HOW OUR BRAINS SOLVE SUDOKU

PRINTABLE SUDOKU BRINGS YOU AN ARTICLE ON HOW OUR FORMIDABLE BRAINS SOLVE SUDOKU...

Scientists have discovered that solving sudoku puzzles depends on neural pathways that even the most powerful computers can't replicate.


Researchers  say that by studying how people solve the puzzles, they might be able to develop more intelligent and brain-like computers.

In a recently published study, Professor John Hopfield of Princeton University explored the unique brain processes we use when playing Sudoku.  This mathematical puzzle involves filling in a grid of 81 squares with varying combinations of the numbers one to nine, something that sounds simple but can be diabolically hard.

To crack Sudoku our brains use a unique set of neural pathways known as associative memory, Hopfield says, which enables us to discover a pattern from a partial clue.

Although computers can store large amounts of information and process it at great speed, they aren't yet capable of sophisticated associative memory. Hopfield provides an algorithm of associative memory in his paper, which he says could be implemented in silicon chips.

Patterns -We all recognise the basic pattern of counting from one to nine, yet the task of completing a Sudoku puzzle is confounded because of the large number of possible permutations of this pattern.

But every time we put the right number in the right place it provides us with a clue, which reduces the number of permutations.

In this way Sudoku is based on a combination of logic and intelligent guesswork based on our abilities of associative memory, Hopfield says.

"In neural terms, the signals developed ... can produce a strong and reasonably accurate feeling of correctness of the item retrieved," Hopfield says.

"This fact may account for our strong psychological feeling of 'right' or 'wrong' when we retrieve a memory from a minimal clue."

Brains versus computers - Associate Professor Andrew Paplinski is an Australian computer scientist who specialises in neural networks at Monash University in Melbourne.

He says the process described in Hopfield's paper helps us to remember a name from a fragment or recognise a partially obscured face.

He says applying Hopfield's model could lead to more accurate facial recognition computer technology.

Being able to mimic associative memory would give computers "extreme robustness of pattern recognition", Paplinski says.

For example, for a computer to recognise a partially visible face it would first have to recognise that the face is obscured, then that it is a face, and then it would have to find a match.

"To answer all these questions takes an enormous amount of computation," he says.

He says we can do this in a fraction of a second in a slow computer like our brain. So there were would be significant implications if we can figure out how this is done and design computers that can replicate it.