# 0.3 Universal coding for classes of sources

 Page 2 / 3

where $Q$ is the prior induced by the coding length function $l$ .

## Minimal redundancy

Note that

$\begin{array}{ccc}\hfill \forall w,l,\phantom{\rule{1.em}{0ex}}\underset{\theta }{sup}{r}_{n}\left(l,\theta \right)& \ge & {\int }_{\Lambda }w\left(d\theta \right){r}_{n}\left(l,\theta \right)\hfill \\ & \ge & \underset{l\in {c}_{n}}{inf}{\int }_{\Lambda }w\left(d\theta \right){r}_{n}\left(l,\theta \right).\hfill \end{array}$

Therefore,

${R}_{n}^{+}=\underset{l}{inf}\underset{\theta }{sup}{r}_{n}\left(l,\theta \right)\ge \underset{w}{sup}\underset{l}{inf}{\int }_{\Lambda }w\left(d\theta \right){r}_{n}\left(l,\theta \right)={R}_{n}^{-}.$

In fact, Gallager showed that ${R}_{n}^{+}={R}_{n}^{-}$ . That is, the min-max and max-min redundancies are equal.

Let us revisit the Bernoulli source ${p}_{\theta }$ where $\theta \in \Lambda =\left[0,1\right]$ . From the definition of [link] , which relies on a uniform prior for the sources, i.e., $w\left(\theta \right)=1,\forall \theta \in \Lambda$ , it can be shown that there there exists a universal code with length function $l$ such that

${E}_{\theta }\left[l\left({x}^{n}\right)\right]\le n{E}_{\theta }\left[{h}_{2},\left(\frac{{n}_{x}\left(1\right)}{n}\right)\right]+log\left(n+1\right)+2,$

where ${h}_{2}\left(p\right)=-plog\left(p\right)-\left(1-p\right)log\left(1-p\right)$ is the binary entropy. That is, the redundancy is approximately $log\left(n\right)$ bits. Clarke and Barron  [link] studied the weighting approach,

$p\left(x\right)={\int }_{\Lambda }dw\left(\theta \right){p}_{\theta }\left(x\right),$

and constructed a prior that achieves ${R}_{n}^{-}={R}_{n}^{+}$ precisely for memoryless sources.

Theorem 5 [link] For memoryless source with an alphabet of size $r$ , $\theta =\left(p\left(0\right),p\left(1\right),\cdots ,p\left(r-1\right)\right)$ ,

$n{R}_{n}^{-}\left(w\right)=\frac{r-1}{2}log\left(\frac{n}{2\pi e}\right)+{\int }_{\Lambda }w\left(d\theta \right)log\left(\frac{\sqrt{|I\left(\theta \right)|}}{w\left(\theta \right)}\right)+{O}_{n}\left(1\right),$

where ${O}_{n}\left(1\right)$ vanishes uniformly as $n\to \infty$ for any compact subset of $\Lambda$ , and

$I\left(\theta \right)\triangleq E\left[\left(\frac{\partial ln{p}_{\theta }\left({x}_{i}\right)}{\partial \theta }\right),{\left(\frac{\partial ln{p}_{\theta }\left({x}_{i}\right)}{\partial \theta }\right)}^{T}\right]$

is Fisher's information.

Note that when the parameter is sensitive to change we have large $I\left(\theta \right)$ , which increases the redundancy. That is, good sensitivity means bad universal compression.

Denote

$J\left(\theta \right)=\frac{\sqrt{|I\left(\theta \right)|}}{{\int }_{\Lambda }\sqrt{|I\left({\theta }^{\text{'}}\right)|}d{\theta }^{\text{'}}},$

this is known as Jeffrey's prior . Using $w\left(\theta \right)=J\left(\theta \right)$ , it can be shownthat ${R}_{n}^{-}={R}_{n}^{+}$ .

