Source Coding and Function Computation: Optimal Rate in Zero-Error and Vanishing Zero-Error Regime

Size: px
Start display at page:

Download "Source Coding and Function Computation: Optimal Rate in Zero-Error and Vanishing Zero-Error Regime"

Transcription

1 Source Coding and Function Computation: Optimal Rate in Zero-Error and Vanishing Zero-Error Regime Solmaz Torabi Dept. of Electrical and Computer Engineering Drexel University Advisor: Dr. John M. Walsh 1/73 1

2 References H. Witsenhausen, The zero-error side information problem and chromatic numbers (corresp.), Information Theory, IEEE Transactions on, vol. 22, no. 5, pp , N. Alon and A. Orlitsky, Source coding and graph entropies, Information Theory, IEEE Transactions on, vol. 42, no. 5, pp , A. Orlitsky and J. Roche, Coding for computing, Information Theory, IEEE Transactions on, vol. 47, no. 3, pp , V. Doshi, D. Shah, M. Médard, and M. Effros, Functional compression through graph coloring, Information Theory, IEEE Transactions on, vol. 56, no. 8, pp , G. Simonyi, Graph entropy: a survey, Combinatorial Optimization, vol. 20, pp , J. Cardinal, S. Fiorini, and G. Joret, Minimum entropy coloring, Journal of combinatorial optimization, vol. 16, no. 4, pp , /73 2

3 Outline Introduction Motivation Some graph theory quantities Zero error source coding (Scalar vs. Block coding) Fixed length coding Variable length coding Asymptotically zero probability of error Optimal rate for function compression Achievable coding scheme Function compression in zero-error regime Compression of specific functions 3/73 3

4 Outline Introduction Motivation Some graph theory quantities Zero-error source coding (Scalar vs. Block coding) Fixed length coding Variable length coding Asymptotically zero probability of error Optimal rate for function compression Achievable coding scheme Compression of specific functions Function compression in zero-error regime 4/73 4

5 Motivation In big data problems How to design computational methodologies suited to statistical calculation Scale easily as the data being processed becomes immense Fundamental problems in massive datasets where some statistical primitives needs to be computed for some tasks. Users submit machine learning tasks Analytics servers translate the tasks into statistical primitives calculation 5/73 5

6 Motivation Central receiver wishes to compute the average temperature of the building Bandwidth limited sensors do not communicate with each other. Considers the recovery not of the sources but the function of sources 6/73 6

7 Outline Introduction Motivation Some graph theory quantities Zero error source coding (Scalar vs. Block coding) Fixed length coding Variable length coding Asymptotically zero probability of error Optimal rate for function compression Achievable coding scheme Function compression in zero-error regime Compression of specific functions 7/73 7

8 Some graph theory quantities Quantities to be discussed for undirected graphs:(undirected, finite and contain neither loops nor multiple edges) Independent set of the graph Coloring of the graph Product of the graph 8/73 8

9 Independent set An independent set or stable set is a set of vertices in a graph, no two of which are adjacent An independent set that is not the subset of another independent set is called maximal f e b a c d 9/73 9

10 Independent set An independent set or stable set is a set of vertices in a graph, no two of which are adjacent. An independent set that is not the subset of another independent set is called maximal. f e b a c Independent set but NOT maximal independent set d 10/73 10

11 Independent set An independent set or stable set is a set of vertices in a graph, no two of which are adjacent. An independent set that is not the subset of another independent set is called maximal. f e b a c Independent set but NOT maximal independent set d 11/73 11

12 Independent set An independent set or stable set is a set of vertices in a graph, no two of which are adjacent An independent set that is not the subset of another independent set is called maximal f e b a c Maximal independent set d 12/73 12

13 Independent set An independent set or stable set is a set of vertices in a graph, no two of which are adjacent. An independent set that is not the subset of another independent set is called maximal. f e b a c n (G) = {f,d,e}{b, d, e} o {c, f, e}{a, d, f} d 13/73 13

14 Probabilistic Graph-Chromatic Number χ(g) is the smallest number of colors needed to color the vertices of G so that no two adjacent vertices share the same color. The chromatic number of a graph clique number(maximum clique size) f b a e f b a e c 0.05 c (G) =3 d d 14/73 14

15 Entropy of the coloring Random variable X P(x), and function c : X N Entropy of the coloring is defined as H(c(X )) = p i = P(c 1 (i)) = i c(x ) p i log p i x V :c(x)=γ P(x) c 1 (i) is a color class (a set of vertices assigned the same color i) Coloring of G is any function c over V such that c 1 (.) induced a partition of V into independent sets of G H X (G, X ) = min{h(c(x )) : c is a coloring of G} 15/73 15

16 Chromatic Entropy Note that χ H(G,P) χ G f e f e f e b 0.05 a 0.05 b a b a c 0.05 c c d d d H(0.333, 0.333, 0.333) = 1.58 H(0.616, 0.333, 0.05) = 1.17 H(0.85, 0.05, 0.05, 0.05) = /73 16

17 Chromatic Entropy Chromatic entropy is known to be NP-hard 1 Find independent sets with high probability mass and establish these as color classes, Unbalance the distribution as much as possible Add a node to the color class with the highest probability if possible. (concavity of f (x) = x log(x) ) If color classes C 1 and C 2 can be merged properly, C = {C 1 C 2, C 3,..., C k } has strictly lower entropy than the original coloring C. 1 J. Cardinal, S. Fiorini, and G. Joret, Minimum entropy coloring, in Algorithms and Computation 17/73 17

18 AND power graph vs OR power graph AND-product graph of G 1 (X 1, E 1 ) and G 2 (X 2, E 2 ) G 1 G 2 has vertex set X 1 X 2 (x 1, x 2) and (x 1, x 2) are connected if {either x i = x i connected to x i in graph G i } for each i = 1, 2 or x i is OR-product graph of G 1 (X 1, E 1 ) and G 2 (X 2, E 2 ) G 1 G 2 has vertex set X 1 X 2 (x 1, x 2) and (x 1, x 2) are connected if either x 1 is adjacent to x 1 or x 2 is adjacent to x 2 in graphs G 1 and G 2, respectively. G 1 G 2 = G 1 G 2 G n = G n 18/73 18

19 AND product graph vs OR product graph (1, 1) (1, 1) (3, 3) (1, 2) (3, 3) (1, 2) (3, 2) (1, 3) (3, 2) (1, 3) 1 (3, 1) (2, 1) (3, 1) (2, 1) 2 3 (2, 3) (2, 2) (2, 3) (2, 2) G original graph 2 AND Power graph 2 OR Power graph 19/73 19

20 Outline Introduction Zero error source coding (Scalar vs. Block coding) Fixed length coding Variable length coding Function compression Optimal rate Achievable coding scheme Function compression in zero-error regime Compression of specific functions 20/73 20

21 Optimal Rate: Source coding, Function compression Source, side info fixed length Source, side info variable length Function scalar code Zero error block code Zero error block code error 21/73 21

22 Zero error source coding In information theory: Increasing the number of repetitions decreases p(e) In some applications no error can be tolerated. Sometimes there are only few source instances available. The error prob. decreases as the no. of instances increases Not useful 22/73 22

23 Zero error- Simple point-to-point communication - Scalar Every source instances, a random variable X is generated according to p(x) independently of all other instance. for a single instance, the smallest number of bits in the worst case is log X log X log X + 1 x Encoder Decoder ˆx X n =[x(1),x(2),...,x(n)] x 2 X x p(x) 23/73 23

24 Zero error- Simple point-to-point, Bocking For n independent instances, the number of bits needed in the worst case is n log X n log X n log X + 1 The asymptotic per-instance In worst case is between log X and log X + 1/n bits X n Encoder Decoder ˆXn X n =[x(1),x(2),...,x(n)] x 2 X x p(x) p{x n 6= ˆX n } =0 24/73 24

