Publication

Network Traffic Reduction Through AI-Assisted Local Data Optimization of Synthetic Data

Matthias Rüb; Jens Grüber; Hans Dieter Schotten

In: Proceedings of the The 8th International Conference on Information and Communications Technology 2025. International Conference on Information and Communications Technology (ICOIACT-2025), December 4, ISBN 979-8-3315-5408-8, IEEE, 2025.

Abstract

In future wireless networks, AI instances are expected to take on autonomous tasks increasingly. In this context, the availability of high-quality synthetic data is becoming more important, as AI-agents in different domains, such as healthcare and natural language processing (NLP) will be expected to run safety and stability tests without the need for real data. Synthetic data is also useful for clinical practitioners to address imbalances in datasets or to augment existing data. However, this increased demand for datasets places an additional burden on the communication infrastructure, adding to the existing traffic. This not only impacts the performance of the network but also negatively affects sustainability. In this work addresses these issues by introducing a new paradigm, which shifts away from transmitting entire datasets and focuses on local data generation and optimization instead. While this approach has been desirable for a long time, recent developments in artificial intelligence (AI), particularly in NLP, have made it now feasible by empowering end-users without technical expertise to perform data optimization. As an exemplary use case, this work demonstrates a large language model (LLM) curated data optimization using synthetic smart insole gait data. Several lowweight LLMs are tasked to assist the curation. It is shown that local light-weight models, which can be deployed even with lowcost clinical IT infrastructure, can support non-experts with the local optimization process of otherwise insufficient synthetic data.

Network Traffic Reduction Through AI-Assisted Local Data Optimization of Synthetic Data

Abstract

Projects

More links