly.

More information

Lecture 2. The Lovász Local Lemma

Lecture 2. The Lovász Local Lemma Staford Uiversity Sprig 208 Math 233A: No-costructive methods i combiatorics Istructor: Ja Vodrák Lecture date: Jauary 0, 208 Origial scribe: Apoorva Khare Lecture 2. The Lovász Local Lemma 2. Itroductio

More information

If a subset E of R contains no open interval, is it of zero measure? For instance, is the set of irrationals in [0, 1] is of measure zero?

If a subset E of R contains no open interval, is it of zero measure? For instance, is the set of irrationals in [0, 1] is of measure zero? 2 Lebesgue Measure I Chapter 1 we defied the cocept of a set of measure zero, ad we have observed that every coutable set is of measure zero. Here are some atural questios: If a subset E of R cotais a

More information

10-704: Information Processing and Learning Spring Lecture 10: Feb 12

10-704: Information Processing and Learning Spring Lecture 10: Feb 12 10-704: Iformatio Processig ad Learig Sprig 2015 Lecture 10: Feb 12 Lecturer: Akshay Krishamurthy Scribe: Dea Asta, Kirthevasa Kadasamy Disclaimer: These otes have ot bee subjected to the usual scrutiy

More information

1. Universal v.s. non-universal: know the source distribution or not.

1. Universal v.s. non-universal: know the source distribution or not. 28. Radom umber geerators Let s play the followig game: Give a stream of Ber( p) bits, with ukow p, we wat to tur them ito pure radom bits, i.e., idepedet fair coi flips Ber( / 2 ). Our goal is to fid

More information

Solutions to Math 347 Practice Problems for the final

Solutions to Math 347 Practice Problems for the final Solutios to Math 347 Practice Problems for the fial 1) True or False: a) There exist itegers x,y such that 50x + 76y = 6. True: the gcd of 50 ad 76 is, ad 6 is a multiple of. b) The ifiimum of a set is

More information

Week 5-6: The Binomial Coefficients

Week 5-6: The Binomial Coefficients Wee 5-6: The Biomial Coefficiets March 6, 2018 1 Pascal Formula Theorem 11 (Pascal s Formula For itegers ad such that 1, ( ( ( 1 1 + 1 The umbers ( 2 ( 1 2 ( 2 are triagle umbers, that is, The petago umbers

More information

ACO Comprehensive Exam 9 October 2007 Student code A. 1. Graph Theory

ACO Comprehensive Exam 9 October 2007 Student code A. 1. Graph Theory 1. Graph Theory Prove that there exist o simple plaar triagulatio T ad two distict adjacet vertices x, y V (T ) such that x ad y are the oly vertices of T of odd degree. Do ot use the Four-Color Theorem.

More information

Increasing timing capacity using packet coloring

Increasing timing capacity using packet coloring 003 Coferece o Iformatio Scieces ad Systems, The Johs Hopkis Uiversity, March 4, 003 Icreasig timig capacity usig packet colorig Xi Liu ad R Srikat[] Coordiated Sciece Laboratory Uiversity of Illiois e-mail:

More information

Generalized Semi- Markov Processes (GSMP)

Generalized Semi- Markov Processes (GSMP) Geeralized Semi- Markov Processes (GSMP) Summary Some Defiitios Markov ad Semi-Markov Processes The Poisso Process Properties of the Poisso Process Iterarrival times Memoryless property ad the residual

More information

Sequences. Notation. Convergence of a Sequence

Sequences. Notation. Convergence of a Sequence Sequeces A sequece is essetially just a list. Defiitio (Sequece of Real Numbers). A sequece of real umbers is a fuctio Z (, ) R for some real umber. Do t let the descriptio of the domai cofuse you; it

More information

MAT1026 Calculus II Basic Convergence Tests for Series

MAT1026 Calculus II Basic Convergence Tests for Series MAT026 Calculus II Basic Covergece Tests for Series Egi MERMUT 202.03.08 Dokuz Eylül Uiversity Faculty of Sciece Departmet of Mathematics İzmir/TURKEY Cotets Mootoe Covergece Theorem 2 2 Series of Real

More information

Entropy Rates and Asymptotic Equipartition

Entropy Rates and Asymptotic Equipartition Chapter 29 Etropy Rates ad Asymptotic Equipartitio Sectio 29. itroduces the etropy rate the asymptotic etropy per time-step of a stochastic process ad shows that it is well-defied; ad similarly for iformatio,