25 Zero-error source coding with side information The receiver knows some possibly related rv. Y jointly distributed with X Probability of error must be exactly zero. How many bits are needed in the worst case? Any advantages over block coding of any independent instances? X n Encoder Decoder ˆX n P ( ˆX n 6= X n )=0 Y n 25/73 25

26 Zero-error source coding, Fixed length code Receiver wants to know X without error. (No error is tolerated!) X = {1, 2,..., 5} and Y = {1, 2,..., 5} x y X n Encoder Decoder ˆX n P ( ˆX n 6= X n )=0 Y n 26/73 26

27 Confusability graph x, x are confusable if y Y such that p(x, y)p(x, y) > 0 Given y = 1, then x can be either 1 or 2 (1 and 2 are confusable). This confusability relation can be captured by the characteristic graph G. x y (x, x 0 ) 2 E () if x, x 0 are confusable /73 27

28 Encoding 1-instance at a time, Scaler code The smallest number of possible messages is χ(g) In worst case the smallest number of bits needed is log(χ(g)) x y x x 5 x x 4 x 3 (G) =3 28/73 28

29 n-instance encoding Assign different codewords to the confusable vectors x n = (x 1,.., x n ) and x n = (x 1,..., x n) are confusable if y n Y n such that p(x n, y n )p(x n, y n ) > 0 Confusability of the vectors can be captured by the AND power graph. G n = (V n, E n ) (x n, x n ) E n if (x i, x i ) E or x i = x i for all i {1,..., n} The number of bits needed in worst case is log χ(g n ) In general χ(g n ) (χ(g)) n benefit of block encoding 29/73 29

30 2-AND power graph of the pentagon The five individual pentagons represent the confusability relation for x 2, each for fixed x 1 meta-pentagon represents the confusability relation for x 1 alone (1, 1) (1, 5) (1, 2) (1, 4) (1, 3) log 2 χ(g 2 ) = log 2 5 < log 2 χ 2 (G) = log 2 9 (5, 5) (5, 4) (5, 2) (2, 5) (2, 1) (2, 2) Benefit from blocking can be obtained (5, 3) (2, 4) (2, 3) (4, 1) (3, 1) (4, 5) (4, 2) (3, 5) (3, 2) (4, 4) (4, 3) (3, 4) (3, 3) 30/73 30

31 Outline Introduction Zero error source coding (Scalar vs. Block coding) Fixed length coding Variable length coding Asymptotically zero probability of error Optimal rate for function compression Achievable coding scheme Compression of specific functions 31/73 31

32 Variable length coding, Simple point-to-point Scalar code: the smallest expected number of bits needed is between H(X ) and H(X ) + 1 Block encoding: the number of bits needed on average is between nh(x ) and nh(x ) + 1 The asymptotic per-instance on average is between H(X ) and H(X ) + 1 n bits x Encoder Decoder ˆx X n =[x(1),x(2),...,x(n)] x 2 X x p(x) 32/73 32

33 Variable length coding, Side information Variable length codes assigns shorter codes to the most frequent symbols and vice versa. Alon and Orlitsky considered two family of codes. 2 No single letter characterization exists for optimal variable length code Complementary graph entropy For a subclass of variable length codes the optimal rate is Graph entropy. X n Encoder Decoder ˆX n P ( ˆX n 6= X n )=0 Y n 2 N. Alon and A. Orlitsky, Source coding and graph entropies, Information Theory, IEEE Transactions 33/73 33

34 Coding method for unrestricted inputs (UI) Encoding function φ : X {0, 1} { for all x, x V φ(x) P φ(x ) (x, x ) E φ(x) φ(x ) The decoder learns x only if p(x, y) > 0. o.w., the decoder may be wrong on the value of x. (x 1), (x 2), (x3), (x 4 ) ) (x 1) (x 2) (x 3) (x 4) (x 1 )= (x 3 ) p(x 1,y 1 ) > 0 p(x 3,y 1 )=0 ) x 1 is detected x 1 x 2 x 3 x 4 ) Encoder Decoder Lookup table x 1 x 2 x 3 x 4 y 1 y 2 y 3 y 4 34/73 34

35 Coding method for unrestricted inputs (UI) Encoding function φ : X {0, 1} { for all x, x V φ(x) P φ(x ) (x, x ) E φ(x) φ(x ) UI code is a coloring of G followed by a prefix-free coding of the colors. (x 1), (x 2), (x3), (x 4 ) ) (x 1) (x 2) (x 3) (x 4) (x 1 )= (x 3 ) p(x 1,y 1 ) > 0 p(x 3,y 1 )=0 ) x 1 is detected x 1 x 2 x 3 x 4 ) Encoder Decoder Lookup table x 1 x 2 x 3 x 4 y 1 y 2 y 3 y 4 35/73 35

36 Expected number of bits Note The optimal code isn t necessarily induced by the optimal coloring The expected number of bits transmitted is l(ψ) = x V P(x) ψ(x) The minimum Rate of UI codes are: L(G) = min{l(ψ) : ψ is an UI code} call the codes attaining these minima the optimal code 36/73 36

37 UI code-scalar H χ (G, X ) L H χ (G, X ) + 1 Upper bound: Take a coloring of G achieving H χ (G, X ). There is a prefix free encoding of the colors whose expected length is at most H χ (G, X ) + 1 Data compression result H(X ) l p.f (X ) H(X ) + 1 lower bound Every UI protocol can be viewed as a coloring of G and prefix free encoding of it s colors. The entropy of a coloring is at least H χ (G, X ) The expected length is at least H χ (G, X ) 37/73 37

38 n-independent instance x 1 x 2 x 3 x 4 Encoder Decoder? x 2? x 4 p(x 1,y 3 )=0 p(x 3,y 1 )=0 y 3 y 2 y 1 y 4 Decoder needs to learn x i for exactly those i s where p(x i, y i ) > 0 and can be wrong about x i for the other i s. This may require more bits. x n, x n need to be distinguished if i such that p(x i, y i )p(x i, y i) > 0 (x n, x n ) E n if (x i, x i ) E for some i {1,.., n} This is the definition of OR-Power graph of G 38/73 38

39 Optimal rate for UI (variable length code) Remember for one instance the optimal rate is for n-instances Asymptotic per instance 3 : H χ (G, X ) L H χ (G, X ) + 1 H χ (G n, X (n) ) L n H χ (G n n, X (n) ) L am = lim n n H χ(g n, X (n) ) = H G (X ) 3 N. Alon and A. Orlitsky, Source coding and graph entropies, Information Theory, IEEE Transactions 39/73 39

40 Graph entropy Was originally defined by Körner(1973). H G (X ) = min I (W, X ) X W Γ(G) Γ(G) is the set of independent sets of the confusability graph G W is an independent set of G The minimization can be restricted to W ranging over the independent sets. 40/73 40

41 Source, side info fixed length Source, side info variable length Function scalar code Zero error log (G) H (G, X)? block code Zero error log (G^n ) H G (X)? block code error?? H G (X Y ) apple H(X Y ) apple H G (X) apple H (G, X) apple H(X) apple log (G) 41/73 41

42 Outline Introduction Zero error source coding (Scalar vs. Block coding) Fixed Length coding Variable Length coding Asymptotically zero error Optimal rate for function compression Achievable coding scheme Function compression in zero-error regime Compression of specific functions 42/73 42

43 Function computing Compute a function f (X, Y ) at the decoder, and we want this encoding to be as efficient as possible f (X, Y ) = Z must be determined for a (single) block of multiple instances Vanishing (bit) block probability of error is allowed H(f (X, Y ) Y ) Rate= R H(X Y ) : X n! {0, 1} k : {0, 1} k Y n! Z n p(x, Y ) X encoder decoder ˆf(X, Y ) Y p( ( (x), y) 6= f(x, y)) < 43/73 43

