“At present, AI (Artificial Intelligence) does not totally replace human beings (or only in rare cases). It helps increase them, to assist them in their daily lives (no, this is nothing new, a person who wears glasses is also “increased”). AIs are specific and they are designed to solve specific problems. We are still very far from being able to create a generalist AI (i.e. that would be able to perform tasks without any link and multiply applications). However, these AIs already find multiple applications that improve the performance of the companies that integrate them, whether in their production cycle, quality control, marketing, sales, or HR departments among others.”
— Excerpt from the article : Our definition of AI, Neovision’s expert viewpoint
ACQUISITION & STRUCTURE
Data is the raw material needed to create AI. It must be representative of the real data that the algorithm will have to analyze. Without data, there can’t be any AI. The data must then be cleaned and annotated before being exploited by an algorithm.
TO CREATE A MODEL
Once the database has been prepared, it is necessary to find the appropriate algorithm. Whether it is Open Source or designed by us, an algorithm will be adapted to certain types of data and use cases. Once selected, the algorithm will be trained on the prepared data.
TO OPTIMIZE PERFORMANCE
Following the training of the algorithm, we obtain a model, a mathematical function capable of applying a treatment on new data. This model is then tested and validated on the dataset. If the results are unsatisfactory, we will work again on the datasets and the models will need to be re-trained.
OF THE SOFTWARE SOLUTION
When the model is satisfactory, it is time to put it into production. First of all, a beta version of the solution must be quickly set up, so that the technology can be tested in a real environment. Subsequently, software engineering work is required to interface the software brick with the customer’s existing IT system.