The software that enables these features is called Whetstone and it has been configured to allow neural computer networks to process information some 100 times more efficiently compared with current imaging standards. This enables increased use of artificial intelligence in mobile devices plus autonomous vehicles. The development comes from Sandia National Laboratories, based in the U.S.
The intelligent system reduces the amount of circuitry needed to perform autonomous tasks. According to lead researcher, neuroscientist Brad Aimone: “Instead of sending out endless energy dribbles of information…artificial neurons trained by Whetstone release energy in spikes, much like human neurons do.”
The circuits are formed of artificial neurons, and these are essentially capacitors capable of absorbing and summing electrical charges. These charges are then releases as tiny bursts of electricity. These work with computer chips called “neuromorphic systems,” which act to group neural networks into larger clusters that mimic the human brain.
Neuromorphic engineering involves building electronic analog circuits to mimic neuro-biological architectures present in the nervous system. The key difference is that neuromorphic systems can operate a billion times faster than biologic neural systems.
These systems can sometimes be less efficient. The software Whetstone deploys machine learning to train and to improve artificial neurons through leveraging those that spike only when a sufficient amount of energy (in the form of data, for computing) has been collected.
In trials, this training process has been effective in improving standard neural networks and it will improve the required future-state technology needed for neuromorphic systems. Microprocessors configured more like human brains than traditional computer chips could soon make computers far more astute about what’s going on around them. Such visual awareness is of particular use to the development of autonomous vehicles. Autonomous vehicles with embedded artificial intelligence will be required to assess massive amounts of information and data points in real-time to effectively make decisions and this type of technology is regarded by many researchers as a critical development.
The software has been provided as an open-source code. The development has been described in the journal Nature Machine Intelligence, and the peer reviewed paper is headed “Training deep neural networks for binary communication with the Whetstone method.”
