Ai

Ce qui suit est une opinion personnelle, un ressenti de mon expérience, et peut ne pas refléter la réalité ou même le ressenti de l’ensemble de mes camardes, merci de ne pas leur en tenir rigueur :) Lorsque j’ai débuté mon parcours professionnel, voire depuis mes études, nous avions une image assez négative des chercheurs en informatique. Ils étaient certes très intelligents, et avaient des connaissances approfondies, mais inutiles pour le quotidien. Savoir comment fonctionne un compilateur pouvait être passionnant, et servir dans quelques cas d’optimisation. De là à dire que c’était ce qui allait nous servir au quotidien… Durant les 15 premières années de ce siècle, la tendance a perduré. Ce que j’ai pu observer autour de moi n’était pas très glorieux pour les chercheurs et universitaires. Nous les trouvions déconnectés de la réalité, perdus dans des théories ou sur des problématiques très éloignées des nôtres. Quelques frémissements se sont fait sentir dans certains domaines avec la montée en puissance des grands acteurs actuels, Google en tête. Les questions d’analyse sémantique et de volumétrie de données à traiter ont amené ces acteurs à travailler directement avec la recherche scientifique, car aucun produit sur étagère n’était prévu pour ce genre de cas. Vu de mon fauteuil, cela aura été le début discret du changement que nous pouvons observer aujourd’hui. Les chercheurs sont sollicités, approchés, séduits. Nous avons besoin de leur vision en pointe, voire en avance sur la pointe, pour résoudre des problématiques spécifiques. Ce qui a changé, selon moi, est l’état d’esprit, probablement poussé par les start-ups et la digitalisation massive. Nous sommes passés d’une approche “produit” (qu’est-ce que je peux faire avec ce que je connais) à une approche “solution métier” (que faut-il pour résoudre le problème posé par le business). Et cela change tout. Là où nous nous limitions à utiliser les capacités de quelques produits et à les mettre en service pour des fonctions prédéfinies, désormais nous sommes en mesure de creuser la problématique métier, qui n’a souvent rien à voir avec un problème IT. Cette problématique, nous la traduisons ensuite en critères techniques, et nous allons à la recherche du meilleur compromis pour résoudre ladite problématique. Et s’il le faut nous nous tournons vers les chercheurs. Du côté des laboratoires, encore une fois selon moi, ce qui a changé en France est que ces équipes doivent maintenant aller chercher la plus grande part de leur budget dans des financements extérieurs. Et l’issue positive est que nous nous sommes rapprochés. Comme dans une belle histoire Disney de Noël (c’est de saison !), chacun a fait un pas vers l’autre et ensemble nous sommes plus forts. ;) Le marché privé se rend compte que le mode de fonctionnement et de financement de la recherche publique est particulier. Le privé est capable d’entendre cela et de s’y adapter, car cela permet de créer des nouvelles solutions, avec l’appui des meilleurs cerveaux et technologies, même si elles n’existent pas encore. Et la recherche publique a admis qu’elle devait travailler avec des projets peut-être plus précis, en termes de planning et d’objectifs, et surtout de ROI. ...

