What mechanism can we devise to solve the problems of tomorrow – even problems which we cannot envisage today? The challenge of predictive data analytics has been exercising Saffron Technology. As Forrester’s Boris Evelson reported in ‘The Dawning of a new age of BI-DBMS‘, Saffron’s use of associative memories may be the key.
Anyone who has used statistical analysis tools to try and predict consumer goods sales orders will simply recite the old mantra: ‘one thing you can predict about sales forecasts is that they will always be wrong’. Forecasting sales demand may be easier than the weather, but at the detail level most forecasts are unreliable.
Using associative memories when you can’t predict the future
Having to be prepared for anything is a new approach to business intelligence, but has other applications too. The enquiries showed that the events of 9/11 were predictable. There was adequate intelligence data but consolidation and analysis was poor, and several cues were missed. Hindsight is 20/20, as they say, but what of memories?
Using the techniques of predictive analytics and associative memories may help to reduce the threat of such future catastrophes. Databases comprised of both structured and unstructured data are used to create memories in a way which mimics human memory, with autoassociative memory being able to retrieve the whole memory based on a small sample of itself – in simple terms, self referencing. Hetero-associative is a form of cross referencing memories.
These memory types form the basis of man made neural networks which are agile and flexible, learning in real time and making connections.With associative memories, patterns are established and cross matched, locked into a computation, almost like two-key code, with one piece being able to unlock the whole.
What have we seen before?
And based on that, what is likely to happen next?
This is a completely different approach to the traditional statistical methods including noisy and often inaccurate ‘pair-wise associations’, as well as the rigid need to frame questions:
“And what about the weather tomorrow?”
“Yes, there will be weather.”
The ‘noise’ in pair-wise associations reduces the accuracy of predictions, requiring an ever-increasing number of variations to the rule.
Rich context is the most significant factor in understanding how to use data in an actionable sense – allowing us to move beyond causal relationships between event A and event B.
By unlocking a deep understanding of context, Associative Memories help us to better anticipate what could happen, and plan accordingly.