As a member of Yoshua Bengio’s lab in Montréal, Canada, Grégoire completed his PhD in Deep Learning, while collaborating closely with research teams at Google, Microsoft and Facebook, in the US.
Grégoire is now working on his second venture based in Paris and Montréal. His consulting firm, Incalia, builds custom Machine and Deep Learning solutions for companies where the client owns the IP. The team works on a wide variety of solutions, from analyzing medical image to integrating user feedback into search engines. He and his team mentor Data Science teams want to become proficient in Deep Learning.
When I asked him, Grégoire gladly accepted to answer some questions about his field of interest, which I find fascinating.
How do Artificial Intelligence (AI) and Machine & Deep Learning bring value to companies?
AI is relevant and very useful for organizations that would like to streamline a human decision process into an automated process at scale. Today, machines are reaching an accuracy equal or superior to human performance on certain tasks. AI is creating value in diverse markets by allowing new ways of collaborating between traditional wisdom and algorithms. AI is or will be present in most industries, with applications as varied as self-driving trucks to detecting early stage cancers.
What is the difference between Machine Learning and Deep Learning?
Research and industry have made huge progress over the last three decades in the field of Machine Learning, but Deep Learning is now adding on another significant layer of change. Deep Learning allows machines to bridge the perceptual gap. In the past, it was hard to extract the best characteristics from sensory input like images, sounds or even texts into machine-readable format. Now, the machine can automatically learn the features that are most suited for the challenge being tackled.
Another advantage of Deep Learning is versatility. The high-level architecture of Deep Learning is composed of plug and play blocks that can be easily combined. For instance, you have two algorithms: one can generate text from a large, unstructured set of text and the other one can easily detect an object in an image. You can plug those two models together and obtain an algorithm that will automatically generate textual captions for images, given the detected objects in the image.
With AI and DL being more and more introduced in our daily lives, should we be scared that robots will take over?
Machines are still dumb. Indeed, they can become extremely competitive when world-renowned scientists spend significant time working on a complex problem, as we recently realized it with the Go Game victory of AlphaGo versus a world-class human champion. This victory heavily relied on human input, as numerous sequences from expert human player games were fed into the machine during its training process. So, we are currently very far away from a computer that we feed with minimal data, and that becomes intelligent by itself.
In nature, animals and plants adapt and learn very quickly from their environment, showing a form of intelligence that we cannot find in machines.
Explore this fascinating topic in greater depth.
Here are some ideas: