AI in practice: Master of Science in Engineering students train the models of the future

Combining studies with real research projects is a dream for many students. For MHD Anas Alebrahim, it became a reality when he was hired to work with AI model training in two different projects.
"The first project is about building an AI model that can identify different types of road cracks. The other focuses on recognizing the end of fiber cables," he says.

What exactly is an AI model?

According to Anas, an AI model is a mathematical description of data, often in the form of a machine learning or statistical model. Once the model has been trained on data, it can be used to make predictions or classifications of new data.
But training an AI model requires more than just technical know-how.
"You have to understand the area in which the model is to be used, be able to program, have good knowledge of mathematics and statistics, and also be able to handle data efficiently," he explains.

How is an AI model trained?

Training an AI model is a multi-step process. It starts with formulating the problem by defining clear goals and how the result will be measured. The data is then collected and labeled, which means that the data is provided with information that the model can learn from. Before the model can be trained, the data needs to be pre-processed – it can be about removing errors, normalizing values or handling missing data. Once the data is prepared, it is divided into two parts: a training set used to teach the model, and a test set used to evaluate how well the model works.
"Time management is one of the most important factors in successfully creating an AI model," says Anas. There is a lot of data to be labeled, but with structure and focus, it goes well.
The next step is to select the type of model to use, and then the model is trained by adjusting its parameters. Finally, the model is evaluated and fine-tuned to improve its accuracy and performance.
Ana's role in the projects is to work with labeling and data annotation – a crucial step for the AI model to learn to recognize and classify different types of data.
"The goal is to manually create a classification template that the model can be trained on. The better the marking, the more accurate the model will be," he says.

Challenges and insights

The biggest challenge? The amount of data.
"There is a lot to be noticed, but with good time management, it is manageable," says Anas.
A normal working day involves concentrated work in front of the screen, with regular breaks to avoid eye and muscle fatigue.
"I've learned how important time management is, and I've gained a deeper understanding of how AI models work – especially how small deviations in data can affect the outcome.
Once the marking is complete, the supervisor will review the work before the programming phase begins.
"It feels exciting to be part of something that can actually be used in practice," Anas concludes.