CONTEXT
Airbus wishes to increase reliability of the software mass data classfication decisions. The important volume of data makes manual classification impossible. Decision-making rules as defined by engineers resulting in decisions less reliable than a Human expert.
SOLUTION
• Optimise the use of several descriptions for data objects.
• Speed up tremendously learning through HPC techniques.
• Enhance caution: better generalizing, quicker classification
• Improve the administration of local descriptions for objects.
RESULT
• Automation of repetitive tasks which are hardly modelled by an expert. Thus, the expert doesn’t need to model his knowledge, the software learns by observing the expert.
• Target detection and recognition on aerial imagery.
• Automatic vehicle recognition on pictures.
• Classification of ~10M aircraft parts to speed up the simulation: 97% accurate, ~x1000 speed increase compared to the expert.