Deep Mind Advanced Project
20
Sep
Projects that attempted to simulate brain synapses, communication between neurons, were formerly
called neural or neural networks, and had a large development and applications.
Gradually these projects were moving to studies of the mind and the code was being directed to Machine Learning that now using neural networks happened to be called deep learning, an advanced project is Google Brain.
Basically it is a system for the creation and training of neural networks that detect and decipher patterns and correlations in applied systems, although analogous, only imitate the way humans learn and reason about certain patterns.
Deep Learning is a branch of Machine Learning that operates a set of algorithms used to model data in a deep graph (complex networks) with several layers of processing, and that, unlike the training of neural networks, operate with both linear and non-linear patterns .
One platform that works with this concept is Tensor Flow, originated from an earlier project called DistBelief, is now an open source system, released by the Apache 2.0 team in November 2015, Google Brain uses this platform.
In May 2016, Google announced to this system the TPU (Tensor Processing Unit), a programmable artificial intelligence program accelerator with high transfer rate ability for low precision arithmetic (8 bts), which runs models and does not more training as neural networks did, a Deep Compute Engine stage begins.
The second step of this process in Google Compute Engine, the second generation of TPUs achieves up to 180 teraflops (10 ^ 12 floating point operations), and mounted in clusters of 64 TPUs, work up to 11.5 petaflops.
