[LUM#22] Veni, vidi, industrie
The year is 2024 AD. Artificial intelligence has taken over every sector... Every sector? No! Industry still holds out against this revolution. A resistance that mathematician Bijan Mohammadi could well overcome with his magic formula.
He's neither a druid nor a warrior, but a researcher at the Alexander Grothendieck Institute in Montpellier. By swapping the cauldron for a computer and the potions for an algorithm, Bijan Mohammadi has developed an AI that is unique in the world, and convinced industrialists such as Renault of its relevance. " It uses physical signals such as noise and vibrations to detect anomalies and carry out preventive maintenance," he explains.
Hearing the tearing of a tiny rubber connector to keep a faulty car out of the way, or signalling the wear and tear of a milling head by tracking changes in vibration - these are just some of the applications this AI is capable of. Not enough to break a wild boar's back, some would say, except that ... "... This enables us to produce up to 300% more parts with the same milling head, which represents a saving of several hundred thousand euros per year and per production line", stresses the mathematician "accustomed to the industrial world and its logic".
AI and industry: the zizzy
Saving money, time and quality. That's all it took to convince this sector, because while we might imagine AI ruling industry like Caesar ruling Gaul, it's not the case. There are many reasons for this reluctance, starting with the lack of robustness of conventional AIs, whose fixed architectures are not designed to adapt to the variability of industrial sound environments. " As Bijan Mohammadi illustrates, "We can see this with telephone voice assistants, which bug if the user has a strong accent or if there's background noise.
Added to this is the problem of data. For reasons of sovereignty, manufacturers refuse to turn to the Gafams to share the data needed to train AI. Then there's the question of scaling up to industrial scale, and the lack of proof of return on investment, in short. ..". A lot of difficulties, few guarantees, an equation that doesn't appeal to the business," sums up the mathematician, who has therefore opted for another formula.
Caesar's laurels at Bondzai
In the summer of 2016 A.D., while Gaul was taking a tour of itself, Bijan Mohammadi was working on his 12 works and developing the algorithm that would open up the realm of the gods to him. This replaces the fixed architecture of conventional AIs, which comprises a fixed and usually gigantic number of layers of neurons, with a neural generator that stacks the layers in real time like a 3D printer. "It's a kind of miniature AI that creates its own architecture. The advantage is twofold: it needs very few examples and adapts completely to the environment in which it is placed. It's also very reliable, since it involves very few specific parameters."
And where there's little data, there's no cloud, so there's no privacy problem, and no colossal energy expenditure either. "It's a frugal AI that runs on classic processors that require very little energy and no investment in terms of IT infrastructure." Reliability, adaptability, profitability... the ingredients of the glory marketed since 2021 by the start-up Bondzai, whose laurels Bijan Mohammadi(x) has not finished picking.
Read also:
UM podcasts are now available on your favorite platforms (Spotify, Deezer, Apple podcasts, Amazon Music...).