Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Refinement of Two Fundamental Tools in Information Theory Raymond W. Yeung Institute of Network Coding The Chinese University of Hong Kong Joint work with Siu Wai Ho and Sergio Verdu Discontinuity of Shannon’s Information Measures Shannon’s information measures: H(X), H(X|Y), I(X;Y) and I(X;Y|Z). They are described as continuous functions [Shannon 1948] [Csiszár & Körner 1981] [Cover & Thomas 1991] [McEliece 2002] [Yeung 2002]. All Shannon's information measures are indeed discontinuous everywhere when random variables take values from countably infinite alphabets [Ho & Yeung 2005]. e.g., X can be any positive integer. P.2 Discontinuity of Entropy Let PX = {1, 0, 0, ...} and 1 1 1 PXn 1 , , ,...,0,0,.... log n n log n n log n As n , we have i PX (i) PXn (i) However, 2 0. log n lim H X n . n P.3 Discontinuity of Entropy Theorem 1: For any c 0 and any X taking values from a countably infinite alphabet with H(X) < , PXn s.t. V PX , PXn i PX (i ) PXn (i ) 0 but H X n H X c H X n H X c H X 0 n P.4 Discontinuity of Entropy Theorem 2: For any c 0 and any X taking values from countably infinite alphabet with H(X) < , PX (i ) PXn s.t. D PX || PXn i PX (i ) log 0 PXn (i ) but H X n H X c H X n H X c H X 0 n P.5 Pinsker’s inequality 1 D(p || q) ³ V 2 (p, q) 2 ln 2 By Pinsker’s inequality, convergence w.r.t. D(× || ×) implies convergence w.r.t. V(×,×) . Therefore, Theorem 2 implies Theorem 1. P.6 Discontinuity of Entropy 1 1 1 4 1 2 4 P.7 Discontinuity of Shannon’s Information Measures Theorem 3: For any X, Y and Z taking values from countably infinite alphabet with I(X;Y|Z) < , PXnYnZn s.t. limn D PXYZ || PXnYnZn 0 but limn I X n ; Yn | Z n . P.8 Discontinuity of Shannon’s Information Measures Applications: channel coding theorem lossless/lossy source coding theorems, etc. Fano’s Inequality Typicality Shannon’s Information Measures P.9 To Find the Capacity of a Communication Channel Alice Bob Channel Capacity C1 Typicality Capacity C2 Fano’s Inequality P.10 On Countably Infinite Alphabet Applications: channel coding theorem lossless/lossy source coding theorems, etc. Fano’s Inequality Typicality Shannon’s Information Measures discontinuous! P.11 Typicality Weak typicality was first introduced by Shannon [1948] to establish the source coding theorem. Strong typicality was first used by Wolfowitz [1964] and then by Berger [1978]. It was further developed into the method of types by Csiszár and Körner [1981]. Strong typicality possesses stronger properties compared with weak typicality. It can be used only for random variables with finite alphabet. 12 Notations Consider an i.i.d. source {Xk, k 1}, where Xk taking values from a countable alphabet X . Let PX PXk for all k. Assume H(PX) < . Let X = (X1, X2, …, Xn) For a sequence x = (x1, x2, …, xn) X n , N(x; x) is the number of occurrences of x in x q(x; x) = n-1N(x; x) and QX = {q(x; x)} is the empirical distribution of x e.g., x = (1, 3, 2, 1, 1). N(1; x) = 3, N(2; x) = N(3; x) =1 QX = {3/5, 1/5, 1/5}. 