44 Example Satellite has 10 parameters to report to the base Base needs to learn only one parameter(only the base knows which one) The lower bound gives H(x Y Y ) = 1 bit and the upper bound gives H(X Y ) = 10 bits. R = 10 f(x, Y )=x Y Y Unif(1 : 10) X =(x 1,..,x 10 ) 44/73 44

45 Slepian Wolf In the case where f (X, Y ) = (X, Y ), the well-established rate result is the Slepian-Wolf bound X Y encoder encoder R 1 R 2 R 1 >H(X Y ) ( ˆX,Ŷ ) R 2 >H(Y X) R 1 + R 2 >H(X, Y ) When Y is available at the decoder, the optimal rate for the transmission of X is the conditional entropy of X given Y, H(X Y ). 45/73 45

46 Slepian Wolf To achieve this rate H(X Y ), encoder use more efficient encoding by taking advantage of its correlation with Y. knowing that the decoder can use its knowledge of Y to disambiguate the transmitted data This analysis so far has ignored the fact that we are computing f (X, Y ) at the decoder, rather than reporting the two values directly! 46/73 46

47 Can we do better than Slepian Wolf bound? Example: X and Y are integers and f = X + Y (mod 4). Regardless of how large X and Y might be, just encode the last two bits of each. The SW bound only takes advantage of the correlation between the two rvs. SW doesn t take advantage of the properties of the function To analyze a given f, we construct f-confusability graph. 47/73 47

48 f -Confusability graph f-confusability graph of X given Y, f, and p(x, y) is a graph G = (V, E) V x = X and, (x, x ) E if y Y, such that, both p(x, y) > 0 and p(x, y) > 0 and f (x, y) f (x, y). Negation If two nodes are not confusable, for all possible values of y either p(x, y) = 0 or p(x 0,y)=0 (x, x 0 ) 62 E or f(x, y) =f(x 0,y) 48/73 48

49 Encoding scheme 1-instance (x, x ) E can safely be given the same encoding Color the f-confusibility graph G f with minimum entropy Transmit the name of the color class that contains each value SW: for asymptotically zero-error, it suffices to send the conditional entropy H(X Y ) bits. Use SW encoding on the distribution over the color classes. This rate is H χ (G f, X Y ) = min p(y)h(c(x) Y = y) c y Y Slepian-Wolf Code X Graph Coloring ENCODER DECODER Lookup Table f(x, Y ) Y 49/73 49

50 Encoding scheme n-instance How to define the notion of f-confusability for vector? (x n, x n ) are f-confusable, if y n Y n such that p(x n, y n )p(x n, y n ) > 0 and f (x n, y n ) f (x n, y n ) (x n, x n ) E n if (x i, x i ) E for some i {1, 2,..., n} This is the definition of OR power graph 50/73 50

51 Optimal rate-achievable scheme For 1-instance Asymptotic per instance H χ (G f, X Y ) 1 lim n n H χ(g n f, X (n) Y (n) ) = H Gf (X Y ) we can compute the function with vanishing probability of error by first coloring a sufficiently large power graph of the characteristic graph encode the colors using any code that achieves the entropy bound on the colored source 51/73 51

52 Optimal rate optimal rate for the zero distortion functional compression problem with side information equals R = H G (X Y ) 4 H G (X Y ) = min I (W, X Y ) W X Y X W Γ(G) where W X Y indicates that W, X, Y is a Markov chain. W should not contain any information about Y that is not available through X The minimization can be restricted to W ranging over the maximal independent sets. 4 A. Orlitsky and J. Roche, Coding for computing, Information Theory, IEEE Transactions 52/73 52

53 Summary Source, side info fixed length Source, side info variable length Function scalar code Zero error log (G) H (G, X)? block code Zero error log (G^n ) H G (X)? block code error H(X Y ) H Gf (X Y ) H Gf (X Y ) apple H(X Y ) apple H G (X) apple H (G, X) apple H(X) apple log (G) 53/73 53

54 Outline Introduction Zero-error source coding (Scalar vs. Block coding) Fixed Length coding Variable Length coding Asymptotically zero-error Optimal rate for function compression Achievable coding scheme Function Computation in zero-error regime Compression of specific functions 54/73 54

55 Function Compression, Scalar code, Zero-error Compute a function f (X, Y ) at the decoder. f (X, Y ) must be determined for a (single) block instance. No error is allowed (zero-error regime) Fundamental limit of computing a function with exactly zero probability of error? Is motivated by Alon, and Orlitsky ( source coding with side information) p(x, Y ) X encoder decoder ˆf(X, Y ) Y 55/73 55

56 Achievable scheme The protocol that guarantees the zero probability of error, and to compute f (x, y) only when p(x, y) > 0 (x, x ) E f can safely be given the same encoding Either p(x, y) = 0 or p(x, y) = 0 f (x, y) = f (x, y) Color the f-confusibility graph G f with minimum entropy Transmit the name of the color class that contains each value Using a prefix free encoding on the distribution over the color classes yields a valid encoding scheme for the problem 56/73 56

57 Achievable coding scheme Hu man Code X Graph Coloring ENCODER DECODER Lookup Table f(x, Y ) Y 57/73 57

58 Optimal scalar code for Function computing Theorem Optimal rate for computing a function f (X, Y ) with zero probability of error, using scalar variable length code is H χ (G f, X ) R H χ (G f, X ) + 1 Proof. Upper bound: Take a coloring of G achieving H χ (G f, X ). There is a prefix free encoding of the colors whose expected length is at most H χ (G f, X ) + 1 lower bound This protocol can be viewed as a coloring of G f prefix free encoding of it s colors. and The entropy of a coloring is at least H χ (G f, X ) The expected length is at least H χ (G f, X ) 58/73 58

59 Block encoding To protect against loss of synchronization if the side information at the decoder is occasionally wrong: For a sequence x 1 x 2,..x n, the function needs to be computed correctly only for those coordinates that p(x i, y i ) > 0. For other positions decoder may not be able to compute f (x i, y i ) correctly. (x n, x n ) E (n) if (x i, x i ) E for some i {1,.., n} (x n, x n ) E (n) φ(x n ) φ(x n ) (1) x n, x n V (n) φ(x n ) P φ(x n ) (2) 59/73 59

60 Optimal rate for function compression in zero-error regime One instance n-instance Asymptotic per instance H χ (G f, X ) R H χ (G f, X ) + 1 H χ (G n f, X (n) ) R n H χ (G n f, X (n) ) lim n n H χ(g n f, X (n) ) = H Gf (X ) We can compute the function with zero-error by first coloring a sufficiently large power graph of the characteristic graph Encode the colors using Huffman codes 60/73 60

61 function compression in zero-error regime Proposition In exactly zero-error regime, using block code, the optimal rate for computing f (X, Y ) is less than recovering the source X. Proof. G f = (V, E f ) and G = (V, E), and E f E Let (X, W ) be random variable achieving H G (X ) W is an independent set containing X, and G f G it s also an independent set of G f. H Gf (X ) = min I (X ; W ) I (X ; W ) = min I (X ; W ) = H G (X ) X W Γ(G f ) X W Γ(G) 61/73 61

62 function compression in zero-error regime Proposition In exactly zero-error regime, using scalar code, the optimal rate for computing f (X, Y ) is less than recovering the source X. G f G Any coloring of G is a coloring of G f The coloring with minimum entropy of G is also a coloring of G f. H χ (G f, X ) H(c (G f )) = H χ (G, x) H(c(G)) 62/73 62

