Linköping University - LU
speciality: Artificial Intelligence

I enrolled in LiU’s MSc in Statistics and Machine Learning partly because the programme is presented as suitable for students aiming to work as machine learning engineers. Unfortunately, my experience has been very disappointing.
My main concern is that the curriculum feels much more like a traditional statistics degree with some machine learning added to it than a modern ML engineering programme. A lot of time is spent on mature statistical methods such as Bayesian methods, MCMC, graphical models, state-space models, Kalman filters and particle filters. These topics are useful, but the balance feels outdated.
Modern deep learning receives much less attention. Transformers were covered for roughly two weeks with one lab assignment, while LLMs were not covered as a separate topic at all. There are also no advanced electives in transformers, LLMs, modern generative AI, computer vision, reinforcement learning or ML systems. For a programme advertised as relevant to future ML engineers, I find this a significant weakness.
The programme also relies very heavily on R. R is still useful in statistics, but it is not the main language used in modern ML engineering. Students are often required to learn specific R libraries in depth, and this sometimes carries over into exams. In practice, this can mean spending exam time debugging unfamiliar R behaviour rather than demonstrating understanding of the underlying ML or statistical concepts.
Another issue is the way many statistical methods are taught. In my experience, there is often a lot of focus on mathematical derivations, but much less explanation of why a method is useful, when it should be chosen, what alternatives exist, and how different approaches compare in practice. As a result, the material can feel unnecessarily abstract and disconnected from real applications.
I am also concerned about examination outcomes in some courses. In 732A99 Machine Learning, some 2026 exam sittings appear to have failure rates approaching 80–90%, with very few A or B grades. What I found especially worrying is that some teachers I spoke to said they were not aware of how high the failure rates were in the courses they taught, and I am not aware of any systematic effort to investigate the causes.
Students also have very little flexibility to compensate for gaps in the curriculum. I tried to take a relevant ML/computer-vision course outside my programme. Both my programme coordinator and the course coordinator approved it academically, but the university still refused to allow it because of faculty/programme administrative rules. Despite involving many university employees and repeatedly asking for an exception process, no practical solution was found.
This experience also reflects my broader impression of LiU administration. Response times can be extremely slow, escalation routes are unclear, and staff often seem more focused on explaining why something cannot be done than on finding a solution. In one case, the issue eventually reached the Rector’s level, where the expected waiting time for a decision was six weeks—long after the relevant enrolment deadline had passed.
Overall, I would not recommend this programme to someone specifically looking for a modern, engineering-oriented ML master’s. It may suit students who want a statistics-heavy education with strong emphasis on probabilistic and traditional statistical methods, but that is very different from what I expected based on how the programme was presented.