About this role
Join our Team
About this opportunity:
Modern mobile networks are increasingly operated by autonomous control loops. Before an automated agent is allowed to adjust a live network, its proposed actions must be evaluated, and world models are one way to achieve this.
For critical telecommunications infrastructure, predictive accuracy alone is not enough. Operators and regulators increasingly demand models whose behaviour can be inspected, audited, and justified. At Ericsson Research, we are exploring intrinsically interpretable machine learning for network dynamics. Kolmogorov–Arnold Networks (KANs) have demonstrated the potential to balance performance, interpretability, and model size.
Black-box sequence models, such as RNNs and transformers, often outperform feedforward predictors on network data because they exploit information carried in the trajectory history. This suggests that the observed indicators have strong temporal correlations.
The goal of this thesis is to investigate whether this advantage can be recovered inside an interpretable model: a nonlinear state-space model with a low-dimensional learned latent state, parameterised entirely by KANs. The full learned realisation—including the state transition and observation maps—should be possible to prune and express in closed form.
The thesis will be conducted by one student, corresponds to 30 hp, and is expected to start in early January 2027. The location is Stockholm, Sweden. Supervisors are Agustín Valencia and Maxime Bouton.
What you will do:
- Formulate and implement a KAN-parameterised state-space model with a learned hidden state, trained on multi-step rollouts.
- Validate the approach under controlled partial observability in standard Gymnasium environments, where hidden variables are known and can be masked. This will allow recovery of latent structure to be measured against ground truth.
- Apply the validated method to a real Ericsson Radio Access Network dataset.
- Benchmark the approach against recurrent models and state-of-the-art interpretable baselines.
- Analyse the learned model symbolically through closed-form extraction and from a control-theoretic perspective.
- Present the results to the Ericsson research team and deliver a repository with reproducible experiments.
The skills you bring:
- You are a final-year Master's student in Mathematical Statistics, Applied Mathematics, Engineering Physics, Electrical Engineering, Computer Science, or a related programme.
- You have a solid foundation in machine learning and probability or statistics.
- You have strong Python skills and experience with a deep-learning framework, preferably PyTorch.
- You have coursework or project experience in at least one of the following areas: dynamical systems, control theory, system identification, or time-series analysis.
- You can work independently, structure an open research question, and communicate your findings clearly.
Why join Ericsson? At Ericsson, you´ll have an outstanding opportunity. The chance to use your skills and imagination to push the boundaries of what´s possible. To build solutions never seen before to some of the world’s toughest problems. You´ll be challenged, but you won’t be alone. You´ll be joining a team of diverse innovators, all driven to go beyond the status quo to craft what comes next.
What happens once you apply? Click Here to find all you need to know about what our typical hiring process looks like.Encouraging a diverse and inclusive organization is core to our values at Ericsson, that's why we champion it in everything we do. We truly believe that by collaborating with people with different experiences we drive innovation, which is essential for our future growth. We encourage people from all backgrounds to apply and realize their full potential as part of our Ericsson team. Ericsson is proud to be an Equal Opportunity Employer. learn more.
Primary country and city: Sweden (SE) || Stockholm
Req ID: 791197