Master's Thesis: Learning Progressively Smaller Models for Mobile Networks via Self-Distillation
Ericsson ABSammanfattning
This thesis opportunity at Ericsson Research in Stockholm focuses on developing a self-distillation framework for tabular foundation models (TFMs) in mobile networks. The role involves reviewing existing research, designing and implementing a self-distillation pipeline, and evaluating its performance on real-world datasets. Candidates will have the chance to explore innovative ideas in AI while contributing to the field of efficient AI for mobile networks.Det här erbjuder vi
Opportunity to develop innovative ideas in AI.Experience working with real-world datasets and mobile network conditions.
Ansök senast: Öppet tillsvidare
Publicerad: 2026-10-29
Beskrivning
Join our Team
About this opportunity:
Mobile networks generate large volumes of continuously evolving data for tasks such as traffic prediction, anomaly detection, energy optimization, and network automation. Tabular foundation models (TFMs) can use large-scale pretraining and in-context learning to adapt to new tasks and changing data without extensive retraining. However, their computational demands can make deployment difficult in latency-sensitive, resource-constrained network environments, where fast and interpretable decisions are important.
This thesis will explore a self-distillation framework in which a large TFM acts as a teacher, adapting to data drift and emerging tasks using contextual information. Its knowledge will guide progressively smaller models, with the aim of maintaining high accuracy while reducing latency, energy use, and model size without sacrificing interpretability.
What you will do:
You will also have the opportunity to develop your own ideas at the intersection of foundation models, continual learning, and efficient AI for future mobile networks.
The skills you bring:
Please include your CV, cover letter, and transcripts for your bachelor's degree and current master's studies.
About this opportunity:
Mobile networks generate large volumes of continuously evolving data for tasks such as traffic prediction, anomaly detection, energy optimization, and network automation. Tabular foundation models (TFMs) can use large-scale pretraining and in-context learning to adapt to new tasks and changing data without extensive retraining. However, their computational demands can make deployment difficult in latency-sensitive, resource-constrained network environments, where fast and interpretable decisions are important.
This thesis will explore a self-distillation framework in which a large TFM acts as a teacher, adapting to data drift and emerging tasks using contextual information. Its knowledge will guide progressively smaller models, with the aim of maintaining high accuracy while reducing latency, energy use, and model size without sacrificing interpretability.
What you will do:
- Review research on tabular foundation models, continual learning, and knowledge distillation for mobile networks.
- Design a self-distillation pipeline that adapts to incoming data and tasks under realistic mobile-network conditions.
- Implement the framework and evaluate it on real-world datasets with evolving data distributions and new task scenarios.
- Analyze trade-offs among prediction accuracy, adaptation speed, computational cost, model size, and interpretability.
- Present your findings in a written thesis and at Ericsson Research.
You will also have the opportunity to develop your own ideas at the intersection of foundation models, continual learning, and efficient AI for future mobile networks.
The skills you bring:
- You are pursuing a master's degree in Computer Science, Artificial Intelligence, Data Science, Electrical or Computer Engineering, or a related field.
- Strong programming skills in Python and PyTorch.
- Knowledge of AI; a background in machine learning and neural networks is preferred.
- Knowledge of model compression, distillation, or tabular foundation models is an advantage.
- Interest in AI research or building end-to-end AI prototypes.
- Familiarity with Git is a plus.
- Fluency in English.
Please include your CV, cover letter, and transcripts for your bachelor's degree and current master's studies.
- Extent: One student, 30 ECTS credits.
- Preferred start: Spring 2027.
- Location: Ericsson Research, Stockholm.
- Keywords: Knowledge distillation, tabular foundation models.
Ansök till tjänsten
Master's Thesis: Learning Progressively Smaller Models for Mobile Networks via Self-Distillation
OM FÖRETAGET
Ericsson AB