Let us derive the Fisher information $I\left(\theta \right)$ for the Bernoulli source,

$\begin{array}{ccc}& & {p}_{\theta }\left(x\right)={\theta }^{{n}_{x}\left(1\right)}·{\left(1-\theta \right)}^{{n}_{x}\left(0\right)}\hfill \\ & ⇒& ln{p}_{\theta }\left(x\right)={n}_{x}\left(1\right)ln\theta +{n}_{x}\left(0\right)ln\left(1-\theta \right)\hfill \\ & ⇒& \frac{\partial ln{p}_{\theta }\left(x\right)}{\partial \theta }={n}_{x}\left(1\right)\frac{1}{\theta }-{n}_{x}\left(0\right)\frac{1}{1-\theta }\hfill \\ & ⇒& {\left(\frac{\partial ln{p}_{\theta }\left(x\right)}{\partial \theta }\right)}^{2}=\frac{{n}_{x}^{2}\left(1\right)}{{\theta }^{2}}+\frac{{n}_{x}^{2}\left(0\right)}{{\left(1-\theta \right)}^{2}}-\frac{2{n}_{x}\left(1\right){n}_{x}\left(0\right)}{\theta \left(1-\theta \right)}\hfill \\ & ⇒& E\left[{\left(\frac{\partial ln{p}_{\theta }\left(x\right)}{\partial \theta }\right)}^{2}\right]=\frac{\theta }{{\theta }^{2}}+\frac{1-\theta }{{\left(1-\theta \right)}^{2}}-\frac{2}{\theta \left(1-\theta \right)}E\left[{n}_{x}\left(1\right){n}_{x}\left(0\right)\right]\hfill \\ & & =\frac{1}{\theta }+\frac{1}{1-\theta }-0\hfill \\ & & =\frac{1}{\theta \left(1-\theta \right)}.\hfill \end{array}$

Therefore, the Fisher information satisfies $I\left(\theta \right)=\frac{1}{\theta \left(1-\theta \right)}$ .

Recall the Krichevsky–Trofimov coding, which was mentioned in  [link] . Using the definition of Jeffreys' prior [link] , we see that $J\left(\theta \right)\propto \frac{1}{\sqrt{\theta \left(1-\theta \right)}}$ . Taking the integral over Jeffery's prior,

$\begin{array}{ccc}\hfill {p}_{J}\left({x}^{n}\right)& =& {\int }_{0}^{1}c\frac{d\theta }{\sqrt{\theta \left(1-\theta \right)}}{\theta }^{{n}_{x}\left(1\right)}{\left(1-\theta \right)}^{{n}_{x}\left(0\right)}\hfill \\ & =& c{\int }_{0}^{1}{\theta }^{{n}_{x}\left(1\right)-\frac{1}{2}}{\left(1-\theta \right)}^{{n}_{x}\left(0\right)-\frac{1}{2}}d\theta \hfill \\ & =& \frac{\Gamma \left({n}_{x}\left(0\right)+\frac{1}{2}\right)\Gamma \left({n}_{x}\left(1\right)+\frac{1}{2}\right)}{\pi \Gamma \left(n+1\right)},\hfill \end{array}$

where we used the gamma function. It can be shown that

${p}_{J}\left({x}^{n}\right)=\prod _{t=0}^{n}{p}_{J}\left({x}_{t+1}|{x}_{1}^{t}\right),$

where

$\begin{array}{ccc}\hfill {p}_{J}\left({x}_{t+1}|{x}_{1}^{t}\right)& =& \frac{{p}_{J}\left({x}_{1}^{t+1}\right)}{{p}_{J}\left({x}_{1}^{t}\right)},\hfill \\ \hfill {p}_{J}\left({x}_{t+1}=0|{x}_{1}^{t}\right)& =& \frac{{n}_{x}^{t}\left(0\right)+\frac{1}{2}}{t+1},\hfill \\ \hfill {p}_{J}\left({x}_{t+1}=1|{x}_{1}^{t}\right)& =& \frac{{n}_{x}^{t}\left(1\right)+\frac{1}{2}}{t+1}.\hfill \end{array}$

Similar to before, this universal code can be implemented sequentially. It is due to Krichevsky and Trofimov  [link] , its redundancy satisfies   Theorem 5 by Clarke and Barron  [link] , and it is commonly used in universal lossless compression.

