What is a WISARD Neural Network? I have been trying to find good resources on the internet describing what Wisard Neural Networks are and their differences with traditional Neural Networks but failed to find anything substantial. Can anybody please explain their main characteristics please? 
 A: IIRC WISARD was a RAM based neural network method developed by Igor Aleksander at Imperial College in the 1980s and early 1990s.  RAM based neural networks essentially use look up tables to store the function computed by each neuron, and hence are easily implemented in digital hardware and have efficient training algorithms.  They don't seem to be used much these days.  In think the best source of information would be Aleksander and Morton's book "Introduction to neural computing" (sadly I can't find my copy).  I don't think these sort of networks are used much these days, which is a pity as they are rather interesting;  my favourite variant was the "Single layer look up perceptron", which is related to n-tuple classifiers (see this paper, and this one, for details.)  They seem reminiscent in some ways to random forest methods and also "extreme learning" methods (as a linear model is trained using a fixed set of random basis functions).
A Google scholar search for Igor Aleksander WISARD should find most of the relevant material.
A: A major working development to date (1985) in this area however is the N-tuple classifier: "WISARD" (WIlkie, Stoneham and Aleksander, Recognition Device), which has been developed at Brunel University and is produced commercially by Computer Recognition Systems Limited. 
The WISARD system acts upon images from standard grey scale TV cameras working at 30 fps. For each frame, the image is binarised. Then each of the pixels of the stored image are pseudo randomly sampled (mapped - ie chosen from the whole frame area using a pseudo random number to choose each pixel in the frame) and formed into K groups (tuples) of N (where N is an integer - normally ranged between 1 .. . . .. 8). The data pattern formed by each N-tuple is applied as an address to a Random Access Memory (RAM) element. (Note: a pseudo random number generator will always deliver an identical non repeating stream of numbers starting from the same seed number).
The size of this memory element is 2^N bits where N is the size of the tuple being used (thus a 4-tuple will use a 16 bit sized memory element and an 8-tuple will use a 256-bit memory element), and a "1" will be written to the bit in this memory space corresponding to the N sized address formed by the tuple (all other spaces initially set to zero). The memory space of the machine is partitioned into C groups of K addresses (where K = the number of pixels in the image and K*N = the vector space of one stored image). Each of these groups of memory addresses is termed a Discriminator.
During the training stage, each object that is to be recognised by the system in the classification mode is sampled a number of times (a training  set) and allocated one of the C groups of memory and becomes a discriminator for that object.
If during the repeated sampling (using the identical pseudo random mapping) the addresses formed by the N-tuples are slightly different, additional "1"s are placed in the memory elements (it is expected that this will occur otherwise the method will be little better than Template matching).
When the machine is used for recognition, using identical pseudo random mapping the generated N-tuples from the object to be recognised are applied to their corresponding memory elements within all the trained discriminators. Where there is a match (of “1”s) a memory element will return a "1" or otherwise "0". For each discriminator a summation is taken and a histogram produced of the results. A discriminator producing a high response is likely to be the object being classified.
from my 1985 Dissertation “A study of the INMOS Transputer in robotic vision” pp7 – 11 
A: WISARD was  actually  built   by   Bruce   Wilkie  , Sonham asreader  , Alexandre  getting the funding .  Eric Conan being one of technicians  building it  .
It was  presented at  Royal Institution , Maggie  had a  look
Wi     Wilkie
S      Stonham
A      Alexandre 
R      recognition
D      discriminator -
two  19 inch  racks  of electronics  -  massive transformer  with massive  copper  busbars  and   smoothing capacitor  - linerar   regulator  - with  tv camera   input to TV   display 
Sets of  RAM     -  images to  compatre  .  Randomly  selecting  content  ( bit wise )   and  counting hits to misses  D
A: WISARD stamds for WILKIE STONHAM AND ALEXANDER. A Neural network based on hardware RAMs hardwired to be a perceptron system. This is described in one of ALEXANDERS own books. The difference that it made was to bring back the industry's attention to Neural Networks after the disastrous paper issued by Prof. Marvin and Minsky on the linearity of Neural Networks in other words that any problem that they could solve would be linear and non linear. The WISARD system proved that NN work and it strated a new silemt revolution at the time. 
