By akademiotoelektronik, 04/10/2022

How to train in artificial intelligence

The last number

In 2017, the Association for Executive Employment identified 2,400 job offers in artificial intelligence (AI), twice as many as in 2016. And not all recruitment goes through Apec. Faced with the difficulty of finding these rare pearls, companies are looking to convert their internal resources. But the job is demanding and the needs are difficult to identify! Overview of training courses, according to profiles.

1. Acculturate executives

According to a study by the Boston Consulting Group, a third of companies that have launched AI projects have failed. “There are a lot of fantasies about what artificial intelligence can do. Before launching a project, senior executives must understand what they can get interesting out of it,” explains Olivier Ezratty. An engineer turned consultant, he provides two-day training sessions on AI for Capgemini Institute, at a cost of 2,400 euros. "The Uses of Artificial Intelligence", his very complete ebook, free to download, is a good start to self-study. More technical, Polytechnique offers top managers "learning expeditions" of one to three days, to meet start-ups and industrial companies using AI.

Sorbonne University (ex-UPMC) gives access to a university degree (DU), much more in-depth, but still generalist, in 98 hours spread out from October to April, at the rate of two days per month (8,000 euros) . The first promotion welcomes 17 trainees this year.

Half of them come from industry and research, half from services (IT, banking, insurance), they all have at least a bac +5. Like engineers, human resources or customer relations managers start with two days of math and computer science that will not make them data scientists, but will allow them to know the foundations of AI. Courses will follow on the management of massive data, machine learning, AI applications by sector... "These are business experts, confronted with the collection, analysis and processing of data, who come to seek a general vision what AI can bring them", explains Chantal Viger, from Sorbonne University.

2. Bringing AI bricks to technical professions

"It is difficult for an employee to train in data science. It is a very technical job, which requires advanced skills in mathematics and statistics. Even retraining data analysts is complicated. We prefer to recruit young graduates “, testifies David Remaud, the marketing director of Dhatim.

On the other hand, developers, data analysts and product managers of the French software publisher have been trained in the use of AI in their profession. Eight out of thirty employees took the Stanford Mooc (free) on machine learning, fifty hours leading to a certificate. Cornell University, in the United States, also provides online access to a MOOC on this subject.

A group of engineering schools (Ensta, schools of the Institut Mines-Télécom) will launch in April a certifying training in artificial intelligence of twenty-five days, over nine months (10,600 euros). Intended for engineers and computer scientists, it addresses deep learning, neural networks, learning for robotics. The Institut Mines-Télécom also offers a certificate of the same type dedicated to data science and around ten two-day specialization training courses.

3. A certifying training at Polytechnique

Since 2015, Polytechnique has had a 20-day training course in data science for technical profiles (statisticians, programmers) (12,000 euros). "The objective is to give them the keys to understanding to interact with specialists", explains Larbi Touahir, the director of the "digital" theme in continuing education. Next fall, two programs will allow students to discover Natural language processing and web marketing in twelve days.

"The needs of companies guide our offer, and text analysis is one of them", specifies Larbi Touahir. Which provides on-demand training in visual computing and is considering training at the crossroads between blockchain and AI. Developers are introduced to new AI tools such as Google's TensorFlow program library. Or follow the free training courses set up by the companies that deploy these new tools and wish to make them essential. Microsoft, for example, trains young people with the Simplon school and the Nvidia institute dedicated to deep learning announced in 2017 that it wanted to train 100,000 developers, particularly for the medical sector.

4. Train the coaches

The simpler the AI ​​interfaces become, the less training the end user needs. But his tasks are changing. A new profession appears, that of AI coach. "Someone has to teach the machine to read the invoices and place them in the right account", underlines David Remaud. A job done at the software publisher when he creates it, to “feed” the AI ​​with data, then by the customer himself. A less technical task, but meticulous, which still requires training, often provided by the publisher. "The machine is 95% autonomous, but in 5% of cases, the trained accountant must intervene", continues David Remaud. In the same idea, Systran, a machine translation company, simplifies its models so that its customers can "train" the machine themselves by entering their own data. To create a translation engine on a specific lexical field, for example.