More information

Lecture 9: Expanders Part 2, Extractors

Lecture 9: Expanders Part 2, Extractors Lecture 9: Expaders Part, Extractors Topics i Complexity Theory ad Pseudoradomess Sprig 013 Rutgers Uiversity Swastik Kopparty Scribes: Jaso Perry, Joh Kim I this lecture, we will discuss further the pseudoradomess

More information

Math 61CM - Solutions to homework 3

Math 61CM - Solutions to homework 3 Math 6CM - Solutios to homework 3 Cédric De Groote October 2 th, 208 Problem : Let F be a field, m 0 a fixed oegative iteger ad let V = {a 0 + a x + + a m x m a 0,, a m F} be the vector space cosistig

More information

REGRESSION WITH QUADRATIC LOSS

REGRESSION WITH QUADRATIC LOSS REGRESSION WITH QUADRATIC LOSS MAXIM RAGINSKY Regressio with quadratic loss is aother basic problem studied i statistical learig theory. We have a radom couple Z = X, Y ), where, as before, X is a R d

More information

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 6 9/23/2013. Brownian motion. Introduction

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 6 9/23/2013. Brownian motion. Introduction MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/5.070J Fall 203 Lecture 6 9/23/203 Browia motio. Itroductio Cotet.. A heuristic costructio of a Browia motio from a radom walk. 2. Defiitio ad basic properties

More information

Entropy and Ergodic Theory Lecture 5: Joint typicality and conditional AEP

Entropy and Ergodic Theory Lecture 5: Joint typicality and conditional AEP Etropy ad Ergodic Theory Lecture 5: Joit typicality ad coditioal AEP 1 Notatio: from RVs back to distributios Let (Ω, F, P) be a probability space, ad let X ad Y be A- ad B-valued discrete RVs, respectively.

More information

Regression with quadratic loss

Regression with quadratic loss Regressio with quadratic loss Maxim Ragisky October 13, 2015 Regressio with quadratic loss is aother basic problem studied i statistical learig theory. We have a radom couple Z = X,Y, where, as before,

More information

Basics of Probability Theory (for Theory of Computation courses)

Basics of Probability Theory (for Theory of Computation courses) Basics of Probability Theory (for Theory of Computatio courses) Oded Goldreich Departmet of Computer Sciece Weizma Istitute of Sciece Rehovot, Israel. oded.goldreich@weizma.ac.il November 24, 2008 Preface.

More information

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 19 11/17/2008 LAWS OF LARGE NUMBERS II THE STRONG LAW OF LARGE NUMBERS

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 19 11/17/2008 LAWS OF LARGE NUMBERS II THE STRONG LAW OF LARGE NUMBERS MASSACHUSTTS INSTITUT OF TCHNOLOGY 6.436J/5.085J Fall 2008 Lecture 9 /7/2008 LAWS OF LARG NUMBRS II Cotets. The strog law of large umbers 2. The Cheroff boud TH STRONG LAW OF LARG NUMBRS While the weak

More information

Sequences A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence

Sequences A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence Sequeces A sequece of umbers is a fuctio whose domai is the positive itegers. We ca see that the sequece 1, 1, 2, 2, 3, 3,... is a fuctio from the positive itegers whe we write the first sequece elemet

More information

Injections, Surjections, and the Pigeonhole Principle

Injections, Surjections, and the Pigeonhole Principle Ijectios, Surjectios, ad the Pigeohole Priciple 1 (10 poits Here we will come up with a sloppy boud o the umber of parethesisestigs (a (5 poits Describe a ijectio from the set of possible ways to est pairs

More information

Simple Polygons of Maximum Perimeter Contained in a Unit Disk

Simple Polygons of Maximum Perimeter Contained in a Unit Disk Discrete Comput Geom (009) 1: 08 15 DOI 10.1007/s005-008-9093-7 Simple Polygos of Maximum Perimeter Cotaied i a Uit Disk Charles Audet Pierre Hase Frédéric Messie Received: 18 September 007 / Revised:

More information

Sieve Estimators: Consistency and Rates of Convergence

Sieve Estimators: Consistency and Rates of Convergence EECS 598: Statistical Learig Theory, Witer 2014 Topic 6 Sieve Estimators: Cosistecy ad Rates of Covergece Lecturer: Clayto Scott Scribe: Julia Katz-Samuels, Brado Oselio, Pi-Yu Che Disclaimer: These otes

