DFF1 Project NETML: Networks for Reliable Machine Learning

As of now, machine learning (ML) algorithms are adopted in wide-range of applications, including time-critical analytics and decision-making applications, such as in industries, health sector (e.g., early diagnosis of terminal illness), finance, agriculture and so on. These ML applications rely and benefits from data whose availability is heavily dependent on the underlying communication network architecture. The project "NETML" contributes to the adoption of ML-driven applications through collective intelligence using massively distributed "high-quality" private data; however, without violating data privacy, and exploiting dense connectivity and computing infrastructure offered in 6G, to guide for accurate and reliable decision-making. The developed methodology offers valuable insight in understanding privacy-reliability tradeoffs in networked systems and builds protocols for communication-efficient ML training that uses only relevant and high-quality data; hence, equally promoting energy frugal connectivity by limiting unnecessary computations and exchange of data.

Machine learning (ML) is currently utilized in many areas that require timely analysis and decision-making, such as industry, finance, agriculture, and healthcare (for early disease detection). For ML models to operate optimally, they need access to relevant data, which depends on the structure of the communication network.

The "NETML" project will expand the use of machine learning by working with large amounts of "high-quality" private data from various sources, without compromising personal privacy protections. Additionally, it will leverage the fast and efficient connectivity of 6G networks to make precise and reliable decisions.

A crucial aspect of the project is finding ways to balance privacy protection with the need for reliable data. This involves training ML models with the best and most relevant information. Such an approach can also help save energy by minimizing unnecessary computations and data exchanges.

Protecting data and managing the network during data collection can be challenging. Therefore, this project will develop new communication methods that ensure only high-quality and relevant data is used for machine learning.

The goal is for the project's results to facilitate the transfer of critical data and provide quick and reliable responses in urgent situations.