63 Summary Theorem We can get saving over the block encoding in zero-error function compression regime. H Gf (X ) H χ (G f, X ) Proof. if c is a coloring of G, then c 1 (c(x )), the color class of X, induced a partition of V into independent sets of G, and always contains X. H χ (G f, X ) = min{h(c(x ) c 1 (c(x )) are the color classes} a = min {H(W ) disjoint union of W s are the color classes} X W Γ(G f ) b = min {I (W ; X ) disjoint union of W s are the color classes} X W Γ(G f ) c min {I (W ; X )} = H G f (X ) X W Γ(G f ) 63/73 63

64 Summary Source, side info fixed length Source, side info variable length Function scalar code Zero error log (G) H (G, X) H (G f,x) block code Zero error log (G^n ) H G (X) H Gf (X) block code error H(X Y ) H Gf (X Y ) H G (X Y ) apple H(X Y ) apple H G (X) apple H (G, X) apple H(X) apple log (G) 64/73 64

65 Outline Introduction Zero-error source coding (Scalar vs. Block coding) Fixed Length coding Variable Length coding Asymptotically zero-error Optimal rate for function compression Achievable coding scheme Function compression zero-error regime Compression of specific functions 65/73 65

66 Compression of specific functions In environmental monitoring, a relevant statistic of temperature sensor readings x 1, x 2,..., x n may be the mean temperature, or median, or mode. In alarm networks, the quantity of interest might be the maximum, max 1 i n (x i ) of n temperature readings. Frequency histogram function counts the number of occurrences of each argument among measurements. It yields many good results such as mean, variance, maximum, minimum, median and other important statistics Function compression on specific functions by graph coloring which provide benefit over compressing each source separately 66/73 66

67 mod r-sum We have K users which has access to independent sequences Each node (user) periodically makes measurements, which belongs to a fixed finite set X = {1, 2,..., X }. f : X K {0,..., r 1} X 1 = x1(1),x1(2),...,x1(n) Enc 1 X 2 = x2(1),x2(2),...,x2(n) Enc 2 Dec f(x 1(j),x 2(j),...,x K(j)) = x 1(j) x 2(j)... x K(j) for j 2 {1,...,N} X K = xk(1),xk(2),...,xk(n) Enc K 67/73 67

68 The characteristic graph of the i-th user Theorem (p, q) E i (p, q X ), if and only if, p q mod r Proof. : Since (p, q) E i, we obtain that for all x \i X N 1, f (X ) = f (p, x \i ) = f (q, x \i ). p + j i x j q + j i x j mod r p q mod r. : If p q mod r, and x j x j mod r, then for all x \i X N 1 we j i j i have p + j i x j q + j i x j mod r. Hence, p, q E i 68/73 68

69 The characteristic graph of the i-th user Note that maximal independent sets form a partition. W0 = {x X x = 0 (modr)} W1 = {x X x = 1 (modr)}... Wr 1 = {x X x = r 1 (modr)} H χ (G, X ) can be achieved by assigning different colors to different maximal independent sets. The minimum entropy coloring can be achieved by optimal coloring χ(g) 69/73 69

70 Optimal rate for Mod r sum H G (X Y ) = min I (W, X Y ) = min{h(w Y ) H(W XY )} = min{h(w Y ) H(W X )} = H(W Y ) where P i = x i W i p(x i ) = H(W ) = (P 0 log P P r 1 log P r 1 ) If the distribution over the vertices is uniform, and X = Cr, and C Z +, the total benefit is H(X Y ) H G (X Y ) = log X log r = log C 0 70/73 70

71 Summary Source coding problem with side information in zero-error regime is discussed. The optimal rate for function compression in zero-error regime is derived Function computation in asymptoticly zero error regime is discussed. The optimal rate for computing mod r sum is derived. 71/73 71

72 Thank you Questions? 72/73 72

Distributed Functional Compression through Graph Coloring

Distributed Functional Compression through Graph Coloring Distributed Functional Compression through Graph Coloring Vishal Doshi, Devavrat Shah, Muriel Médard, and Sidharth Jaggi Laboratory for Information and Decision Systems Massachusetts Institute of Technology

More information

Cases Where Finding the Minimum Entropy Coloring of a Characteristic Graph is a Polynomial Time Problem

Cases Where Finding the Minimum Entropy Coloring of a Characteristic Graph is a Polynomial Time Problem Cases Where Finding the Minimum Entropy Coloring of a Characteristic Graph is a Polynomial Time Problem Soheil Feizi, Muriel Médard RLE at MIT Emails: {sfeizi,medard}@mit.edu Abstract In this paper, we

More information

Graph Coloring and Conditional Graph Entropy

Graph Coloring and Conditional Graph Entropy Graph Coloring and Conditional Graph Entropy Vishal Doshi, Devavrat Shah, Muriel Médard, Sidharth Jaggi Laboratory for Information and Decision Systems Massachusetts Institute of Technology Cambridge,

More information

On Network Functional Compression

On Network Functional Compression On Network Functional Compression Soheil Feizi, Student Member, IEEE, Muriel Médard, Fellow, IEEE arxiv:0.5496v2 [cs.it] 30 Nov 200 Abstract In this paper, we consider different aspects of the problem

More information

Broadcasting With Side Information

Broadcasting With Side Information Department of Electrical and Computer Engineering Texas A&M Noga Alon, Avinatan Hasidim, Eyal Lubetzky, Uri Stav, Amit Weinstein, FOCS2008 Outline I shall avoid rigorous math and terminologies and be more

More information

A Genetic Algorithm to Minimize Chromatic Entropy

A Genetic Algorithm to Minimize Chromatic Entropy A Genetic Algorithm to Minimize Chromatic Entropy Greg Durrett 1,MurielMédard 2, and Una-May O Reilly 1 1 Computer Science and Artificial Intelligence Laboratory 2 Research Laboratory for Electronics Massachusetts

More information

A Graph-based Framework for Transmission of Correlated Sources over Multiple Access Channels

A Graph-based Framework for Transmission of Correlated Sources over Multiple Access Channels A Graph-based Framework for Transmission of Correlated Sources over Multiple Access Channels S. Sandeep Pradhan a, Suhan Choi a and Kannan Ramchandran b, a {pradhanv,suhanc}@eecs.umich.edu, EECS Dept.,

More information

Lecture 4 Noisy Channel Coding

Lecture 4 Noisy Channel Coding Lecture 4 Noisy Channel Coding I-Hsiang Wang Department of Electrical Engineering National Taiwan University ihwang@ntu.edu.tw October 9, 2015 1 / 56 I-Hsiang Wang IT Lecture 4 The Channel Coding Problem

More information

Coding for Computing. ASPITRG, Drexel University. Jie Ren 2012/11/14

Coding for Computing. ASPITRG, Drexel University. Jie Ren 2012/11/14 Coding for Computing ASPITRG, Drexel University Jie Ren 2012/11/14 Outline Background and definitions Main Results Examples and Analysis Proofs Background and definitions Problem Setup Graph Entropy Conditional

More information

Network coding for multicast relation to compression and generalization of Slepian-Wolf

Network coding for multicast relation to compression and generalization of Slepian-Wolf Network coding for multicast relation to compression and generalization of Slepian-Wolf 1 Overview Review of Slepian-Wolf Distributed network compression Error exponents Source-channel separation issues

More information

ECE Information theory Final (Fall 2008)

ECE Information theory Final (Fall 2008) ECE 776 - Information theory Final (Fall 2008) Q.1. (1 point) Consider the following bursty transmission scheme for a Gaussian channel with noise power N and average power constraint P (i.e., 1/n X n i=1

More information

Index coding with side information

Index coding with side information Index coding with side information Ehsan Ebrahimi Targhi University of Tartu Abstract. The Index Coding problem has attracted a considerable amount of attention in the recent years. The problem is motivated

More information

Bandwidth: Communicate large complex & highly detailed 3D models through lowbandwidth connection (e.g. VRML over the Internet)

