In July, a bunch of synthetic intelligence researchers showcased a self-driving bicycle that might navigate round obstacles, comply with an individual, and reply to voice instructions. Whereas the self-driving bike itself was of little use, the AI know-how behind it was outstanding. Powering the bicycle was a neuromorphic chip, a particular form of AI laptop.

Neuromorphic computing shouldn’t be new. In actual fact, it was first proposed within the 1980s. However latest developments within the synthetic intelligence trade have renewed curiosity in neuromorphic computer systems.

The rising recognition of deep learning and neural networks has spurred a race to develop AI {hardware} specialised for neural community computations. Among the many handful of traits which have emerged up to now few years is neuromorphic computing, which has proven promise due to its similarities to organic and synthetic neural networks.

Learn: [How machines see: everything you need to know about computer vision]

How deep neural networks work

On the coronary heart of latest advances in synthetic intelligence are artificial neural networks (ANN), AI software program that roughly follows the construction of the human mind. Neural networks are composed of synthetic neurons, tiny computation items that carry out easy mathematical capabilities.

Synthetic neurons aren’t of a lot use alone. However if you stack them up in layers, they’ll carry out outstanding duties, reminiscent of detecting objects in pictures and reworking voice audio to textual content. Deep neural networks can include a whole lot of hundreds of thousands of neurons, unfold throughout dozens of layers.