Maybe the 8’th of August will be remembered as a historical date, as significant to the revolution of information technology as the introduction of the first transistor based computers, in the early 40’s of last century was.
IBM introduced a chip that radically changes conventional design patterns for hardware and software and might be a driving force for SAP HANA & Big Data analytics.
Simply speaking the chip is a artificial neural network on a chip. Neural Networks are very useful in pattern recognition and classification tasks. The problem so far was that they only existed in software simulations. They can be already used in the combination of SAP HANA and R for Big Data analytics. If the size if such a network is reasonably small they can even process in real time.
For more complex tasks however the processing speed of conventional computer chips is too slow. For example some years ago one group of students of mine tried to implement a neural network that will tell a blind person, wearing a device like Google Glass, if a traffic light is green or red. Another group implemented a neural network that should predict the outcome of a football match by in game parameters. Both projects were far from real world applications because the computer, required to calculate the simulation of the neural networks, was too slow, big and energy hungry. And the networks were small, not more than some hundreds of neurons.
Hardware so far was tailored according to the Von Neumann architecture, software followed suit which can be still seen in early programming languages like ASSEMBLER, FORTRAN and COBOL. Only in the recent decades new concepts like object oriented programming were widely used, which are very useful and quite abstract from the actual hardware implementation.
IBM’s TrueNorth brings now a new hardware concept along with a object oriented programming language describing so called corelets, by this describing possible operations of the chip. It is specialized for writing code for a hardware implemented neural network. It is described as:
"Leveraging this notion, we have developed an entirely new programming paradigm that can permit construction of complex cognitive algorithms and applications while being efficient for TrueNorth and effective for programmer productivity." from:
"Cognitive Computing Programming Paradigm: A Corelet Language for Composing Networks of Neurosynaptic Cores", Amir, A. et al., IBM Research - Almaden, San Jose, CA 95120
So we have now a chip that can compute predictive analytics in real time, a database that can deliver the data in real time and a programming language that can describe neural network architectures for the new chip.
I think bringing the speed and efficiency of silicium based neural networks together with the data volume and speed of SAP HANA could lead to exciting applications.
Some years ago graphic cards were widely introduced, mainly with the task of making visualization of 3D scenes possible. I think it is possible that we will have neural net cards in near future. Likely the will be based on descendants of today’s TrueNorth technology and making complex big data analysis available instantly at a fingertip.
