SMALL NEURAL NETWORKS, SMARTER PHONES
by As it turns out, it’s not just the human brain that is worth modeling. Researchers at IBM have created a compact system modeled after the 48 million nerve cells in the brain of a small-sized rodent. By mimicking the network of neurons and synapses in a living brain, computer systems and algorithms can be fed data and learn to perform specific tasks.
While deep learning technology is already available in the form of Facebook’s face recognition feature and Skype’s real-time translation service, IBM’s chip could allow smaller devices like smartphones, hearing aids, and even watches to harness the technology and related algorithms more effectively. In the case of Google Now, most of the computing takes place on Google’s servers, which means they’re inaccessible if you happen to be working off the grid.
However, IBM chips, which have been dubbed TrueNorth, open up the possibility of moving at least some of the processing heft onto phones and other personal devices. A single TrueNorth chip contains 5.4 billion transistors, but draws only 70 milliwatts of power.
So while a company like Google could train neural networks in its own data centers to recognize say, pictures of cats, the smaller-scale neural network on your phone would build off this capability and simply help to execute it.