Europe & the Machine Learning Train

I had a look to the recent keynotes of Google I/O and since more than 2 years we only ear “Machine Learning”, “Deep Learning” all around. I am excited by what happens in this field, there are real improvements on important topics, even if there is so much hype.

It will be a big challenge for Europe to be a leader on Machine Learning applications from a business point of view. Of course, there are many top-notch scientists working hard and companies making thing happens. But that is not the point of this short note.

The most important part is the size of your business: datasets and data-centers.

Availability of huge quality datasets are critical, those are the seeds to feed your ML (Machine Learning) algorithms. If you do not have a business big enough, or a big enough user base, it will not work. (NB: this can change if there are break-through at this level, and probably there will be improvement over time)

For products: think about Alibaba and Amazon.

You can say that you just see correlation between existing products, and if there are new products with a lot of hype you can win. But it is difficult now to bypass those Logistics platforms or to go direct to the customer, the window of profit is narrow, the information spreads quickly, even if you have a kind of monopoly. Also, do not forget this is not only about products, there are having all different kind of services and much more information than just products, so they know more and more stuff over time about customers.

For users: think about Facebook, Google, Baidu, Microsoft and Apple.

This is the same story, people are continuously feeding private datasets by using many different services improving not only the accuracy/performances of the system, but also generating new knowledge from those streams of data. Also, it is stimulated by the creation of new services, new applications that further generate new streams of data and it increases opportunities for revenues.

Availability of data-centers is critical too, ML is really super computation thirsty. Europe is missing leaders big enough and having the money to have such huge data-centers. Performances matters a lot in this area, there are huge streams of continuous data coming and training/testing those ML systems are extremely computation/data intensive. This is why Google is developing its own ASICs to further push the performance, reduce the energy. Such improvements are key enablers, it can make things not possible before coming possible, and more important it helps to iterate further and faster.

There is a race between all the big names of the IT industry in the world: winners will take more customers and higher retention. ML can disrupt the existing business in one way or another, they need to spread out using it otherwise someone else will do.

You can see there are more and more initiatives to make ML tools available for developers. Google is the best in class with TensorFlow that spreads everywhere. It is an interesting way to secure the core business for data-centers: leadership with the best API that is already a standard, nice hardware infrastructure, biggest community in ML if not now very soon.

It does not mean there are no opportunities for Europe, there are many but it will very difficult to compete in the major league already tackled by the big boys of the IT industry. We are living in quiet exciting times.