Daily Telegraph report (2007)
Why you never forget how to ride a bike
Wednesday 3 January 2007
By Roger Highfield
It is said that you never forget to ride a bike. And when you get back on a saddle after many New Year celebrations, relearning how to cycle does indeed occur surprisingly quickly.
Now researchers think they know why we pick up unused skills so rapidly, providing new insights into human memory that could help give machines improved powers of recall, paradoxically by pruning brain connections.
Using mathematics and computer models of the brain called neural networks, two scientists at the universities of Sheffield and St Andrews have produced a proof in the journal Neural Computation that relearning some parts of a partially forgotten set of memories induces the spontaneous and lasting recovery of the remaining memories.
As riding a bike consists of a set of memories of feelings and movements, so called "sensory-motor memories", the results of Drs Jim Stone and Peter Jupp's results provide an account of why such memories seem to recover with so little practice. Or, as they put it, "relearning parts of an old skill could in principle provide a free lunch in the form of spontaneous recovery of the skill."
The reason that memories seem to be so enduring relies crucially on the fact that memories are diffuse – they are locked up in a pattern of connections between many brain cells.
As the brain stores information in the form of these "distributed representations", each brain cell contributes to the storage of many associations, so that relearning some old and partially forgotten associations affects other old associations.
Using neural network models, Drs Stone and Jupp have shown that relearning some associations does not disrupt other stored memories, but actually restores them. Because there are many types of artificial neural network, it could be that this effect applies only to certain types of network. But, remarkably, their mathematical proofs are independent of the particular type of network considered, so their results ought to apply to the vast networks of neurons in the human brain too.
The current work builds on earlier research that has shown free lunch learning in people. "This is the first mathematical proof that a lasting spontaneous recovery is not just possible, but inevitable.", said Dr Stone.
As an added bonus, their research also shows that, paradoxically, extra connections reduce the size of the free-lunch effect. This provides a neat account of the pruning of synapses (connections) in the brain, where redundant synaptic connections between brain cells are removed. Synaptic pruning eliminates unused synaptic connections while frequently used connections are kept and strengthened.
Our experiences shape which connections will be strengthened and which will be pruned; connections that have been activated most frequently are preserved. Until today's work, this pruning was thought to occur only to minimise the energy demands of maintaining unnecessary neuronal machinery.