Abstract :
[en] Today, there is a high demand on intelligent systems (e.g chatbots, ob-
ject decetors, translators, etc). Engineers have to develop these systems
in a lots of di erent domains (e.g. medicine, nance, car industry). More-
over, these intelligent systems are trained on data collected from these do-
mains using an iterative training process. Et each training iteration, the
parameters of such system are updated intuitivly based on the engineer's
experience. However, gathering and labelling these data is very costly and
time consuming. Moreover, the systems are often complex. It is recom-
mended to have a strong mathematical background. Thus, engineers often
design these systems based on their own experience and collected informa-
tion about the system. We present the road towards a novel methodology,
called SEMKIS, for the design ang generation of intelligent systems and
synthetic learning data. We use the model-driven engineering approach in
our methodology to specify and design our systems. We generate speci -
cations, designs and implementation of our intelligent systems. We used
the mathematical set theory to de ne the concepts for the speci cation of
intelligent systems and data synthetis within a formal conceptual frame-
work. The concepts have been used in a small executable illustration that
focuses on the recognition of handwritten digits on a picture. The results
show that our concepts are usable and that we reduce the complexitiy of
specifying and designing intelligent systems.