More information

Entropies & Information Theory

Entropies & Information Theory Etropies & Iformatio Theory LECTURE I Nilajaa Datta Uiversity of Cambridge,U.K. For more details: see lecture otes (Lecture 1- Lecture 5) o http://www.qi.damtp.cam.ac.uk/ode/223 Quatum Iformatio Theory

More information

CHAPTER 10 INFINITE SEQUENCES AND SERIES

CHAPTER 10 INFINITE SEQUENCES AND SERIES CHAPTER 10 INFINITE SEQUENCES AND SERIES 10.1 Sequeces 10.2 Ifiite Series 10.3 The Itegral Tests 10.4 Compariso Tests 10.5 The Ratio ad Root Tests 10.6 Alteratig Series: Absolute ad Coditioal Covergece

More information

6.867 Machine learning, lecture 7 (Jaakkola) 1

6.867 Machine learning, lecture 7 (Jaakkola) 1 6.867 Machie learig, lecture 7 (Jaakkola) 1 Lecture topics: Kerel form of liear regressio Kerels, examples, costructio, properties Liear regressio ad kerels Cosider a slightly simpler model where we omit

More information

The multiplicative structure of finite field and a construction of LRC

The multiplicative structure of finite field and a construction of LRC IERG6120 Codig for Distributed Storage Systems Lecture 8-06/10/2016 The multiplicative structure of fiite field ad a costructio of LRC Lecturer: Keeth Shum Scribe: Zhouyi Hu Notatios: We use the otatio

More information

Fall 2013 MTH431/531 Real analysis Section Notes

Fall 2013 MTH431/531 Real analysis Section Notes Fall 013 MTH431/531 Real aalysis Sectio 8.1-8. Notes Yi Su 013.11.1 1. Defiitio of uiform covergece. We look at a sequece of fuctios f (x) ad study the coverget property. Notice we have two parameters

More information

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 21 11/27/2013

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 21 11/27/2013 MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 21 11/27/2013 Fuctioal Law of Large Numbers. Costructio of the Wieer Measure Cotet. 1. Additioal techical results o weak covergece

More information

On Random Line Segments in the Unit Square

On Random Line Segments in the Unit Square O Radom Lie Segmets i the Uit Square Thomas A. Courtade Departmet of Electrical Egieerig Uiversity of Califoria Los Ageles, Califoria 90095 Email: tacourta@ee.ucla.edu I. INTRODUCTION Let Q = [0, 1] [0,

More information

Solution. 1 Solutions of Homework 1. Sangchul Lee. October 27, Problem 1.1

Solution. 1 Solutions of Homework 1. Sangchul Lee. October 27, Problem 1.1 Solutio Sagchul Lee October 7, 017 1 Solutios of Homework 1 Problem 1.1 Let Ω,F,P) be a probability space. Show that if {A : N} F such that A := lim A exists, the PA) = lim PA ). Proof. Usig the cotiuity

More information

STAT Homework 1 - Solutions

STAT Homework 1 - Solutions STAT-36700 Homework 1 - Solutios Fall 018 September 11, 018 This cotais solutios for Homework 1. Please ote that we have icluded several additioal commets ad approaches to the problems to give you better

More information

Definition 4.2. (a) A sequence {x n } in a Banach space X is a basis for X if. unique scalars a n (x) such that x = n. a n (x) x n. (4.

Definition 4.2. (a) A sequence {x n } in a Banach space X is a basis for X if. unique scalars a n (x) such that x = n. a n (x) x n. (4. 4. BASES I BAACH SPACES 39 4. BASES I BAACH SPACES Sice a Baach space X is a vector space, it must possess a Hamel, or vector space, basis, i.e., a subset {x γ } γ Γ whose fiite liear spa is all of X ad

More information

Lecture 6: Source coding, Typicality, and Noisy channels and capacity

Lecture 6: Source coding, Typicality, and Noisy channels and capacity 15-859: Iformatio Theory ad Applicatios i TCS CMU: Sprig 2013 Lecture 6: Source codig, Typicality, ad Noisy chaels ad capacity Jauary 31, 2013 Lecturer: Mahdi Cheraghchi Scribe: Togbo Huag 1 Recap Uiversal

More information

1 Introduction. 1.1 Notation and Terminology

1 Introduction. 1.1 Notation and Terminology 1 Itroductio You have already leared some cocepts of calculus such as limit of a sequece, limit, cotiuity, derivative, ad itegral of a fuctio etc. Real Aalysis studies them more rigorously usig a laguage

