Have you ever wondered to what extent your business applications might benefit from functions based upon artificial intelligence (AI) and deep learning? This question has long seemed irrelevant given the complexity of implementing these algorithms, but things have changed. Take the example of Microsoft’s Cognitive Services, which offer access to an effective image recognition engine via a hosted service.
AI as the cornerstone of tomorrow’s cloud services?
All of the main players in cloud technology are working flat out to develop deep learning tools, which are then broken down into functional components tasked with solving a specific problem, such as understanding a question posed using natural language, transforming voice into text, performing instant translation or detecting the emotions displayed on a face in a photo.
Google, Amazon, IBM, Microsoft or Apple are then implementing their algorithms by means of their own products: of course this brings to mind personal assistants such as Siri, Cortana or Alexa, but also such tools as Windows Hello or Face ID, thanks to which a user can unlock their session simply by showing their face to the camera.
The next step is of course making these new resources available and charging for their use, following the cloud strategy. This means that the client has access to a turnkey AI component without having to worry about creating the immense data records required to train the algorithm.
These offerings are generally based around two main elements. Firstly, a service layer distributed by way of programming interfaces (APIs) which allows clients to integrate preconceived models with their own applications. The second level consists of a machine learning platform designed to create and apply custom algorithms. In both scenarios, the infrastructure is made available upon request.
Cognitive Services: APIs and a toolkit for accessing AI
At Microsoft, ready-to-use components are being marketed under the title of Cognitive Services. They are based around broad groupings of intended uses, such as image recognition, linguistic processing, or the prediction or retrieval of information. In each case, access is provided to custom services assigned to a specific task: in the area of vision, for example, this could be emotional analysis, moderating images before publication, or detecting the objects contained in a photo. All of these services are accessible via common languages thanks to well-documented APIs. Nonetheless, it is difficult to be satisfied with a turnkey offering when solutions to precise business needs are required.
A neighbouring system to the above is the Microsoft Cognitive Toolkit (CNTK), hosted and maintained on Github since 2016. This collection of tools and libraries is designed to allow developers to create their own models dedicated to solving the problem they’re faced with. In this context, there is a free choice of infrastructure: Microsoft naturally suggests turning to Azure to test and train models, but developers can absolutely use their own computing resources.
Whether they’re turnkey services or CNTK, these tools remove the technical barriers which could hold back developments linked to AI and machine learning: consequently, why not consider possible scenarios for implementation? The possibilities are countless and affect all areas of activity, from fraud prevention to customer insight.
This is especially true in the industrial world, where this ability to ‘perceive’ the environment and then interpret the data collected so as to make the right decisions will eventually benefit all processes. An example? Consider the vision sensors which control production lines and imagine that instead of simply sending an alert in cases of non-compliance, they inform the machine tool which settings to change in order to resolve the problem…
Derniers articles parMichel Sebag (voir tous)
- The ‘Cognitive Services’ bringing AI to business applications - 4 October 2017
- Deep learning gives us more intelligent apps - 25 October 2016
- Truly, Madly, Deeply Learned: How Deep Learning is Advancing AI in Leaps and Bounds - 4 October 2016