When capturing expert knowledge in an application, we speak of artificial intelligence (AI). Statibot is this kind of system and maps expert knowledge on inferential statistics to an application: a decision tree created with the help of TreeMaker simulates the statistical consultancy process of an expert.
Many modern AI applications, however, require a system to acquire expertise on their own, based on training data. This process is called machine learning (ML). My showcase ML application tries to detect locomotives on an image, to narrow down their positions within the image and then to derive the propulsion type and the year of construction, entirely from the pixel data.
The above image generated by the machine learning application shows a figure called a saliency map. It contains an Ae 6/6 of the Swiss Federal Railways on the lower left part of the image. The locomotive has been correctly located in the image and the electric propulsion type is correct. Based purely on the pixel information, the application identifies a year of construction in the 1960s, and in fact the Ae 6/6 was built between 1955 and 1966.
This showcase application can be translated into a variety of use cases in industrial analysis and optimization processes.