Bandwidth: Communicate large complex & highly detailed 3D models through lowbandwidth connection (e.g. VRML over the Internet) Compression Motivation Bandwidth: Communicate large complex & highly detailed 3D models through lowbandwidth connection (e.g. VRML over the Internet) Storage: Store large & complex 3D models (e.g. 3D scanner

More information

Distributed Lossy Interactive Function Computation

Distributed Lossy Interactive Function Computation Distributed Lossy Interactive Function Computation Solmaz Torabi & John MacLaren Walsh Dept. of Electrical and Computer Engineering Drexel University Philadelphia, PA 19104 Email: solmaz.t@drexel.edu &

More information

Computing and Communicating Functions over Sensor Networks

Computing and Communicating Functions over Sensor Networks Computing and Communicating Functions over Sensor Networks Solmaz Torabi Dept. of Electrical and Computer Engineering Drexel University solmaz.t@drexel.edu Advisor: Dr. John M. Walsh 1/35 1 Refrences [1]

More information

Multi-coloring and Mycielski s construction

Multi-coloring and Mycielski s construction Multi-coloring and Mycielski s construction Tim Meagher Fall 2010 Abstract We consider a number of related results taken from two papers one by W. Lin [1], and the other D. C. Fisher[2]. These articles

More information

Chapter 3 Source Coding. 3.1 An Introduction to Source Coding 3.2 Optimal Source Codes 3.3 Shannon-Fano Code 3.4 Huffman Code

Chapter 3 Source Coding. 3.1 An Introduction to Source Coding 3.2 Optimal Source Codes 3.3 Shannon-Fano Code 3.4 Huffman Code Chapter 3 Source Coding 3. An Introduction to Source Coding 3.2 Optimal Source Codes 3.3 Shannon-Fano Code 3.4 Huffman Code 3. An Introduction to Source Coding Entropy (in bits per symbol) implies in average

More information

Coding of memoryless sources 1/35

Coding of memoryless sources 1/35 Coding of memoryless sources 1/35 Outline 1. Morse coding ; 2. Definitions : encoding, encoding efficiency ; 3. fixed length codes, encoding integers ; 4. prefix condition ; 5. Kraft and Mac Millan theorems

More information

On Zero-Error Source Coding With Decoder Side Information

On Zero-Error Source Coding With Decoder Side Information IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 49, NO. 1, JANUARY 2003 99 On Zero-Error Source Coding With Decoder Side Information Prashant Koulgi, Student Member, IEEE, Ertem Tuncel, Student Member, IEEE,

More information

Entropies & Information Theory

Entropies & Information Theory Entropies & Information Theory LECTURE I Nilanjana Datta University of Cambridge,U.K. See lecture notes on: http://www.qi.damtp.cam.ac.uk/node/223 Quantum Information Theory Born out of Classical Information

More information

Chapter 2 Date Compression: Source Coding. 2.1 An Introduction to Source Coding 2.2 Optimal Source Codes 2.3 Huffman Code

Chapter 2 Date Compression: Source Coding. 2.1 An Introduction to Source Coding 2.2 Optimal Source Codes 2.3 Huffman Code Chapter 2 Date Compression: Source Coding 2.1 An Introduction to Source Coding 2.2 Optimal Source Codes 2.3 Huffman Code 2.1 An Introduction to Source Coding Source coding can be seen as an efficient way

More information

Lecture 4 Channel Coding

Lecture 4 Channel Coding Capacity and the Weak Converse Lecture 4 Coding I-Hsiang Wang Department of Electrical Engineering National Taiwan University ihwang@ntu.edu.tw October 15, 2014 1 / 16 I-Hsiang Wang NIT Lecture 4 Capacity

More information

Communication Cost of Distributed Computing

Communication Cost of Distributed Computing Communication Cost of Distributed Computing Abbas El Gamal Stanford University Brice Colloquium, 2009 A. El Gamal (Stanford University) Comm. Cost of Dist. Computing Brice Colloquium, 2009 1 / 41 Motivation

More information

EE5585 Data Compression May 2, Lecture 27

EE5585 Data Compression May 2, Lecture 27 EE5585 Data Compression May 2, 2013 Lecture 27 Instructor: Arya Mazumdar Scribe: Fangying Zhang Distributed Data Compression/Source Coding In the previous class we used a H-W table as a simple example,

More information

MARKOV CHAINS A finite state Markov chain is a sequence of discrete cv s from a finite alphabet where is a pmf on and for

MARKOV CHAINS A finite state Markov chain is a sequence of discrete cv s from a finite alphabet where is a pmf on and for MARKOV CHAINS A finite state Markov chain is a sequence S 0,S 1,... of discrete cv s from a finite alphabet S where q 0 (s) is a pmf on S 0 and for n 1, Q(s s ) = Pr(S n =s S n 1 =s ) = Pr(S n =s S n 1

More information

SHARED INFORMATION. Prakash Narayan with. Imre Csiszár, Sirin Nitinawarat, Himanshu Tyagi, Shun Watanabe

SHARED INFORMATION. Prakash Narayan with. Imre Csiszár, Sirin Nitinawarat, Himanshu Tyagi, Shun Watanabe SHARED INFORMATION Prakash Narayan with Imre Csiszár, Sirin Nitinawarat, Himanshu Tyagi, Shun Watanabe 2/40 Acknowledgement Praneeth Boda Himanshu Tyagi Shun Watanabe 3/40 Outline Two-terminal model: Mutual

More information

Reliable Computation over Multiple-Access Channels

Reliable Computation over Multiple-Access Channels Reliable Computation over Multiple-Access Channels Bobak Nazer and Michael Gastpar Dept. of Electrical Engineering and Computer Sciences University of California, Berkeley Berkeley, CA, 94720-1770 {bobak,

More information

1 Introduction to information theory

1 Introduction to information theory 1 Introduction to information theory 1.1 Introduction In this chapter we present some of the basic concepts of information theory. The situations we have in mind involve the exchange of information through

More information

Multimedia Communications. Mathematical Preliminaries for Lossless Compression

Multimedia Communications. Mathematical Preliminaries for Lossless Compression Multimedia Communications Mathematical Preliminaries for Lossless Compression What we will see in this chapter Definition of information and entropy Modeling a data source Definition of coding and when

More information

Run-length & Entropy Coding. Redundancy Removal. Sampling. Quantization. Perform inverse operations at the receiver EEE

Run-length & Entropy Coding. Redundancy Removal. Sampling. Quantization. Perform inverse operations at the receiver EEE General e Image Coder Structure Motion Video x(s 1,s 2,t) or x(s 1,s 2 ) Natural Image Sampling A form of data compression; usually lossless, but can be lossy Redundancy Removal Lossless compression: predictive

More information

Non-linear index coding outperforming the linear optimum

Non-linear index coding outperforming the linear optimum Non-linear index coding outperforming the linear optimum Eyal Lubetzky Uri Stav Abstract The following source coding problem was introduced by Birk and Kol: a sender holds a word x {0, 1} n, and wishes

More information

SIGNAL COMPRESSION Lecture Shannon-Fano-Elias Codes and Arithmetic Coding

SIGNAL COMPRESSION Lecture Shannon-Fano-Elias Codes and Arithmetic Coding SIGNAL COMPRESSION Lecture 3 4.9.2007 Shannon-Fano-Elias Codes and Arithmetic Coding 1 Shannon-Fano-Elias Coding We discuss how to encode the symbols {a 1, a 2,..., a m }, knowing their probabilities,

More information

Lecture 1 : Data Compression and Entropy

Lecture 1 : Data Compression and Entropy CPS290: Algorithmic Foundations of Data Science January 8, 207 Lecture : Data Compression and Entropy Lecturer: Kamesh Munagala Scribe: Kamesh Munagala In this lecture, we will study a simple model for