13 Weak Typicality Definition (Weak typicality): For any e > 0, the weakly typical set Wn[X]e with respect to PX is the set of sequences x = (x1, x2, …, xn) X n such that 14 Weak Typicality Definition 1 (Weak typicality): For any e > 0, the weakly typical set Wn[X]e with respect to PX is the set of sequences x = (x1, x2, …, xn) X n such that | D(QX || PX )+ H(QX )- H(PX ) |£ e Note that H(QX ) = -åQX (x)logQX (x) x while åQ Empirical entropy = - X (x)log PX (x) x 15 Asymptotic Equipartition Property Theorem 4 (Weak AEP): For any e > 0: 1) If x Wn[X]e , then 2 n ( H ( X ) e ) p ( x) 2 n ( H ( X ) e ) 2) For sufficiently large n, Pr X W[nX ]e 1 e 3) For sufficiently large n, (1 e )2n( H ( X ) e ) W[nX ]e 2n( H ( X ) e ) 16 Illustration of AEP X n − Set of all n-sequences Typical Set of n-sequences: Prob. ≈ 1 ≈ Uniform distribution 17 Strong Typicality Strong typicality has been defined in slightly different forms in the literature. Definition 2 (Strong typicality): For |X | < and any d > 0, the strongly typical set Tn[X]d with respect to PX is the set of sequences x = (x1, x2, …, xn) X n such that V ( PX , QX ) x | PX ( x) q( x; x) | d the variational distance between the empirical distribution of the sequence x and PX is small. 18 Asymptotic Equipartition Property Theorem 5 (Strong AEP): For a finite alphabet X and any d > 0: 1) If x Tn[X]d , then n ( H ( X ) d ) 2 p(x) 2 n( H ( X ) d ) 2) For sufficiently large n, Pr X T[nX ]d 1 d 3) For sufficiently large n, (1- d )2n(H ( X )-g ) £ T[ nX ]d £ 2 n(H (X )+g ) 19 Breakdown of Strong AEP If strong typicality is extended (in the natural way) to countably infinite alphabets, strong AEP no longer holds Specifically, Property 2 holds but Properties 1 and 3 do not hold. P.20 Typicality X n finite alphabet Weak Typicality: | D(QX || PX ) H (QX ) H ( PX ) | e Strong Typicality: V ( PX , QX ) d 21 Unified Typicality X n countably infinite alphabet Weak Typicality: | D(QX || PX ) H (QX ) H ( PX ) | e Strong Typicality: V ( PX , QX ) d x s.t. D(QX || PX ) is small but | H (QX ) H ( PX ) | is large 22 Unified Typicality X n countably infinite alphabet Weak Typicality: | D(QX || PX ) H (QX ) H ( PX ) | e Strong Typicality: V ( PX , QX ) d Unified Typicality: D(QX || PX ) | H (QX ) H ( PX ) | . 23 Unified Typicality Definition 3 (Unified typicality): For any > 0, the unified typical set Un[X] with respect to PX is the set of sequences x = (x1, x2, …, xn) X n such that D(QX || PX )+ | H(QX ) - H(PX ) |£ h Weak Typicality: | D(Q || P )+ H(Q )- H(P ) |£ e X X X X Strong Typicality: V(PX ,QX ) £ d Each typicality corresponds to a “distance measure” Entropy is continuous w.r.t. the distance measure induced by unified typicality 24 Asymptotic Equipartition Property Theorem 6 (Unified AEP): For any > 0: 1) If x Un[X] , then 2 n( H ( X ) ) p(x) 2 n( H ( X ) ) 2) For sufficiently large n, Pr X U[nX ] 1 3) For sufficiently large n, (1- h )2 n(H (X )-m ) £ U[nX ]h £ 2n(H ( X )+m ) 25 Unified Typicality Theorem 7: For any x X n, if x Un[X] , then x Wn[X]e and x Tn[X]d , where e and d 2 ln 2. 26 Unified Jointly Typicality Consider a bivariate information source {(Xk, Yk), k 1} where (Xk, Yk) are i.i.d. with generic distribution PXY . We use (X, Y) to denote the pair of generic random variables. Let (X, Y) = ((X1, Y1), (X2, Y2), …, (Xn, Yn)). For the pair of sequence (x, y), the empirical distribution is QXY = {q(x,y; x,y)} where q(x,y; x,y) = n-1N(x,y; x,y). 27 Unified Jointly Typicality Definition 4 (Unified jointly typicality): For any > 0, the unified typical set Un[XY] with respect to PXY is the set of sequences (x, y) X nY n such that D(Q XY || PXY ) | H (Q XY ) H ( PXY ) | | H (Q X ) H ( PX ) | | H (QY ) H ( PY ) | . This definition cannot be simplified. 