More information

Rademacher Complexity

Rademacher Complexity EECS 598: Statistical Learig Theory, Witer 204 Topic 0 Rademacher Complexity Lecturer: Clayto Scott Scribe: Ya Deg, Kevi Moo Disclaimer: These otes have ot bee subjected to the usual scrutiy reserved for

More information

Advanced Analysis. Min Yan Department of Mathematics Hong Kong University of Science and Technology

Advanced Analysis. Min Yan Department of Mathematics Hong Kong University of Science and Technology Advaced Aalysis Mi Ya Departmet of Mathematics Hog Kog Uiversity of Sciece ad Techology September 3, 009 Cotets Limit ad Cotiuity 7 Limit of Sequece 8 Defiitio 8 Property 3 3 Ifiity ad Ifiitesimal 8 4

More information

7.1 Convergence of sequences of random variables

7.1 Convergence of sequences of random variables Chapter 7 Limit Theorems Throughout this sectio we will assume a probability space (, F, P), i which is defied a ifiite sequece of radom variables (X ) ad a radom variable X. The fact that for every ifiite

More information

Integrable Functions. { f n } is called a determining sequence for f. If f is integrable with respect to, then f d does exist as a finite real number

Integrable Functions. { f n } is called a determining sequence for f. If f is integrable with respect to, then f d does exist as a finite real number MATH 532 Itegrable Fuctios Dr. Neal, WKU We ow shall defie what it meas for a measurable fuctio to be itegrable, show that all itegral properties of simple fuctios still hold, ad the give some coditios

More information

Measure and Measurable Functions

Measure and Measurable Functions 3 Measure ad Measurable Fuctios 3.1 Measure o a Arbitrary σ-algebra Recall from Chapter 2 that the set M of all Lebesgue measurable sets has the followig properties: R M, E M implies E c M, E M for N implies

More information

Davenport-Schinzel Sequences and their Geometric Applications

Davenport-Schinzel Sequences and their Geometric Applications Advaced Computatioal Geometry Sprig 2004 Daveport-Schizel Sequeces ad their Geometric Applicatios Prof. Joseph Mitchell Scribe: Mohit Gupta 1 Overview I this lecture, we itroduce the cocept of Daveport-Schizel

More information

CS284A: Representations and Algorithms in Molecular Biology

CS284A: Representations and Algorithms in Molecular Biology CS284A: Represetatios ad Algorithms i Molecular Biology Scribe Notes o Lectures 3 & 4: Motif Discovery via Eumeratio & Motif Represetatio Usig Positio Weight Matrix Joshua Gervi Based o presetatios by

More information

Lecture 12: November 13, 2018

Lecture 12: November 13, 2018 Mathematical Toolkit Autum 2018 Lecturer: Madhur Tulsiai Lecture 12: November 13, 2018 1 Radomized polyomial idetity testig We will use our kowledge of coditioal probability to prove the followig lemma,

More information

lim za n n = z lim a n n.

lim za n n = z lim a n n. Lecture 6 Sequeces ad Series Defiitio 1 By a sequece i a set A, we mea a mappig f : N A. It is customary to deote a sequece f by {s } where, s := f(). A sequece {z } of (complex) umbers is said to be coverget

More information

Singular Continuous Measures by Michael Pejic 5/14/10

Singular Continuous Measures by Michael Pejic 5/14/10 Sigular Cotiuous Measures by Michael Peic 5/4/0 Prelimiaries Give a set X, a σ-algebra o X is a collectio of subsets of X that cotais X ad ad is closed uder complemetatio ad coutable uios hece, coutable

More information

M A T H F A L L CORRECTION. Algebra I 1 4 / 1 0 / U N I V E R S I T Y O F T O R O N T O

M A T H F A L L CORRECTION. Algebra I 1 4 / 1 0 / U N I V E R S I T Y O F T O R O N T O M A T H 2 4 0 F A L L 2 0 1 4 HOMEWORK ASSIGNMENT #4 CORRECTION Algebra I 1 4 / 1 0 / 2 0 1 4 U N I V E R S I T Y O F T O R O N T O P r o f e s s o r : D r o r B a r - N a t a Correctio Homework Assigmet

More information

A Proof of Birkhoff s Ergodic Theorem

A Proof of Birkhoff s Ergodic Theorem A Proof of Birkhoff s Ergodic Theorem Joseph Hora September 2, 205 Itroductio I Fall 203, I was learig the basics of ergodic theory, ad I came across this theorem. Oe of my supervisors, Athoy Quas, showed

More information

MA131 - Analysis 1. Workbook 3 Sequences II

MA131 - Analysis 1. Workbook 3 Sequences II MA3 - Aalysis Workbook 3 Sequeces II Autum 2004 Cotets 2.8 Coverget Sequeces........................ 2.9 Algebra of Limits......................... 2 2.0 Further Useful Results........................

