Master thesis: Key-Value Cache Management in Mobile Networks

Join our Team

About this opportunity:

The key-value (KV) cache is a core optimization in transformer-based models. It accelerates token generation by reusing previously computed information instead of repeating the same computations.

Much of the existing research assumes fixed hardware and communication conditions. Mobile networks are dynamic: wireless links vary with fading and interference, congestion affects latency, and users move between base stations. During handovers, accumulated session state may need to be transferred or rebuilt.

This thesis will investigate KV cache management in mobile networks, considering user mobility, concurrent users, and wireless channel conditions.

What you will do:

The project will move from literature review to testbed development and systematic evaluation:

• Review the literature: Examine KV cache management for LLM inference, including load balancing, quantization and compression, prefix caching, and prefix-cache-aware request routing.

• Build a testbed: Simulate realistic mobile-network conditions, including time-varying wireless links, fading, interference, throughput and latency variation, congestion, user mobility, and handovers between base stations. Support multiple users sharing cache and network resources, different KV cache strategies, and different network traces.

• Evaluate strategies: Benchmark existing and proposed approaches, such as transferring versus rebuilding the cache during handover and mobility-aware prefix routing. Compare compute cost, latency, and energy efficiency across network conditions, and analyze the trade-offs.

• Report your findings: Write a thesis report and present your results at Ericsson Research.

You will also have the opportunity to propose and develop your own ideas within the project.

The skills you bring:

• You are pursuing a master’s degree in Computer Science, Artificial Intelligence, Data Science, Electrical Engineering, Computer Engineering, or a related field.

• Strong programming skills in Python and PyTorch.

• Coursework or experience in AI; a background in machine learning and neural networks is preferred.

• Interest in AI research or building end-to-end AI prototypes.

• Knowledge of Git is an advantage; familiarity with network simulators such as ns-3 is beneficial.

• Fluency in English.

Application and Practical Details

Please include your CV, cover letter, and transcripts for both your bachelor’s degree and current master’s studies.

• Extent: One student, 30 ECTS credits.

• Preferred start: Spring 2027.

• Location: Ericsson Research, Stockholm.

• Keyword: KV cache.