28 Conditional AEP Definition 5: For any x Un[X] , the conditional typical set of Y is defined as U[nY | X ] (x) y U[nY ] : (x, y ) U[nXY ] Theorem 8: For x Un[X], if U[nY | X ] 1, then 2n( H (Y | X ) ) U[nY | X ] 2n( H (Y | X ) ) where 0 as 0 and then n 29 Illustration of Conditional AEP 2 2nH(X) x S[nX] nH(Y) y S[nY] . . . . . . . . . . . . . . . 2nH(X,Y) n (x,y) T[XY ] P.30 Applications Rate-distortion theory A version of rate-distortion theorem was proved by strong typicality [Cover & Thomas 1991][Yeung 2008] It can be easily generalized to countably infinite alphabet Multi-source network coding The achievable information rate region in multisource network coding problem was proved by strong typicality [Yeung 2008] It can be easily generalized to countably infinite alphabet 31 Fano’s Inequality Fano's inequality: For discrete random variables X and Y taking values on the same alphabet X = {1, 2, }, let e P[ X Y ] 1 PXY ( w, w) wX Then H ( X | Y ) e log(| X | 1) h(e ), where 1 1 h( x) x log (1 x) log x 1 x for 0 < x < 1 and h(0) = h(1) = 0. P.32 Motivation 1 H ( X | Y ) e log(| X | 1) h(e ) This upper bound on H(X | Y ) is not tight. For fixed e and | X | , can always find X such that H(X |Y ) £ H(X) < e log(| X | -1)+ h(e ) Then we can ask, for fixed PX and e, what is max PY | X :P[ X Y ]e H ( X | Y ) e log(| X | 1) h(e ) P.33 Motivation 2 If X is countably infinite, Fano’s inequality no longer gives an upper bound on H(X|Y). It is possible that H ( X | Y ) 0 as e 0 which can be explained by the discontinuity of entropy. PX 1 n 1 1 1 , ,..., and PYn 1,0,0,... log n n log n n log n 1 0 Then H(Xn|Yn) = H(Xn) but e n log n Under what conditions e 0 H(X|Y) 0 for countably infinite alphabets? P.34 Tight Upper Bound on H(X|Y) Theorem 9: Suppose e P[ X Y ] 1 PX (1) , then H ( X | Y ) e H (Q ( PX , e )) h(e ) where the right side is the tight bound dependent on eand PX. (This is the simplest of the 3 cases.) e q1 q2 q3 q4 Q ( PX , e ) {e 1q1, e 1q2 , e 1q3 ,} Let X (e ) e H (Q ( PX , e )) h(e ) P.35 Generalizing Fano’s Inequality Fano's inequality [Fano 1952] gives an upper bound on the conditional entropy H(X|Y) in terms of the error probability ePrX Y. e.g. PX = [0.4, 0.4, 0.1, 0.1] [Fano 1952] H(X|Y) [Ho & Verdú 2008] e P.36 Generalizing Fano’s Inequality e.g., X is a Poisson random variable with mean equal to 10. Fano's inequality no longer gives an upper bound on H(X|Y). H(X|Y) e P.37 Generalizing Fano’s Inequality e.g. X is a Poisson random variable with mean equal to 10. Fano's inequality no longer gives an upper bound on H(X|Y). H(X|Y) [Ho & Verdú 2008] e P.38 Joint Source-Channel Coding ( S1, S2 , Sk ) Encoder ( X 1 , X 2 , X n ) Channel ( Sˆ1, Sˆ2 , Sˆk ) Decoder (Y1, Y2 ,Yn ) k-to-n joint source-channel code P.39 Error Probabilities The average symbol error probability is defined as 1 k k P[ Si Sˆi ] k i 1 The block error probability is defined as k P[( S1, S 2 , S k ) ( Sˆ1, Sˆ2 , Sˆk )] P.40 Symbol Error Rate Theorem 10: For any discrete memoryless source and general channel, the rate of a k-to-n joint sourcechannel code with symbol error probability k satisfies sup 1 n n n I ( X ;Y ) n k 1 X k n k H ( S ) S * (k ) where S* is constructed from {S1, S2, ..., Sk} according to PS * (1) min j PSj (1), PS * (a) min j ia1 PSj (i ) ia11 PS * (i ) a 2. P.41 Block Error Rate Theorem 11: For any general discrete source and general channel, the block error probability k of a kto-n joint source-channel code is lower bounded by 1 S k H ( S k ) sup Xn I ( X ;Y ) k n n P.42 Information Theoretic Security 1 n n lim n I ( X ;Y ) 0 Weak secrecy n has been considered in [Csiszár & Körner 78, Broadcast channel] and some seminal papers. [Wyner 75, Wiretap channel I] only stated that “a large value of the equivocation implies a large value of Pew”, where the equivocation refers to n 1H ( X n | Y k ) and Pew means n . It is important to clarify what exactly weak secrecy implies. P.43 Weak Secrecy E.g., PX = (0.4, 0.4, 0.1, 0.1). H(X) H(X|Y) [Fano 1952] [Ho & Verdú 2008] e= P[X Y] P.44 Weak Secrecy Theorem 12: For any discrete stationary memoryless source (i.i.d. source) with distribution PX , if lim n 1I ( X n ; Y n ) 0, n Then lim n max and n lim n 1. n Remark: Weak Secrecy together with the stationary source assumption is insufficient to show the maximum error probability. The proof is based on the tight upper bound on H(X|Y) in terms of error probability. P.45 Summary Applications: channel coding theorem lossless/lossy source coding theorems Fano’s Inequality Typicality Weak Strong Typicality Typicality Shannon’s Information Measures P.46 On Countably Infinite Alphabet Applications: channel coding theorem lossless/lossy source coding theorem Typicality Weak Typicality Shannon’s Information Measures discontinuous! P.47 Unified Typicality Applications: channel coding theorem MSNC/lossy SC theorems Typicality Unified Typicality Shannon’s Information Measures P.48 Generalized Fano’s Inequality Applications: results on JSCC, IT security MSNC/lossy SC theorems Generalized Typicality Fano’s Inequality Unified Typicality Shannon’s Information Measures P.49 Perhaps... A lot of fundamental research in information theory are still waiting for us to investigate. P.50 References S.-W. Ho and R. W. Yeung, “On the Discontinuity of the Shannon Information Measures,” IEEE Trans. Inform. Theory, vol. 55, no. 12, pp. 5362–5374, Dec. 2009. S.-W. Ho and R. W. Yeung, “On Information Divergence Measures and a Unified Typicality,” IEEE Trans. Inform. Theory, vol. 56, no. 12, pp. 5893–5905, Dec. 2010. S.-W. Ho and S. Verdú, “On the Interplay between Conditional Entropy and Error Probability,” IEEE Trans. Inform. Theory, vol. 56, no. 12, pp. 5930–5942, Dec. 2010. S.-W. Ho, “On the Interplay between Shannon's Information Measures and Reliability Criteria,” in Proc. 2009 IEEE Int. Symposium Inform. Theory (ISIT 2009), Seoul, Korea, June 28-July 3, 2009. S.-W. Ho, “Bounds on the Rates of Joint Source-Channel Codes for General Sources and Channels,” in Proc. 2009 IEEE Inform. Theory Workshop (ITW 2009), Taormina, Italy, Oct. 11-16, 2009. P.51 Q&A P.52 Why Countably Infinite Alphabet? An important mathematical theory can provide some insights which cannot be obtained from other means. Problems involve random variables taking values from countably infinite alphabets. Finite alphabet is the special case. Benefits: tighter bounds, faster convergent rates, etc. In source coding, the alphabet size can be very large, infinite or unknown. P.53 Discontinuity of Entropy