ActivityTitle: Deep Learning with Google
A Google engineer answers the question of how artificial intelligence, and in particular deep learning, will be used in business in the future and how it will change the world.
Deep learning is a type of machine learning that mimics the neural structure of the human brain. Its main feature is that, given a large amount of data, it is possible to build complex processing functions, such as image recognition and machine translation, using relatively simple techniques without humans having to give them a way to process the data.
A number of services are already using deep learning. For example, “Gmail” uses deep learning to identify unwanted emails. It has also made a significant contribution to improving the quality of translation in Google Translate. In addition, the company’s data centre uses deep learning to optimise the settings of its cooling equipment, which has resulted in a significant reduction in power consumption.
Google now and in the future
One of the areas where deep learning is being used is in image recognition. Google Photos, which stores images and videos, can already classify photos by the people in them, as well as by events such as Christmas and birthdays. Google is now taking this image recognition a step further and is working on recognising line drawings that people have drawn by hand, as well as image recognition for videos, such as lip reading.
What Google has to offer
Google provides a machine learning API (application programming interface), which is a pre-trained model for machine learning, and a machine learning library, TensorFlow, to enable both its own and external companies to use deep learning technology. TensorFlow is a machine learning library. The machine learning API provides functions such as image recognition, speech recognition and natural language processing. TensorFlow also allows users to build application-specific deep learning mechanisms on Google’s cloud platform by writing simple code in the Python language, without any specialist knowledge.
Image recognition use cases
One of the main areas of application is image recognition through deep learning. For example, a construction company is using deep learning to estimate data such as the hardness of rock for tunneling. By learning from a large number of photos of excavation sites, the company is able to determine the properties of the rock even when there are no experts on site.
A used car company also uses Google’s Tensorflow to recognise photos of used cars. The system automatically categorises the parts of the car that appear in the photo, such as the front, side and interior, to streamline the process of registering photos on used car sales websites. However, the automatic classification of these photos was made possible through a series of trial and error processes. At first, the system was unable to identify even the most basic of vehicles, such as whether they were facing left or right. Now, however, the system can even automatically identify the model of some vehicles.
In concrete terms, we have identified four types of data that can be used in deep learning: images, text, voice and sensors, and three objectives: reducing costs, adding value and creating new business opportunities, and improving creativity. The development of systems through machine learning, such as deep learning, is different from the traditional ones. The key to its use is to recognise the characteristics that allow results from one field to be “transferred” to other fields.