Machine Learning with Desjardins Lab and Google’s Jean-Philippe Gauthier
On November 23rd, we attended a talk by Google’s Jean-Philippe Gauthier at Desjardins Lab on the theme of Machine Learning and its potential. A number of topics were discussed, including Artificial Intelligence, Neural Networks, the importance of data in today’s world and so much more.
Early on, Mr. Gauthier mentioned that data is becoming “the new oil.” In the world of machine learning, everything starts with data, and Mr. Gauthier defines Machine Learning as a type of artificial intelligence that allows computers to learn using data. In the last two years, Machine Learning has progressed by leaps and bounds thanks to Deep Learning, which simulates the way the human brain learns and processes information.
According to Mr. Gauthier, data sets are about to become so massive that they will most likely surpass our capacity to make sense of them. To resolve this problem, Google has already started incorporating Machine Learning into its products. For example, Google Photos on mobile can now analyze picture taken with your phone and identify the elements they contain, thus eliminating the need to create tags and classify each photo manually. If you want to look at photos of your dog, all you need to do is search “Dog” and Google Photos will present to you all your photos that contain a dog. Moreover, Gmail now includes a “Smart Reply” feature that uses Machine Learning to automatically generate an email draft as a response to incoming emails, which you can then send manually.
And this is just the tip of the iceberg. Google is also working on a personal assistant that will be able to remind you of upcoming appointments, help you coordinate tasks you need to accomplish and retrieve information from the web for you quickly and precisely. Since the Google Assistant can learn, the more you use it, the better it understands your needs.
Data is king
To fully understand Machine Learning, we need to take a step back and re-evaluate how computer programs are designed. Without Machine Learning, to create an application, you need to program all functionalities and give precise instructions to the application as to how to interpret various data sets. With Machine Learning, the machine receives data and determines by itself how to classify or use this information, becoming more efficient over time. As a result, the machine can surpass us and imagine entirely new possibilities.
Generally speaking, three types of algorithms can allow machines to learn:
1) Supervised learning, in which the machine learns using data that has been properly labeled and identified (Example: This is a photo of a dog, this is a photo of a cat).
2) Unsupervised learning, in which the machine learns using unidentified, unlabeled data.
3) Learning by reinforcement, in which the machine learns by trial and error in a specific environment, exploring new possibilities while also using previously acquired knowledge.
For example, an artificial intelligence used learning by reinforcement to study Go, an incredibly complex board game that’s popular in Asia. Eventually, the AI was able not only to defeat some of the best Go players in the world, but also to imagine new strategies and ways of playing that surprised and delighted the human players.
A growing market
The sheer power of the CPUs and GPUs inside modern computers is another factor that allowed some of the new developments in the world of Artificial Intelligence and Machine Learning. Today, Google is even experimenting with a new type of processor, called Tensor Processing Unit (TPU), which was designed specifically for Machine Learning.
The number of applications on the market that use Machine Learning in some form is growing quickly. In the coming years, Machine Learning could, for example, allow breast cancers to be diagnosed faster and more efficiently, analyze texts to extract the important information, to determine the tone of a text or voice message (positive, negative, etc) and much more. In the long run, Machine Learning will optimize the world around us by analyzing large data sets and identifying the best solutions. The real world is complex and messy, but Machine Learning will help us improve it.