More information

Assignment 5: Solutions

Assignment 5: Solutions McGill Uiversity Departmet of Mathematics ad Statistics MATH 54 Aalysis, Fall 05 Assigmet 5: Solutios. Let y be a ubouded sequece of positive umbers satisfyig y + > y for all N. Let x be aother sequece

More information

# fixed points of g. Tree to string. Repeatedly select the leaf with the smallest label, write down the label of its neighbour and remove the leaf.

# fixed points of g. Tree to string. Repeatedly select the leaf with the smallest label, write down the label of its neighbour and remove the leaf. Combiatorics Graph Theory Coutig labelled ad ulabelled graphs There are 2 ( 2) labelled graphs of order. The ulabelled graphs of order correspod to orbits of the actio of S o the set of labelled graphs.

More information

Resolution Proofs of Generalized Pigeonhole Principles

Resolution Proofs of Generalized Pigeonhole Principles Resolutio Proofs of Geeralized Pigeohole Priciples Samuel R. Buss Departmet of Mathematics Uiversity of Califoria, Berkeley Győrgy Turá Departmet of Mathematics, Statistics, ad Computer Sciece Uiversity

More information

Feedback in Iterative Algorithms

Feedback in Iterative Algorithms Feedback i Iterative Algorithms Charles Byre (Charles Byre@uml.edu), Departmet of Mathematical Scieces, Uiversity of Massachusetts Lowell, Lowell, MA 01854 October 17, 2005 Abstract Whe the oegative system

More information

Summary. Recap ... Last Lecture. Summary. Theorem

Summary. Recap ... Last Lecture. Summary. Theorem Last Lecture Biostatistics 602 - Statistical Iferece Lecture 23 Hyu Mi Kag April 11th, 2013 What is p-value? What is the advatage of p-value compared to hypothesis testig procedure with size α? How ca

More information

A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence

A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence Sequeces A sequece of umbers is a fuctio whose domai is the positive itegers. We ca see that the sequece,, 2, 2, 3, 3,... is a fuctio from the positive itegers whe we write the first sequece elemet as

More information

Math 475, Problem Set #12: Answers

Math 475, Problem Set #12: Answers Math 475, Problem Set #12: Aswers A. Chapter 8, problem 12, parts (b) ad (d). (b) S # (, 2) = 2 2, sice, from amog the 2 ways of puttig elemets ito 2 distiguishable boxes, exactly 2 of them result i oe

More information

The natural exponential function

The natural exponential function The atural expoetial fuctio Attila Máté Brookly College of the City Uiversity of New York December, 205 Cotets The atural expoetial fuctio for real x. Beroulli s iequality.....................................2

More information

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 3 9/11/2013. Large deviations Theory. Cramér s Theorem

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 3 9/11/2013. Large deviations Theory. Cramér s Theorem MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/5.070J Fall 203 Lecture 3 9//203 Large deviatios Theory. Cramér s Theorem Cotet.. Cramér s Theorem. 2. Rate fuctio ad properties. 3. Chage of measure techique.

More information

Series III. Chapter Alternating Series

Series III. Chapter Alternating Series Chapter 9 Series III With the exceptio of the Null Sequece Test, all the tests for series covergece ad divergece that we have cosidered so far have dealt oly with series of oegative terms. Series with

More information

First Year Quantitative Comp Exam Spring, Part I - 203A. f X (x) = 0 otherwise

First Year Quantitative Comp Exam Spring, Part I - 203A. f X (x) = 0 otherwise First Year Quatitative Comp Exam Sprig, 2012 Istructio: There are three parts. Aswer every questio i every part. Questio I-1 Part I - 203A A radom variable X is distributed with the margial desity: >

More information

LONG SNAKES IN POWERS OF THE COMPLETE GRAPH WITH AN ODD NUMBER OF VERTICES

LONG SNAKES IN POWERS OF THE COMPLETE GRAPH WITH AN ODD NUMBER OF VERTICES J Lodo Math Soc (2 50, (1994, 465 476 LONG SNAKES IN POWERS OF THE COMPLETE GRAPH WITH AN ODD NUMBER OF VERTICES Jerzy Wojciechowski Abstract I [5] Abbott ad Katchalski ask if there exists a costat c >

More information

w (1) ˆx w (1) x (1) /ρ and w (2) ˆx w (2) x (2) /ρ.

w (1) ˆx w (1) x (1) /ρ and w (2) ˆx w (2) x (2) /ρ. 2 5. Weighted umber of late jobs 5.1. Release dates ad due dates: maximimizig the weight of o-time jobs Oce we add release dates, miimizig the umber of late jobs becomes a sigificatly harder problem. For

More information

What is Probability?