About two years ago, Google announced the availability of TensorFlow processing units in its cloud. They are dedicated microcontrollers built for training and running Machine Learning models. TPU are available within Gcloud as an execution platform for ML (of course, optimized for TensorFlow). During the summer, they unveiled the edge equivalent of these TPU, which are named… Edge-TPU :) These are very specific ASIC designed to execute ML models on an edge device, i.e. a small device close to the sensors gathering the data. This allows for a fast decision, without the need to send a truckload of data back up to the cloud. But wait for it… Microsoft did just uncover a device called DataBox Edge. I know, the main purpose of this device is to provide a storage gateway to help you use Azure storage locally, and move the data between the device and Azure, hence the name. Bear with me, the path is a bit convoluted, and I would like you to enjoy every turn of it. Databox Edge is also equipped with what has been called IoT Edge. This nifty piece of technology will enable you to run Azure-based workloads on an edge device, such as Azure Functions, Azure ML, Azure Stream Analytics etc. IoT Edge has been out in the open for about a year now, to be deployed onto compatible devices. And, and that’s where we hit the Edge-TPU spot, also included in Databox Edge is a shiny new Microsoft hardware, called Brainwave. The name kind of gives away the purpose, especially after I guided you through the maze. Anyway, this chip is designed to run AI models on an edge device, and do it with impressive performance and efficiency. I know, at this point, you would point out at the fact that it might again be a case of “We did it first!” from Google. I’d like to focus a big difference between the two approaches. For once, I could not say which would win in the long term. In theory I prefer the approach from Microsoft, but that does not mean it will prevail (or that they would not change tactics and build something more like Edge-TPU). The difference is that Google built an ASIC, whereas Microsoft used Intel FPGA to deploy its Brainwave architecture. OK, this needs some explaining. First the names : ASIC means Application Specific Integrated Circuit. FPGA means Field Programmable Gate Array. You see where this is going? An ASIC is a very specific chip, designed to do only one thing, but optimized to its core. I should be able to execute one kind of job, but do it perfectly. One the other hand, an FPGA is reprogrammable after its deployment, to be able to adapt to future needs. Its performance is close to an ASIC, but not quite equal. To complete the panorama, going from specific to general use, we would then add GPU (Graphical Processing Units, as in your graphics cards) and then CPUs (ye good ol’ Pentium). Microsoft took the path of versatility, whereas Google focused on a particular use. As I mentioned, I’m not sure who has the best strategy, and whether there will even be a fight, but I am very curious to see both chips in the wild!

Wow, it has already been almost a month since I started! My new playground covers IoT and AI, and I am supposed to have a broad understanding of both. Regarding IoT, my recent background helped me grow a solid groundwork for that. I am fairly comfortable with the concepts, and with the involved technologies. Moreover, I have a colleague whose sole purpose is to understand and build IoT solutions, so my bases are well covered. When it comes to Artificial Intelligence, the coast is less clear. First, it is not a domain where I have any background, neither in the theory (math, bio science…) nor practical (any implementation of AI). Second, AI is the 2018 version of the Cloud in 2014 : everyone wants to do it, but not one has a clear definition of what we are talking about. Last but not least, the very term AI covers almost anything, from a chatbot to augmented reality to self-driving cars. My process has been a bit convoluted so far. First thing I have tried was to register for e-learning (MOOC or otherwise) sessions on the topic. I have tried several, from OpenEDX to Microsoft AI school, to Google and Tensorflow. The content ranged from very high level, which was mostly too high for me, to algebra (which was a bit too deep for me). Then I tried to read about the market. So I read a lot of whitepapers, from Microsoft, from DataIKU, from Forrester etc. This was rather useful, as it gave me basic understanding of where the situation was. I recommend Dataiku Machine Learning Demystified : https://pages.dataiku.com/machine-learning-basics-illustrated-guidebook But still, I felt I was stuck in the theory and could not find the practical applications. After some discussions with my usual suspect, Microsoft, I did have a look at their business uses cases and testimonies. I have to admit, some of them were pretty interesting… however there is absolutely no information about the architecture or implementation of the solution, which left me wanting. I finally found two Microsoft websites who did a good job of describing architectural templates, along with potential uses cases. https://azure.microsoft.com/en-us/solutions/architecture/?solution=big-data https://docs.microsoft.com/en-us/azure/architecture This is where I started digging, and it made my mind spin with all the possibilities. You will have to wait a bit for the outcomes, and follow what SCC will be doing on this market in the coming weeks ;) Last note, one of the smartest guy I have met at Microsoft, Frederic Wickert has started an AI business, and is writing, in French, to help debunk AI for us. I definitely recommend reading his posts!