Learn Deep with Google’s AutoML

May 22, 2017, 1:55 p.m. By: Hasan Ali Gumani


In the Google’s I/O Conference 2017, Sundar Pichai (CEO, Google) talked about the new project called – AutoML. This new project focuses on deep learning - a technique that involves passing the data through layers of neural networks.

Creating neural networks is easier said than done. Hence, Google’s idea was to create an artificial intelligence that could shoulder this responsibility for them.

In AutoML, a controller neural network proposes a child model architecture which is trained and evaluated. The result is used to inform the child model architecture how to improve its proposal for the second round and so on. The process is repeated various times, thus, generating new architectures, doing testing on them and finally giving information to the controller.

Till now, the company has used AutoML to design a neural network for both images as well as speech recognition task. This new tech has the potential to impact many other AI-Driven applications that were announced in the conference, for instance, Google Lens. It will also lead to improvements in speech recognition, facial recognition, image recognition, etc.

Image Source: IoT For All