They say that one of the reasons why in our smartphones with tablets still appeared real, but artificial neural network is the lack of power supply the required power.
The thing is that all these so-called brain-like, i.e. “mothodologie” artificial intelligence systems, at least in their modern form, dependent on large-scale multicore processors, which consume incredible (in comparison with the human brain) the amount of energy and therefore are still considered an impractical luxury in a mobile device of a custom class. However, the idea of a dead-end no one calls, and scientists are tirelessly looking for its implementation.
For instance, MIT researchers recently officially announced the processor Eyeriss – a new solution enabling the use of neural networks in low-power devices. Eyeriss also multi-core, it 168 cores, however the cent consumes 10 times less energy than any of the GPUs, available on current smartphones.
Why such impressive energy efficiency? According to the developers Eyeriss, a minimized volume of pumped data. Each of the cores (which works as a neuron) has its own memory, in addition all the cores can communicate with each other bypassing system memory. Thus, the entire computing work for recognition of images and speech, the system can be run locally, without involving network resources. In other words, the data is not sent anywhere, which guarantees high functionality devices in the absence of a connection, and increases the level of information security.
At MIT do not mention, when Eyeriss will turn into something more tangible, preferably to such an extent that it could be seen on the shelves of regular stores. However, new development has already been called a significant step in the development of artificial intelligence in General and machine learning technologies in particular. From the point of view of potential consumer, this means that, for example, a smartphone (tablet, computer, car, etc.) with the processor Eyeriss in the future, it would be better to adapt to new situations while solving a variety of problems, active use of the collective experience of their electronic counterparts and to continuously improve.