What is Probability? Quatificatio of ucertaity. What is Probability? Mathematical model for thigs that occur radomly. Radom ot haphazard, do t kow what will happe o ay oe experimet, but has a log ru order. The cocept of probability

More information

Lecture 9: Hierarchy Theorems

Lecture 9: Hierarchy Theorems IAS/PCMI Summer Sessio 2000 Clay Mathematics Udergraduate Program Basic Course o Computatioal Complexity Lecture 9: Hierarchy Theorems David Mix Barrigto ad Alexis Maciel July 27, 2000 Most of this lecture

More information

7 Sequences of real numbers

7 Sequences of real numbers 40 7 Sequeces of real umbers 7. Defiitios ad examples Defiitio 7... A sequece of real umbers is a real fuctio whose domai is the set N of atural umbers. Let s : N R be a sequece. The the values of s are

More information

Hashing and Amortization

Hashing and Amortization Lecture Hashig ad Amortizatio Supplemetal readig i CLRS: Chapter ; Chapter 7 itro; Sectio 7.. Arrays ad Hashig Arrays are very useful. The items i a array are statically addressed, so that isertig, deletig,

More information

1 Convergence in Probability and the Weak Law of Large Numbers

1 Convergence in Probability and the Weak Law of Large Numbers 36-752 Advaced Probability Overview Sprig 2018 8. Covergece Cocepts: i Probability, i L p ad Almost Surely Istructor: Alessadro Rialdo Associated readig: Sec 2.4, 2.5, ad 4.11 of Ash ad Doléas-Dade; Sec

More information

Lecture Overview. 2 Permutations and Combinations. n(n 1) (n (k 1)) = n(n 1) (n k + 1) =

Lecture Overview. 2 Permutations and Combinations. n(n 1) (n (k 1)) = n(n 1) (n k + 1) = COMPSCI 230: Discrete Mathematics for Computer Sciece April 8, 2019 Lecturer: Debmalya Paigrahi Lecture 22 Scribe: Kevi Su 1 Overview I this lecture, we begi studyig the fudametals of coutig discrete objects.

More information

Hoggatt and King [lo] defined a complete sequence of natural numbers

Hoggatt and King [lo] defined a complete sequence of natural numbers REPRESENTATIONS OF N AS A SUM OF DISTINCT ELEMENTS FROM SPECIAL SEQUENCES DAVID A. KLARNER, Uiversity of Alberta, Edmoto, Caada 1. INTRODUCTION Let a, I deote a sequece of atural umbers which satisfies

More information

Math 104: Homework 2 solutions

Math 104: Homework 2 solutions Math 04: Homework solutios. A (0, ): Sice this is a ope iterval, the miimum is udefied, ad sice the set is ot bouded above, the maximum is also udefied. if A 0 ad sup A. B { m + : m, N}: This set does

More information

Information Theory Tutorial Communication over Channels with memory. Chi Zhang Department of Electrical Engineering University of Notre Dame

Information Theory Tutorial Communication over Channels with memory. Chi Zhang Department of Electrical Engineering University of Notre Dame Iformatio Theory Tutorial Commuicatio over Chaels with memory Chi Zhag Departmet of Electrical Egieerig Uiversity of Notre Dame Abstract A geeral capacity formula C = sup I(; Y ), which is correct for

More information

7.1 Convergence of sequences of random variables

7.1 Convergence of sequences of random variables Chapter 7 Limit theorems Throughout this sectio we will assume a probability space (Ω, F, P), i which is defied a ifiite sequece of radom variables (X ) ad a radom variable X. The fact that for every ifiite

More information

Sequences and Series

Sequences and Series Sequeces ad Series Sequeces of real umbers. Real umber system We are familiar with atural umbers ad to some extet the ratioal umbers. While fidig roots of algebraic equatios we see that ratioal umbers

More information