Master thesis: Key-Value Cache Management in Mobile Networks

Key-value (KV) caching speeds up transformer inference by retaining information from earlier computations. In mobile networks, changing wireless conditions, congestion and movement between base stations complicate the handling of this cached state. This thesis explores how to manage KV caches under these conditions while accounting for multiple concurrent users.

The assignment

• Survey research on KV caching for LLM inference, covering load balancing, quantization, compression, prefix caching and cache-aware request routing.
• Develop a testbed that models changing wireless links, interference, congestion, throughput, latency, user movement and base-station handovers.
• Enable experiments with multiple users sharing network and cache resources, alternative caching strategies and different network traces.
• Benchmark existing and proposed methods, including cache transfer versus reconstruction during handovers and mobility-aware prefix routing.
• Assess trade-offs in computational cost, latency and energy efficiency across network conditions.
• Document the work in a thesis and present the findings 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 related discipline.
• Strong Python and PyTorch programming skills.
• AI coursework or practical experience; knowledge of machine learning and neural networks is preferred.
• Interest in AI research or developing complete AI prototypes.
• Git knowledge is an advantage, as is experience with network simulators such as ns-3.
• Fluent in English.

What we offer

• Opportunity to suggest and develop your own research ideas within the scope of the project.
• Opportunity to present your results at Ericsson Research.

Practical details

Scope: 30 ECTS credits for one student. Preferred start: spring 2027. Location: Ericsson Research in Stockholm. Apply with a CV, cover letter and transcripts from both bachelor's and current master's studies. Application keyword: KV cache.