Master's Thesis: Learning Progressively Smaller Models for Mobile Networks via Self-Distillation
publicerat av Ericsson
- Företag
- Ericsson
- Ort
- Stockholm
- Sista ansökningsdag
- Ej angiven
- Nivå
- Master/civilingenjör (30 hp)
- Antal studenter
- 1 student
- Period
- 30 ECTS credits; preferred start in spring 2027.
- Passar program
- Computer Science, Artificial Intelligence, Data Science, Electrical Engineering, Computer Engineering
Beskrivning
Deploying tabular foundation models in mobile networks is challenging because their computational requirements can conflict with tight latency and resource constraints. This thesis investigates self-distillation, using a large model that adapts to changing data and tasks to train progressively smaller models. The goal is to preserve accuracy and interpretability while lowering inference latency, energy consumption, and model size.
The assignment
- Survey literature on tabular foundation models, continual learning, and knowledge distillation in mobile-network applications.
- Develop a self-distillation pipeline that responds to new data and tasks under realistic network conditions.
- Implement and test the approach using real-world datasets featuring distribution shifts and emerging tasks.
- Assess the trade-offs between predictive performance, adaptation speed, computational requirements, model size, and interpretability.
- Document the results in a thesis and present them at Ericsson Research.
Who we are looking for
- Currently studying for a master's degree in Computer Science, Artificial Intelligence, Data Science, Electrical Engineering, Computer Engineering, or a similar discipline.
- Strong Python and PyTorch programming abilities.
- Knowledge of AI, preferably including machine learning and neural networks.
- Experience with model compression, knowledge distillation, or tabular foundation models is beneficial.
- Interest in AI research or developing complete AI prototypes.
- Git experience is an advantage.
- Fluent English.
What we offer
- Scope to explore your own ideas combining foundation models, continual learning, and resource-efficient AI for future mobile networks.
Practical details
One-student thesis worth 30 ECTS credits. Preferred start in spring 2027 at Ericsson Research in Stockholm. Apply with a CV, cover letter, and academic transcripts from your bachelor's degree and ongoing master's studies.
Förkunskapskrav
Applicants should be pursuing a relevant master's degree, have strong Python and PyTorch skills, possess AI knowledge, and be fluent in English. A background in machine learning and neural networks is preferred. Experience with model compression, distillation, tabular foundation models, or Git is advantageous, alongside an interest in AI research or end-to-end prototyping.
Texten är en sammanfattning av Ericssons annons. Läs hela annonsen och ansök hos Ericsson