Entropy is a measure of uncertainty. We can be more and more sure that a particular event will happen as time goes, but at the same time, the uncertainty of the whole picture keeps on increasing. If one found the above statement counter-intuitive, he/she may have the concept that entropy is continuous rooted in his/her mind. The limiting probability distribution may not fully characterize the asymptotic behavior of a Markov chain. P.54 Discontinuity of Entropy Suppose a child hides in a shopping mall where the floor plan is shown in the next slide. In each case, the chance for him to hide in a room is directly proportional to the size of the room. We are only interested in which room the child locates in but not his exact position inside a room. Which case do you expect is the easiest to locate the child? P.55 Case A 1 blue room + 2 green rooms Case B 1 blue room + 16 green rooms Case C 1 blue room + 256 green rooms Case A Case B The chance in Case C Case D 0.698 0.742 0.5 0.622 0.25 0.0326 0.00118 0.000063 the blue room The chance in Case D 1 blue room + 4096 green rooms a green room P.56 Discontinuity of Entropy From Case A to Case D, the difficulty is increasing. By the Shannon entropy, the uncertainty is increasing although the probability of the child being in the blue room is also increasing. We can continue to construct this example and make the chance in the blue room approaching to 1! The critical assumption is that the number of rooms can be unbounded. So we have seen that “There is a very sure event” and “large uncertainty of the whole picture” can exist at the same time. Imagine there is a city where everyone has a normal life everyday with probability 0.99. With probability 0.01, however, any kind of accident that beyond our imagination can happen. Would you feel a big uncertainty about your life if you were living in that P.57 city? 1 n k lim n I ( X ; Y ) 0 n Weak secrecy is insufficient to show the maximum error probability. Example 1: Let W, V and Xi be binary random variables. Suppose W and V are independent and uniform. Let W V 0 Xi independen t and uniform V 1 ~ max 1 max x PX ( x) 0.5 ~ max n lim 1 max xn P Xn ( xn) 1 1 3 lim 1 2 2 4 n P.58 Example 1 Let Y1 Y2 Y3 Y4 = = = = X1 X4 X9 X16 X2 X3 X8 X15 X5 X6 X7 X14 ... 0 lim n 1I ( X n ; Y k ) n lim n 1 n 0 n X10 X11 X12 X13 (0,0,,0) if Yi 0 i Choose xˆ (1,1,,1) if Yi 1 i. Then 1 ~ 3 lim n n P[V 1] max 2 4 1 1 ~ 1 lim n n P[V 1] max 2 4 2 n P.59 Joint Unified Typicality D(Q XY || PXY ) | H (Q XY ) H ( PXY ) | | H (Q X ) H ( PX ) | | H (QY ) H ( PY ) | . Can be changed to Ans: X 1 D(Q XY || PXY ) | H (Q XY ) H ( PXY ) | | H (Q X ) H ( PX ) | . Y 3 X 2 2 Q = {q(xy)} ? Y 2 2 P = {p(xy)} D(Q||P) << 1 P.60 Joint Unified Typicality D(Q XY || PXY ) | H (Q XY ) H ( PXY ) | | H (Q X ) H ( PX ) | | H (QY ) H ( PY ) | . Can be changed to Ans: X 1 D(Q XY || PXY ) | H (Q X ) H ( PX ) | | H (QY ) H ( PY ) | . Y 3 X 1 2 Q = {q(xy)} ? Y 2 2 P = {p(xy)} D(Q||P) << 1 P.61 Asymptotic Equipartition Property Theorem 5 (Consistency): For any (x,y) X nY n, if (x,y) Un[XY] , then x Un[X] and y Un[Y] . Theorem 6 (Unified JAEP): For any > 0: 1) If (x, y) Un[XY] , then 2 n( H ( XY ) ) p(x, y ) 2 n( H ( XY ) ) 2) For sufficiently large n, Pr X, Y U[nXY ] 1 3) For sufficiently large n, (1 )2n( H ( XY ) ) U n n( H ( XY ) ) 2 [ XY ] P.62