More information

Guess & Check Codes for Deletions, Insertions, and Synchronization

Guess & Check Codes for Deletions, Insertions, and Synchronization Guess & Check Codes for Deletions, Insertions, and Synchronization Serge Kas Hanna, Salim El Rouayheb ECE Department, Rutgers University sergekhanna@rutgersedu, salimelrouayheb@rutgersedu arxiv:759569v3

More information

Common Information. Abbas El Gamal. Stanford University. Viterbi Lecture, USC, April 2014

Common Information. Abbas El Gamal. Stanford University. Viterbi Lecture, USC, April 2014 Common Information Abbas El Gamal Stanford University Viterbi Lecture, USC, April 2014 Andrew Viterbi s Fabulous Formula, IEEE Spectrum, 2010 El Gamal (Stanford University) Disclaimer Viterbi Lecture 2

More information

EECS 229A Spring 2007 * * (a) By stationarity and the chain rule for entropy, we have

EECS 229A Spring 2007 * * (a) By stationarity and the chain rule for entropy, we have EECS 229A Spring 2007 * * Solutions to Homework 3 1. Problem 4.11 on pg. 93 of the text. Stationary processes (a) By stationarity and the chain rule for entropy, we have H(X 0 ) + H(X n X 0 ) = H(X 0,

More information

ECE Information theory Final

ECE Information theory Final ECE 776 - Information theory Final Q1 (1 point) We would like to compress a Gaussian source with zero mean and variance 1 We consider two strategies In the first, we quantize with a step size so that the

More information

18.310A Final exam practice questions

18.310A Final exam practice questions 18.310A Final exam practice questions This is a collection of practice questions, gathered randomly from previous exams and quizzes. They may not be representative of what will be on the final. In particular,

More information

Source Coding. Master Universitario en Ingeniería de Telecomunicación. I. Santamaría Universidad de Cantabria

Source Coding. Master Universitario en Ingeniería de Telecomunicación. I. Santamaría Universidad de Cantabria Source Coding Master Universitario en Ingeniería de Telecomunicación I. Santamaría Universidad de Cantabria Contents Introduction Asymptotic Equipartition Property Optimal Codes (Huffman Coding) Universal

More information

Communicating the sum of sources in a 3-sources/3-terminals network

Communicating the sum of sources in a 3-sources/3-terminals network Communicating the sum of sources in a 3-sources/3-terminals network Michael Langberg Computer Science Division Open University of Israel Raanana 43107, Israel Email: mikel@openu.ac.il Aditya Ramamoorthy

More information

Upper Bounds on the Capacity of Binary Intermittent Communication

Upper Bounds on the Capacity of Binary Intermittent Communication Upper Bounds on the Capacity of Binary Intermittent Communication Mostafa Khoshnevisan and J. Nicholas Laneman Department of Electrical Engineering University of Notre Dame Notre Dame, Indiana 46556 Email:{mhoshne,

More information

Design of Optimal Quantizers for Distributed Source Coding

Design of Optimal Quantizers for Distributed Source Coding Design of Optimal Quantizers for Distributed Source Coding David Rebollo-Monedero, Rui Zhang and Bernd Girod Information Systems Laboratory, Electrical Eng. Dept. Stanford University, Stanford, CA 94305

More information

EE376A: Homework #3 Due by 11:59pm Saturday, February 10th, 2018

EE376A: Homework #3 Due by 11:59pm Saturday, February 10th, 2018 Please submit the solutions on Gradescope. EE376A: Homework #3 Due by 11:59pm Saturday, February 10th, 2018 1. Optimal codeword lengths. Although the codeword lengths of an optimal variable length code

More information

COMPLEMENTARY graph entropy was introduced

COMPLEMENTARY graph entropy was introduced IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 55, NO. 6, JUNE 2009 2537 On Complementary Graph Entropy Ertem Tuncel, Member, IEEE, Jayanth Nayak, Prashant Koulgi, Student Member, IEEE, Kenneth Rose, Fellow,

More information

AN INTRODUCTION TO SECRECY CAPACITY. 1. Overview

AN INTRODUCTION TO SECRECY CAPACITY. 1. Overview AN INTRODUCTION TO SECRECY CAPACITY BRIAN DUNN. Overview This paper introduces the reader to several information theoretic aspects of covert communications. In particular, it discusses fundamental limits

More information

SOURCE CODING WITH SIDE INFORMATION AT THE DECODER (WYNER-ZIV CODING) FEB 13, 2003

SOURCE CODING WITH SIDE INFORMATION AT THE DECODER (WYNER-ZIV CODING) FEB 13, 2003 SOURCE CODING WITH SIDE INFORMATION AT THE DECODER (WYNER-ZIV CODING) FEB 13, 2003 SLEPIAN-WOLF RESULT { X i} RATE R x ENCODER 1 DECODER X i V i {, } { V i} ENCODER 0 RATE R v Problem: Determine R, the

More information

Interactive Communication for Data Exchange

Interactive Communication for Data Exchange Interactive Communication for Data Exchange Himanshu Tyagi Indian Institute of Science, Bangalore Joint work with Pramod Viswanath and Shun Watanabe The Data Exchange Problem [ElGamal-Orlitsky 84], [Csiszár-Narayan

More information

Lecture 4 October 18th

Lecture 4 October 18th Directed and undirected graphical models Fall 2017 Lecture 4 October 18th Lecturer: Guillaume Obozinski Scribe: In this lecture, we will assume that all random variables are discrete, to keep notations

More information

Lecture 4 : Adaptive source coding algorithms

Lecture 4 : Adaptive source coding algorithms Lecture 4 : Adaptive source coding algorithms February 2, 28 Information Theory Outline 1. Motivation ; 2. adaptive Huffman encoding ; 3. Gallager and Knuth s method ; 4. Dictionary methods : Lempel-Ziv

More information

On Scalable Coding in the Presence of Decoder Side Information

On Scalable Coding in the Presence of Decoder Side Information On Scalable Coding in the Presence of Decoder Side Information Emrah Akyol, Urbashi Mitra Dep. of Electrical Eng. USC, CA, US Email: {eakyol, ubli}@usc.edu Ertem Tuncel Dep. of Electrical Eng. UC Riverside,

More information

Chapter 4. Data Transmission and Channel Capacity. Po-Ning Chen, Professor. Department of Communications Engineering. National Chiao Tung University

Chapter 4. Data Transmission and Channel Capacity. Po-Ning Chen, Professor. Department of Communications Engineering. National Chiao Tung University Chapter 4 Data Transmission and Channel Capacity Po-Ning Chen, Professor Department of Communications Engineering National Chiao Tung University Hsin Chu, Taiwan 30050, R.O.C. Principle of Data Transmission

More information

Complexity Theory VU , SS The Polynomial Hierarchy. Reinhard Pichler

Complexity Theory VU , SS The Polynomial Hierarchy. Reinhard Pichler Complexity Theory Complexity Theory VU 181.142, SS 2018 6. The Polynomial Hierarchy Reinhard Pichler Institut für Informationssysteme Arbeitsbereich DBAI Technische Universität Wien 15 May, 2018 Reinhard

More information

Outline. Complexity Theory EXACT TSP. The Class DP. Definition. Problem EXACT TSP. Complexity of EXACT TSP. Proposition VU 181.

