2013/05/16 12:48:41 -0500
x - input sequence
n - length of x
x=x1x2...xn, xij=xixi+1...xj
α - discrete alphabet
r - cardinality of α
nx(a) - the number of times that a∈α appears in x
Px(a) - empirical probability
Q - the true i.i.d. distribution of x
H - entropy
D(·∥·) - divergence
⌈·⌉ - rounding up
R - rate of source code
TQ(δ) - set of inputs that are δ-typical with respect to (w.r.t.) Q
C - a code
(·)C - complement of set
·T - transpose of vector or matrix
l(x) - length of code in bits
x'=Sx - step ∀n∈Z, xn'=xn+1
θ - parameters of parametric source (can contain multiple scalar parameters)
M - number of states of unifilar source
S={1,...,M} - states of unifilar source
s1,...,sn, - state sequence
- transition probability
- next state function
Λ - class of parametric models
cn - collection of lossless codes for length-n inputs
I(·;·) - mutual information
rn(l,θ) - how far a coding length function l is from the entropy
h2(·) - binary entropy
I(·) - Fisher information
J(·) - Jeffreys' prior
θML - maximum likelihood parameter
Rn–, Rn+ - min-max and max-min redundancy
T - set of leaves of a context tree source
s - state of context tree
nx(s,a) - number of times that a∈α appears in context s in x
T* - optimal context tree source
D - maximal depth of tree source
Ts* - optimal tree structure to encode symbols whose context ends with s
MDL(s) - minimum description length required for encoding these symbols
KT(·,·) - Krichevsky-Trofimov coding length
(y,i) - BWT output consisting the permuted sequence and index
- approximate version of x
- reproduction alphabet
d(·,·) - distortion metric
- distortion between sequences
D - distortion level
R(D) - minimal rate such that can be described
t - temperature in simulated annealing
Zt - normalization factor in Boltzmann pmf
Nr - recurrence time