Study: neuroscientists develop equation for predicting future disasters

A team of neuroscientists has published a paper claiming it has developed a mathematical calculation that could potentially predict the tipping point of any massive event -- from a market crash to a brain seizure. Neuroscientists are used to working on systems of nodes, whereby one neuron in the brain ignites a stream of connected activity when activated -- a web-like chain reaction that unfolds in seemingly unpredictable ways. Of course, these can be predicted if you have the right information. For instance, when a team of neuroscientists monitored the brain activity of a macaque while its hand was being touched in a recent study, the team could identify which chain reaction of electrical stimulation in the brain signified that sensation. By replicating the same pattern, the team could cause the macaque to "feel" that same sensation artificially. Now, in a study let by professors at the University of Sussex, along with colleagues working in psychology and physics, that kind of pattern hunting has been translated into a computer simulation that those behind it say could one day "predict calamitous events before they happen".

The team, also made up of professors from the Sackler Centre for Consciousness Science and the Centre for Research in Complex Systems at Charles Sturt University in Australia, developed an equation that revealed the effects of information flow between multiple nodes. According to Lionel Barnett, lead author on the paper, they found the fact that all the elements "casually influence each other" to be of most importance. It means we must first identify all the parts of a system, then assess the relationship between individual nodes and then their causal effect on the whole. In doing so, we can find out when the fate of a node is dependant on its own behaviour because it behaves so differently from the others, and when its fate is dependant on all other nodes.

"The dynamics of complex systems -- like the brain and the economy -- depend on how their elements causally influence each other; in other words, how information flows between them," said Barnett.

The team suggests it's possible to measure when a system reaches that tipping point, when it moves from a healthy system to one that is overwhelmingly indicating a change. It occurs when an overwhelming number of nodes have caused an integrated change too big to remain stable. The equation was tested, but not using seizure or financial data. It was tested using a model physicists use to predict "phase transitions" in standard systems, know as the Ising model. It's a model of a ferromagnetic material that depicts how atoms in a one-, two- and three-dimensional lattices interact, and how these small phase transitions together generate a total state change -- creating a magnetic field.

"The Ising model is a very standard model in physics for analysing phase transitions from disordered to ordered states," explains Anil Seth, codirector of the Sackler Centre. "It is based on how magnetic spins 'line up'. What we did was use this model as a very rigorous test of whether different measures of 'information flow' peaked at or to one side of the phase transition. While some of these results could be established by analytical maths, other had to be obtained by computer simulation. And here we were very rigorous, performing 200 simulations for a variety of different sizes of the Ising model -- the latter helps exclude the possibility that our results are due to performing 'finite' simulations leading to ceiling effects and the like."

Using supercomputers at the Charles Sturt University in Australia, the team found that one measure called "global transfer entropy flow" reached a peak, repeatedly, "on the disordered side of the transition -- just before the tipping point".

It's the density of the information flow that anticipates the tipping point -- "all other measures peak strictly at the tipping point itself" explained Seth.

read more at link: http://www.wired.co.uk/news/archive/2013-10/28/predicting-disasters