Innovation

"Deep Learning may need new programming," says Facebook's chief of AI

Feb 26, 2019 | 77 views

The Python programming language may be in contact days. Well, at least that's what Facebook's director of artificial intelligence research Yann LeCun thinks. For him, deep learning may need a new programming language that is more flexible and easier to work with, even though many of the researchers and engineers just roll their nose at it.

"There are several projects on Google, Facebook and elsewhere to design a language so compiled that it can be efficient for the deep learning of the machines but it is not clear if the community will follow because people just want to use Python," said LeCun in an interview with VentureBeat.

According to GitHut's recent Octoverse report, Python is currently the most popular language used by developers working on machine learning projects. The programming language also forms the basis for Google's PyTorch and Google's TensorFlow structures.

Recently, LeCun presented a paper exploring the latest trends and spoke to companies making next-generation chips at the International Solid-State Circuits Conference of the IEEE (Institute of Electrical and Electronics Engineers) in Sao Francisco.

The first part of the article is devoted to the lessons LeCun took after his passing through Bell Labs, including the observation that the creations of AI researchers and computer scientists tend to be tied to hardware and software tools. With more than 50 years of existence, artificial intelligence has its current growth directly linked to the increase in the computational capacity provided by chips and other hardware.

For LeCun, it's a virtuous circle. Better hardware brings better algorithms that boost performance and make people ultimately build even better machines and compounds.

In the early 2000s, after leaving Bell Labs and joining New York University, LeCun worked with other industry experts such as Yoshua Bengio and Geoffrey Hinton conducting research to resurrect interest in neutral networks and increase the popularity of deep learning.

Facebook's IA boss highlighted a number of trends in AI that hardware manufacturers "should" consider in the coming years and made recommendations on "the type of architecture needed in the near future," recommending that the increasing size of deep learning systems be taken into account.

He also believes that future deep learning systems will be largely trained with self-described learning and that new high-performance hardware will be needed to support this learning.