Newsgroups: comp.ai.philosophy
From: ohgs@chatham.demon.co.uk (Oliver Sparrow)
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!pipex!peernews.demon.co.uk!chatham.demon.co.uk!ohgs
Subject: Two impressive snippets
Organization: Royal Institute of International Affairs
Reply-To: ohgs@chatham.demon.co.uk
X-Newsreader: Demon Internet Simple News v1.27
Lines: 69
X-Posting-Host: chatham.demon.co.uk
Date: Fri, 24 Feb 1995 15:44:33 +0000
Message-ID: <793640673snz@chatham.demon.co.uk>
Sender: usenet@demon.co.uk

I was delighted by two papers in the week's science press.

1: Gauging similarity with n-Grams: Language independent categorisation of 
   text. (Damashek M Sceince 267 10 Feb pp 843-848)

   An nGram is a template that can be applied "ABBIT EATS GRA" onto text and
   hits scored. Applying large numbers of nGrams to text and managing the
   statistics of what results from this, the author shows how the similarity of 
   texts can be determined, both in respect of their language and their 
   content. He shows that this technique allows trees to be built which permit 
   clusters to be explored which are found to address closely linked topics, 
   but tha the system which does this has no idea of the content of what is 
   being searched and is completely language independent. (Naturally, it 
   cannot easily match French and English versions of the same doucument, but 
   will find same-language matches for either.

   The lesson is that one can surely get too smart for one's own good. This 
   subtle application of brute strength is PC-implementable, and I could see it 
   interrupting something that I was typing to say that I had already said that 
   a years ago and would I like to look? This would happen rather a lot, I fear.

2: Whittington M Traub R Jefferys J Nature 373 pp 612-615 16 Feb 1995.
   I quote from the covering review paper by  Douglas and Martin.

   " A central problem in memory is how information is represented by neurons. 
   One way in which memory could work is that neurons ...could learn to 
   recognise the various things that make up our world..... But the number of 
   such high level feature detectors scales unfavourably with the number of 
   things to be encoded, and so most theories use combinations of neurons, each 
   of which represents a sngle feature, to code for different things. Each  
   neuron then particpates in the representations of many different things.

   ...Population coding .. seems to require a mechanism by which neurons which 
   represent one things are related or are "bound together". One theory is that 
   neurons signal their relatedness by discharging nerve impulses in synchrony 
   with each other at about 40 Hz. This is an attractive theory because 'brain 
   waves' of various frequencies have been made since ... the 1920s, but the 
   function of these has never been clear.

   Quite simple feedback systems are prone to oscillation, so .. it is not 
   surprising that cortical circuits do the sdame. And it is exactly these 
   recurrent circuits .... that may provide a basis for associative recall. 
   Unfortunately, classicla associative memories recall entire input scenes as 
   single memories. ..This restriction severely limits the ability of the 
   associative networks to generalise because commonly occuring components of 
   complex scenes will not be recognised in new contexts. So the challenge is 
   to find a method that enables an associative memory to segment a scene into 
   a number of components and to represent them seperately and simultaneously . 
   Theoreticians have demonstrated that is associative elements are allowed to 
   oscillate, then this problem can be partly solved.

   ... A key link between collective oscillations and perception was made 
   when ....noticed that the cortical neurons in the visual system responded 
   to coherent visual stimuli by discharging synchonously at about 40 Hz. 
   Incoherent stimuli.. did not elicit .. discharge. This suggests that 
   synchronisation reflects a transient binding together of reverberating 
   groups of neurons, each of which responds to different features in the same 
   perceptual object. Whittington et al have now been able to demonstrate that 
   these 40Hz oscillations can be evoked in thin slices of brain . They have 
   been able to  show ... that only a small subset ... of the neurons are 
   needed to generate the 40 Hz oscillation of the network. The way is now open 
  for direct studies of the details of the oscillatory mechanims and its 
  possible oscillatory function...


   _________________________________________________

  Oliver Sparrow
  ohgs@chatham.demon.co.uk