## Rissanen's bound

Let us consider – on an intuitive level – why ${C}_{n}\approx \frac{r-1}{2}\frac{log\left(n\right)}{n}$ . Expending $\frac{r-1}{2}log\left(n\right)$ bits allows to differentiate between ${\left(\sqrt{n}\right)}^{r-1}$ parameter vectors. That is, we would differentiate between each of the $r-1$ parameters with $\sqrt{n}$ levels. Now consider a Bernoulli RV with (unknown) parameter $\theta$ .

One perspective is that with $n$ drawings of the RV, the standard deviation in the number of 1's is $O\left(\sqrt{n}\right)$ . That is, $\sqrt{n}$ levels differentiate between parameter levels up to a resolution that reflects the randomness of the experiment.

A second perspective is that of coding a sequence of Bernoulli outcomes with an imprecise parameter,where it is convenient to think of a universal code in terms of first quantizing the parameter and then using that (imprecise) parameter to encode the input $x$ . For the Bernoulli example, the maximum likelihood parameter ${\theta }_{ML}$ satisfies

anyone know any internet site where one can find nanotechnology papers?
research.net
kanaga
Introduction about quantum dots in nanotechnology
what does nano mean?
nano basically means 10^(-9). nanometer is a unit to measure length.
Bharti
do you think it's worthwhile in the long term to study the effects and possibilities of nanotechnology on viral treatment?
absolutely yes
Daniel
how to know photocatalytic properties of tio2 nanoparticles...what to do now
it is a goid question and i want to know the answer as well
Maciej
Abigail
for teaching engĺish at school how nano technology help us
Anassong
Do somebody tell me a best nano engineering book for beginners?
there is no specific books for beginners but there is book called principle of nanotechnology
NANO
what is fullerene does it is used to make bukky balls
are you nano engineer ?
s.
fullerene is a bucky ball aka Carbon 60 molecule. It was name by the architect Fuller. He design the geodesic dome. it resembles a soccer ball.
Tarell
what is the actual application of fullerenes nowadays?
Damian
That is a great question Damian. best way to answer that question is to Google it. there are hundreds of applications for buck minister fullerenes, from medical to aerospace. you can also find plenty of research papers that will give you great detail on the potential applications of fullerenes.
Tarell
what is the Synthesis, properties,and applications of carbon nano chemistry
Mostly, they use nano carbon for electronics and for materials to be strengthened.
Virgil
is Bucky paper clear?
CYNTHIA
carbon nanotubes has various application in fuel cells membrane, current research on cancer drug,and in electronics MEMS and NEMS etc
NANO
so some one know about replacing silicon atom with phosphorous in semiconductors device?
Yeah, it is a pain to say the least. You basically have to heat the substarte up to around 1000 degrees celcius then pass phosphene gas over top of it, which is explosive and toxic by the way, under very low pressure.
Harper
Do you know which machine is used to that process?
s.
how to fabricate graphene ink ?
for screen printed electrodes ?
SUYASH
What is lattice structure?
of graphene you mean?
Ebrahim
or in general
Ebrahim
in general
s.
Graphene has a hexagonal structure
tahir
On having this app for quite a bit time, Haven't realised there's a chat room in it.
Cied
what is biological synthesis of nanoparticles
what's the easiest and fastest way to the synthesize AgNP?
China
Cied
types of nano material
I start with an easy one. carbon nanotubes woven into a long filament like a string
Porter
many many of nanotubes
Porter
what is the k.e before it land
Yasmin
what is the function of carbon nanotubes?
Cesar
I'm interested in nanotube
Uday
what is nanomaterials​ and their applications of sensors.
Got questions? Join the online conversation and get instant answers!