Re: various or "Still fuzzy?"

[Thanh Tung Truong <tung.truong@e-technik.tu-chemnitz.de>]

Hi Bac Zung,

> Ba'c Thanh Tung vie^'t ca'i posting sau ve^` fuzzy ho+i kho' hie^?u.

I am sorry but it is not easy to explain some technical problems. I am
trying my best. You may want to read the previous messages in the thread
once again.

> To^i kho^ng tha^'y ta.i sao fuzzy thi` la.i pha?i parallel (hai ca'i
> na`y theo to^i ddo^.c la^.y vo+'i nhau, co`n ne^'u fuzzy models tie^.n
> cho parallel processing kho^ng thi` la` chuye^.n kha'c - cha(?ng ha.n
> to^i tha^'y object-oriented tech. tie^.n cho parralel vi` ca'c objects
> la^'y luo^n la`m thread trong multi-threading la` ra^'t tu+. nhie^n,
> nhu+ng to^i chu+a tha^'y fuzzy tie^.n hay kho^ng tie^.n cho parralel o+?
> cho^~ na`o).

In a fuzzy algorithm, there are several data streams to be independently
processed by a pipeline of several complex operations, i.e. 
fuzzification,
implication, composition and defuzzification. Therefore it is possible 
AND
desirable (from the viewpoint of processing time) to implement a fuzzy
algorithm by parallel techniques.
For example consider a fuzzy system with N inputs and M output. Each 
input
is modelled by P fuzzy sets and each output by Q fuzzy sets. In general,
there are P^N*M linguistic rules with N premises and one conclusion each.
Obviously:
the fuzzification should be applied to N*P independent data streams,
the (vectorial) implication should be applied to P^N*M independent data
streams,
the conclusion should be applied to M*Q independent data streams,
the defuzzification should be applied to M independent data streams.
As can be seen, the number of operations can be very high.

> Theo to^i fuzzy la` 1 technical term bie^?u hie^.n su+.
> chia nho? ho+n ca'c mu+'c dda'nh gia' tho^i (cha? ha.n thay vi` no'i
> thie^.n -a'c ta no'i ra^'t thie^.n, ho+i thie^.n, ho+i a'c, ra^'t a'c).
> Tuy` theo chia nho? bao nhie^u ma` lu+o+.g tho^ng tin co' va` ca^`n xu+?
> ly' ta(ng le^n tu+o+ng u+'ng, va` lu+o+ng primitive gates trong circuits
> ca^`n ta(ng le^n tu+o+ng u+'ng (va^'n dde^` o+? dda^y cha('c kho^ng
> pha?i cha.y nhanh hay cha^.m nhu+ ba'c Vie^.t vie^'t, ma` la` xu+? ly'
> approximate chi'nh xa'c dde^'n bao nhie^u).

The processing speed is one of the main motivation for dedicated fuzzy
hardware. The processing accuracy is not becaus a virtually arbitrary
accuracy can be reach with conventional digital hardware. It may be a
problem only if analog hardware is used owing to nonlinear effects, drift
and error propagation. A high processing speed is very important for
high-impact applications such as image processing or pattern recognition.


> Ca'c implementations cu?a fuzzy hie^.n nay cha('c dde^`u pha?i digital
> tho^i (chi'nh xa'c ho+n: analog -> digital -> mechanic or analog...).

That depends. For low-cost consumer products where the demand for 
accuracy
and/or programability is low and/or very high speed is required, analog
implementation could be the choice.

> To^i ddo.c tho^ng tin ve^`
> ma^'y ca'i ma'y a?nh thi` tha^'y ho. vie^'t fuzzy logic = automatic
> control of light! Kha'i nie^.m fuzzy va` quantum computing  cha('c hoa`n
> toa`n ddo^.c la^p vo+'i nhau (ie you can have fuzzy without quantum and
> vice versa,  and their
> basic ideas are also absolutely different). dda~ la` microprocessor thi`
> ba?n tha^n no' cha('c kho^ng the^? fuzzy, chi? co' the^? du`ng no' ti'nh
> ca'c "qua' tri`nh fuzzy" ddu+o+.c tho^i.

Correct, fuzzy hardware is physically not fuzzy at all. It is just a 
rather
"fuzzy" technical term.

Hope you find that not so fuzzy.

Thanh-Tung Truong