July 6, 2016 at 8:27 pm
Comments posted to this topic are about the item Fuzzy Processing
July 7, 2016 at 12:49 am
I'm not sure about chips having constrained accuracy but there is certainly merit in new search capabilities that approximate a match.
Bloom filters are something I would like to see in SQL Server. Full text search offers huge opportunities for improvement. Geospatial indexes and functionality offers huge potential.
I wonder what other index types are out there that would offer SQL Server a significant boost in capability and performance?
July 7, 2016 at 3:19 am
Peoples' names would be another, especially if it needs to handle misspellings, contractions and variants. There is also the issue as to whether middlenames are used.
July 7, 2016 at 6:11 am
crmitchell (7/7/2016)
Peoples' names would be another, especially if it needs to handle misspellings, contractions and variants. There is also the issue as to whether middlenames are used.
I was using Soundex for this 20 years ago to overcome issues with Mc and Mac.
On a more general note I think to use of 'fuzzy logic' has to be careful and considered. Whilst I can see its uses if someone is wrongly arrested/convicted as a result of 'fuzzy logic' it would do untold damage to the data industry!
July 7, 2016 at 6:39 am
you forgot Mc not that you see it used often nowadays.
There have been enough cases of incorrect identity that you would hope that any warrant to arrest Joanne Bloggs instead of Joseph Bloggs should be checked before being issued.
July 7, 2016 at 6:42 am
Over forty years ago one of my advanced math classes had a problem about how to aim a moving sub's torpedo at a moving ship and hit it. Of course if you aim directly at the ship then by the time the torpedo gets there the ship has moved on. They solved this in various ways one way mathematically is using a formula that you keep re-iterating through using each result set as input into the next operation of the calculation until the margin of error becomes small enough that it is acceptable. There wasn't the exactness of adding up a column of numbers but the idea of approximate mathematically results is certainly not something new as in the image comparison example. You just keep comparing until your margin of error becomes small enough that it reaches a preset value.
July 7, 2016 at 8:00 am
Watch the movie "Brazil" for an example of fuzzy logic and mistaken identity. It's a hoot.
July 7, 2016 at 8:30 am
Iwas Bornready (7/7/2016)
Over forty years ago one of my advanced math classes had a problem about how to aim a moving sub's torpedo at a moving ship and hit it. Of course if you aim directly at the ship then by the time the torpedo gets there the ship has moved on. They solved this in various ways one way mathematically is using a formula that you keep re-iterating through using each result set as input into the next operation of the calculation until the margin of error becomes small enough that it is acceptable. There wasn't the exactness of adding up a column of numbers but the idea of approximate mathematically results is certainly not something new as in the image comparison example. You just keep comparing until your margin of error becomes small enough that it reaches a preset value.
If you are at all interested in this it is worth reading "Dreadnougt Gunnery and the Battle of Jutland" by Professor John Brooks. It is an extremely interesting book about the Dreyer table (think mechanical computer) and its development saga.
July 7, 2016 at 8:35 am
dld (7/7/2016)
Watch the movie "Brazil" for an example of fuzzy logic and mistaken identity. It's a hoot.
Tuttle, Tuttle
One of my favourite films 🙂
July 7, 2016 at 8:49 am
We really need a lot more research into that area, because strict algorithmic approaches are slow and can be error prone.
Thousands of species of animals from insects up through birds and bats manage complex flight 'calculations' without any algorithmic approach at all. Especially birds and bats are dealing with controlling hundreds of muscles, receiving input from hundreds or thousands of sensors (nerves), yet the whole process seems smooth and effortless.
Chips that encompass these more or less analog controls, using rough accuracy to provide efficiency will be necessary before self driving cars are really practical.
...
-- FORTRAN manual for Xerox Computers --
July 7, 2016 at 9:13 am
something along the lines of the work Google undertook to get house numbers into GoogleMaps
July 7, 2016 at 9:59 am
Iwas Bornready (7/7/2016)
Over forty years ago one of my advanced math classes had a problem about how to aim a moving sub's torpedo at a moving ship and hit it. Of course if you aim directly at the ship then by the time the torpedo gets there the ship has moved on. They solved this in various ways one way mathematically is using a formula that you keep re-iterating through using each result set as input into the next operation of the calculation until the margin of error becomes small enough that it is acceptable. There wasn't the exactness of adding up a column of numbers but the idea of approximate mathematically results is certainly not something new as in the image comparison example. You just keep comparing until your margin of error becomes small enough that it reaches a preset value.
You could have just used pigeons! https://en.wikipedia.org/wiki/Project_Pigeon
July 7, 2016 at 10:46 am
This is a really cool concept. It's basically bringing computers one step closer to how people really think.
Right now, in order to compare sets of information (say two images) computers fake fuzzy logic and guesswork through a complex progression of comparing exact values and adding in some sort of room for error whereas a person can immediately (a relative term when comparing humans to computers) discern that two things are similar using approximation. If the processor can skip the exact comparisons and go straight to approximating results then you've eliminated a significant portion of the processing time. Then by adding a good old fashioned digital processor into the mix the approximations can be further refined as needed to clarify the accuracy of any results.
This has implications all over the place but an easy one that comes to mind is real-time data compression in video applications. As sensor resolutions continue to increase the storage demands grow as well. The processing required in your average camcorder already takes up a significant portion of the battery power not to mention space required to store the raw video but if you can apply approximation techniques like this then it becomes easier to compress the data in real-time before having to write it to storage media.
Very exciting times.
July 7, 2016 at 12:17 pm
Diligently using fuzzy logic to produce an answer for the wooly thinkers
July 7, 2016 at 2:41 pm
I am not sure of the background education of everyone here, but my guess is that there are at least a few people who are unaware of the issue with doing math with a processor.
While processors absolutely are accurate when it comes to binary, they are not accurate when it comes to floating point calculations. One of my professors really enjoyed showing us how there is a margin of error. If you understand the conversion between binary and base 10, it becomes obvious why this is true. It has been too long for me to give an educated explanation of the reasoning behind it, but I am sure a Google search would provide insights.
The important thing to note is that it is absolutely possible for a computer to perform calculations and come up with a wrong answer.
Dave
Viewing 15 posts - 1 through 15 (of 20 total)
You must be logged in to reply to this topic. Login to reply