IN THE ETHER

A global archive of independent reviews of everything happening from the beginning of the millennium


Read our Copyright Notice click here

For publication dates click here


Two more for comparison: Pintia Toro 2008 and Vega Sicilia Tinto Valbuena 2006 (right)




PATTERN RECOGNITION


Reviewed by ANDRE BEAUMONT


We are in 2014 now and I am confident that we will soon have very safe, fully autonomous navigation, as in driverless cars, leaning heavily on machine learning and pattern recognition.


A fully autonomous road vehicle needs to control four mechanical variables - acceleration, braking, steering and gear changing - but much of the rest is pattern recognition allied to sufficient low latency computing power. There are alternatives here to machine learning but not to pattern recognition. Driverless cars are likely to avoid 90% or more of the road accidents that humans cause.

I am less confident that the big data enthusiasts wishing to use general practitioner medical records for data crunching will not come away disappointed.

Published literature shows that about one third of diagnoses are wrong or contain some element of error.

So if you went to the doctor with two complaints for diagnosis you would have about a 45% chance of coming away with fully correct diagnoses for both complaints.

Add to this errors resulting from transcribing old handwritten records, often more an aide memoire than a diagnosis, to digital format and futher errors in then understanding what the transcription is meant to mean and the quality and reliability of the data declines further.

Further problems are added when the underlying science on which the diagnosis has been based has - to put it politely - been superseded.

Then there is the great gullibility in so much of the health and care industry which leads it to believe that if something has been written down it must be correct.

A journalist will tell you otherwise. Float an idea at a lunch and you will be treated as eccentric or deluded. Publish it in a newspaper and you will be feted as the world's expert and the font of reliable knowledge at another lunch the next week.


Policemen's training make them much better at not jumping to wrong conclusions from evidence than the medical world.

Having said this what you are looking for in a good doctor is someone good at pattern recognition, who will overlay it with thought and critical analysis - and a good deal more.

This does not do much for the big data boys and girls, though.

A dog's nose is good at pattern recognition as are the noses of so many animals.

The human nose is now thought to be able to distinguish millions of smells or chemical compounds.

Think about how different localities have different smells and how the smell, or even the memory of the smell, brings back associations.

Hyde Park smells different from St James's Park.

If you were kidnapped and brought back blindfold and dumped in Hyde Park (or somewhere equivalent in terms of your prior knowledge) would you recognize the smell and realize you had been brought back to somewhere familiar?

A fair bit says 'yes'.


Throw a bone from literature, architecture or politics and maybe the critical language to describe it can be found but I find attempting wine criticism frustrating.

For sure this or that wine smells of citrus, strawberries or lychees and those pointing it out to you are seriously better at analysis than you but also the aromas are not quite those of these fruits either and composed of a much more variegated set of chemical compounds.

In the scheme of things going to a dinner marking the 150th anniversary of Vega Sicilia two weeks ago deserves some critical attention especially as I chatted with the bodega owner for a quarter of an hour before but how to do it?

Stumped.

Words do not equate with what the nose or the palate are telling.

Then I realized what I do wrong. Long since have I given up using words for wine. All I am doing is engaging in an extensive pattern recognition exercise against memory without word tags. It produces the right answers: the Unico Ribera del Duero 2004 is the one going into permanent memory, drunk and remembered alongside the 1994 for comparison.

Rather than make a comparison with the pattern recognition of driverless cars, can't we just have them soon to take us home?