Outline. Complexity Theory EXACT TSP. The Class DP. Definition. Problem EXACT TSP. Complexity of EXACT TSP. Proposition VU 181. Complexity Theory Complexity Theory Outline Complexity Theory VU 181.142, SS 2018 6. The Polynomial Hierarchy Reinhard Pichler Institut für Informationssysteme Arbeitsbereich DBAI Technische Universität

More information

Linear Codes, Target Function Classes, and Network Computing Capacity

Linear Codes, Target Function Classes, and Network Computing Capacity Linear Codes, Target Function Classes, and Network Computing Capacity Rathinakumar Appuswamy, Massimo Franceschetti, Nikhil Karamchandani, and Kenneth Zeger IEEE Transactions on Information Theory Submitted:

More information

On Dominator Colorings in Graphs

On Dominator Colorings in Graphs On Dominator Colorings in Graphs Ralucca Michelle Gera Department of Applied Mathematics Naval Postgraduate School Monterey, CA 994, USA ABSTRACT Given a graph G, the dominator coloring problem seeks a

More information

Information Leakage of Correlated Source Coded Sequences over a Channel with an Eavesdropper

Information Leakage of Correlated Source Coded Sequences over a Channel with an Eavesdropper Information Leakage of Correlated Source Coded Sequences over a Channel with an Eavesdropper Reevana Balmahoon and Ling Cheng School of Electrical and Information Engineering University of the Witwatersrand

More information

Unequal Error Protection Querying Policies for the Noisy 20 Questions Problem

Unequal Error Protection Querying Policies for the Noisy 20 Questions Problem Unequal Error Protection Querying Policies for the Noisy 20 Questions Problem Hye Won Chung, Brian M. Sadler, Lizhong Zheng and Alfred O. Hero arxiv:606.09233v2 [cs.it] 28 Sep 207 Abstract In this paper,

More information

Capacity of a channel Shannon s second theorem. Information Theory 1/33

Capacity of a channel Shannon s second theorem. Information Theory 1/33 Capacity of a channel Shannon s second theorem Information Theory 1/33 Outline 1. Memoryless channels, examples ; 2. Capacity ; 3. Symmetric channels ; 4. Channel Coding ; 5. Shannon s second theorem,

More information

Sum-networks from undirected graphs: Construction and capacity analysis

Sum-networks from undirected graphs: Construction and capacity analysis Electrical and Computer Engineering Conference Papers Posters and Presentations Electrical and Computer Engineering 014 Sum-networks from undirected graphs: Construction and capacity analysis Ardhendu

More information

Submodularity in Machine Learning

Submodularity in Machine Learning Saifuddin Syed MLRG Summer 2016 1 / 39 What are submodular functions Outline 1 What are submodular functions Motivation Submodularity and Concavity Examples 2 Properties of submodular functions Submodularity

More information

An instantaneous code (prefix code, tree code) with the codeword lengths l 1,..., l N exists if and only if. 2 l i. i=1

An instantaneous code (prefix code, tree code) with the codeword lengths l 1,..., l N exists if and only if. 2 l i. i=1 Kraft s inequality An instantaneous code (prefix code, tree code) with the codeword lengths l 1,..., l N exists if and only if N 2 l i 1 Proof: Suppose that we have a tree code. Let l max = max{l 1,...,

More information

Noisy channel communication

Noisy channel communication Information Theory http://www.inf.ed.ac.uk/teaching/courses/it/ Week 6 Communication channels and Information Some notes on the noisy channel setup: Iain Murray, 2012 School of Informatics, University

More information

Lecture 5 Channel Coding over Continuous Channels

Lecture 5 Channel Coding over Continuous Channels Lecture 5 Channel Coding over Continuous Channels I-Hsiang Wang Department of Electrical Engineering National Taiwan University ihwang@ntu.edu.tw November 14, 2014 1 / 34 I-Hsiang Wang NIT Lecture 5 From

More information

Lecture 3: Channel Capacity

Lecture 3: Channel Capacity Lecture 3: Channel Capacity 1 Definitions Channel capacity is a measure of maximum information per channel usage one can get through a channel. This one of the fundamental concepts in information theory.

More information

5218 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 52, NO. 12, DECEMBER 2006

5218 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 52, NO. 12, DECEMBER 2006 5218 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 52, NO. 12, DECEMBER 2006 Source Coding With Limited-Look-Ahead Side Information at the Decoder Tsachy Weissman, Member, IEEE, Abbas El Gamal, Fellow,

More information

Lecture 16 Oct 21, 2014

Lecture 16 Oct 21, 2014 CS 395T: Sublinear Algorithms Fall 24 Prof. Eric Price Lecture 6 Oct 2, 24 Scribe: Chi-Kit Lam Overview In this lecture we will talk about information and compression, which the Huffman coding can achieve

More information

Chapter 2: Source coding

Chapter 2: Source coding Chapter 2: meghdadi@ensil.unilim.fr University of Limoges Chapter 2: Entropy of Markov Source Chapter 2: Entropy of Markov Source Markov model for information sources Given the present, the future is independent

More information

(Classical) Information Theory III: Noisy channel coding

(Classical) Information Theory III: Noisy channel coding (Classical) Information Theory III: Noisy channel coding Sibasish Ghosh The Institute of Mathematical Sciences CIT Campus, Taramani, Chennai 600 113, India. p. 1 Abstract What is the best possible way

More information

Communications Theory and Engineering

Communications Theory and Engineering Communications Theory and Engineering Master's Degree in Electronic Engineering Sapienza University of Rome A.A. 2018-2019 AEP Asymptotic Equipartition Property AEP In information theory, the analog of

More information

Chapter 9 Fundamental Limits in Information Theory

Chapter 9 Fundamental Limits in Information Theory Chapter 9 Fundamental Limits in Information Theory Information Theory is the fundamental theory behind information manipulation, including data compression and data transmission. 9.1 Introduction o For

More information

Optimal matching in wireless sensor networks

Optimal matching in wireless sensor networks Optimal matching in wireless sensor networks A. Roumy, D. Gesbert INRIA-IRISA, Rennes, France. Institute Eurecom, Sophia Antipolis, France. Abstract We investigate the design of a wireless sensor network

More information

Probabilistic Graphical Models (I)

Probabilistic Graphical Models (I) Probabilistic Graphical Models (I) Hongxin Zhang zhx@cad.zju.edu.cn State Key Lab of CAD&CG, ZJU 2015-03-31 Probabilistic Graphical Models Modeling many real-world problems => a large number of random

More information

Introduction to Information Theory. Uncertainty. Entropy. Surprisal. Joint entropy. Conditional entropy. Mutual information.

Introduction to Information Theory. Uncertainty. Entropy. Surprisal. Joint entropy. Conditional entropy. Mutual information. L65 Dept. of Linguistics, Indiana University Fall 205 Information theory answers two fundamental questions in communication theory: What is the ultimate data compression? What is the transmission rate

More information

Dept. of Linguistics, Indiana University Fall 2015

Dept. of Linguistics, Indiana University Fall 2015 L645 Dept. of Linguistics, Indiana University Fall 2015 1 / 28 Information theory answers two fundamental questions in communication theory: What is the ultimate data compression? What is the transmission

More information

Distributed Lossless Compression. Distributed lossless compression system

Distributed Lossless Compression. Distributed lossless compression system Lecture #3 Distributed Lossless Compression (Reading: NIT 10.1 10.5, 4.4) Distributed lossless source coding Lossless source coding via random binning Time Sharing Achievability proof of the Slepian Wolf

More information

Solutions to Set #2 Data Compression, Huffman code and AEP

Solutions to Set #2 Data Compression, Huffman code and AEP Solutions to Set #2 Data Compression, Huffman code and AEP. Huffman coding. Consider the random variable ( ) x x X = 2 x 3 x 4 x 5 x 6 x 7 0.50 0.26 0. 0.04 0.04 0.03 0.02 (a) Find a binary Huffman code

More information

Distributed Lossy Interactive Function Computation

Distributed Lossy Interactive Function Computation Distributed Lossy Interactive Function Computation Solmaz Torabi & John MacLaren Walsh Dept. of Electrical and Computer Engineering Drexel University Philadelphia, PA 19104 Email: solmaz.t@drexel.edu &

More information

Testing Equality in Communication Graphs

Testing Equality in Communication Graphs Electronic Colloquium on Computational Complexity, Report No. 86 (2016) Testing Equality in Communication Graphs Noga Alon Klim Efremenko Benny Sudakov Abstract Let G = (V, E) be a connected undirected

More information

Asymptotic redundancy and prolixity

Asymptotic redundancy and prolixity Asymptotic redundancy and prolixity Yuval Dagan, Yuval Filmus, and Shay Moran April 6, 2017 Abstract Gallager (1978) considered the worst-case redundancy of Huffman codes as the maximum probability tends

More information

Shannon s noisy-channel theorem

Shannon s noisy-channel theorem Shannon s noisy-channel theorem Information theory Amon Elders Korteweg de Vries Institute for Mathematics University of Amsterdam. Tuesday, 26th of Januari Amon Elders (Korteweg de Vries Institute for

More information

Multiterminal Source Coding with an Entropy-Based Distortion Measure

Multiterminal Source Coding with an Entropy-Based Distortion Measure Multiterminal Source Coding with an Entropy-Based Distortion Measure Thomas Courtade and Rick Wesel Department of Electrical Engineering University of California, Los Angeles 4 August, 2011 IEEE International

More information

Graph-based codes for flash memory

Graph-based codes for flash memory 1/28 Graph-based codes for flash memory Discrete Mathematics Seminar September 3, 2013 Katie Haymaker Joint work with Professor Christine Kelley University of Nebraska-Lincoln 2/28 Outline 1 Background

More information

Lecture 3: Error Correcting Codes

Lecture 3: Error Correcting Codes CS 880: Pseudorandomness and Derandomization 1/30/2013 Lecture 3: Error Correcting Codes Instructors: Holger Dell and Dieter van Melkebeek Scribe: Xi Wu In this lecture we review some background on error

More information

Homework Set #2 Data Compression, Huffman code and AEP

Homework Set #2 Data Compression, Huffman code and AEP Homework Set #2 Data Compression, Huffman code and AEP 1. Huffman coding. Consider the random variable ( x1 x X = 2 x 3 x 4 x 5 x 6 x 7 0.50 0.26 0.11 0.04 0.04 0.03 0.02 (a Find a binary Huffman code

More information

Coding for Discrete Source

Coding for Discrete Source EGR 544 Communication Theory 3. Coding for Discrete Sources Z. Aliyazicioglu Electrical and Computer Engineering Department Cal Poly Pomona Coding for Discrete Source Coding Represent source data effectively

More information

On the minimum neighborhood of independent sets in the n-cube

On the minimum neighborhood of independent sets in the n-cube Matemática Contemporânea, Vol. 44, 1 10 c 2015, Sociedade Brasileira de Matemática On the minimum neighborhood of independent sets in the n-cube Moysés da S. Sampaio Júnior Fabiano de S. Oliveira Luérbio

More information

BASICS OF COMPRESSION THEORY

BASICS OF COMPRESSION THEORY BASICS OF COMPRESSION THEORY Why Compression? Task: storage and transport of multimedia information. E.g.: non-interlaced HDTV: 0x0x0x = Mb/s!! Solutions: Develop technologies for higher bandwidth Find

More information

ECE 4400:693 - Information Theory

ECE 4400:693 - Information Theory ECE 4400:693 - Information Theory Dr. Nghi Tran Lecture 8: Differential Entropy Dr. Nghi Tran (ECE-University of Akron) ECE 4400:693 Lecture 1 / 43 Outline 1 Review: Entropy of discrete RVs 2 Differential

More information

Lecture 1: September 25, A quick reminder about random variables and convexity

Lecture 1: September 25, A quick reminder about random variables and convexity Information and Coding Theory Autumn 207 Lecturer: Madhur Tulsiani Lecture : September 25, 207 Administrivia This course will cover some basic concepts in information and coding theory, and their applications

More information

A Practical and Optimal Symmetric Slepian-Wolf Compression Strategy Using Syndrome Formers and Inverse Syndrome Formers

A Practical and Optimal Symmetric Slepian-Wolf Compression Strategy Using Syndrome Formers and Inverse Syndrome Formers A Practical and Optimal Symmetric Slepian-Wolf Compression Strategy Using Syndrome Formers and Inverse Syndrome Formers Peiyu Tan and Jing Li (Tiffany) Electrical and Computer Engineering Dept, Lehigh

More information

P P P NP-Hard: L is NP-hard if for all L NP, L L. Thus, if we could solve L in polynomial. Cook's Theorem and Reductions

P P P NP-Hard: L is NP-hard if for all L NP, L L. Thus, if we could solve L in polynomial. Cook's Theorem and Reductions Summary of the previous lecture Recall that we mentioned the following topics: P: is the set of decision problems (or languages) that are solvable in polynomial time. NP: is the set of decision problems

More information

6.867 Machine learning, lecture 23 (Jaakkola)

6.867 Machine learning, lecture 23 (Jaakkola) Lecture topics: Markov Random Fields Probabilistic inference Markov Random Fields We will briefly go over undirected graphical models or Markov Random Fields (MRFs) as they will be needed in the context

More information

Information Theory. Lecture 5 Entropy rate and Markov sources STEFAN HÖST

Information Theory. Lecture 5 Entropy rate and Markov sources STEFAN HÖST Information Theory Lecture 5 Entropy rate and Markov sources STEFAN HÖST Universal Source Coding Huffman coding is optimal, what is the problem? In the previous coding schemes (Huffman and Shannon-Fano)it

More information

10-704: Information Processing and Learning Fall Lecture 9: Sept 28

10-704: Information Processing and Learning Fall Lecture 9: Sept 28 10-704: Information Processing and Learning Fall 2016 Lecturer: Siheng Chen Lecture 9: Sept 28 Note: These notes are based on scribed notes from Spring15 offering of this course. LaTeX template courtesy

More information

CS Lecture 4. Markov Random Fields

CS Lecture 4. Markov Random Fields CS 6347 Lecture 4 Markov Random Fields Recap Announcements First homework is available on elearning Reminder: Office hours Tuesday from 10am-11am Last Time Bayesian networks Today Markov random fields

More information

CSCI 2570 Introduction to Nanocomputing

CSCI 2570 Introduction to Nanocomputing CSCI 2570 Introduction to Nanocomputing Information Theory John E Savage What is Information Theory Introduced by Claude Shannon. See Wikipedia Two foci: a) data compression and b) reliable communication

More information

Lecture 22: Final Review

Lecture 22: Final Review Lecture 22: Final Review Nuts and bolts Fundamental questions and limits Tools Practical algorithms Future topics Dr Yao Xie, ECE587, Information Theory, Duke University Basics Dr Yao Xie, ECE587, Information

More information

10-704: Information Processing and Learning Fall Lecture 10: Oct 3

10-704: Information Processing and Learning Fall Lecture 10: Oct 3 0-704: Information Processing and Learning Fall 206 Lecturer: Aarti Singh Lecture 0: Oct 3 Note: These notes are based on scribed notes from Spring5 offering of this course. LaTeX template courtesy of

More information

Lecture 1 : Probabilistic Method

Lecture 1 : Probabilistic Method IITM-CS6845: Theory Jan 04, 01 Lecturer: N.S.Narayanaswamy Lecture 1 : Probabilistic Method Scribe: R.Krithika The probabilistic method is a technique to deal with combinatorial problems by introducing

More information

Entropy as a measure of surprise

Entropy as a measure of surprise Entropy as a measure of surprise Lecture 5: Sam Roweis September 26, 25 What does information do? It removes uncertainty. Information Conveyed = Uncertainty Removed = Surprise Yielded. How should we quantify

More information