Collaborative distributed machine learning: a path to strengthen data privacy

Contemporary machine learning (ML) approaches can substantially benefit society, for example, by predicting natural disasters and health crises. To develop ML models, companies commonly collect and conflate digital data about data subjects from different sources, often without (explicit) consent from data subjects. As data subjects have become increasingly aware of the negative effects of data collection, they have started to strive for more data privacy. In response, governments began to impose data protection regulations, making data collection for ML more difficult. Collaborative distributed machine learning (CDML) is an ML paradigm that aims to make digital data usable for ML while aligning with the striving for more data privacy. In CDML systems (e.g., federated learning systems), parties collaboratively develop ML models without exchanging plain training data. This chapter offers a foundation for understanding the CDML paradigm and how CDML systems can be combined with privacy-enhancing technologies.