Thursday 13 March 2025
Researchers have made a significant breakthrough in the field of vertical federated learning, a technique that allows multiple organizations to collaborate on machine learning projects without sharing their sensitive data. By developing an efficient method for unlearning, or removing, specific information from trained models, scientists have overcome one of the biggest challenges facing this technology.
Federated learning involves training artificial intelligence models using data from multiple sources, such as hospitals, banks, and social media platforms. This approach has numerous benefits, including improved accuracy and reduced risk of data breaches. However, it also raises concerns about data privacy and security, particularly when dealing with sensitive information.
To address these issues, researchers have been working on developing methods for unlearning, or removing, specific information from trained models. This involves identifying the parts of the model that are responsible for learning certain patterns or relationships in the data, and then modifying or removing those components to prevent them from being used to infer sensitive information.
The new method, known as vertical federated unlearning, is designed specifically for use in vertical federated learning scenarios, where different organizations contribute their own data to a shared model. The approach involves using knowledge distillation, a technique that allows one model to learn from another by mimicking its behavior, to remove specific information from the trained model.
The researchers tested their method on several real-world datasets, including adult income prediction, credit scoring, and disease diagnosis. They found that their approach was able to effectively unlearn specific information from the models, without sacrificing accuracy or performance.
One of the key benefits of vertical federated unlearning is its ability to protect sensitive data while still allowing organizations to collaborate on machine learning projects. This could have significant implications for industries such as healthcare and finance, where data privacy is a major concern.
The researchers are now working to further refine their method, with plans to explore its use in more complex scenarios, such as multi-task learning and transfer learning. They also hope to develop methods for unlearning specific information from models that have already been deployed in production environments.
Overall, the development of vertical federated unlearning represents a significant step forward in the field of machine learning, and has the potential to revolutionize the way organizations collaborate on AI projects while protecting sensitive data.
Cite this article: “Breakthrough in Vertical Federated Learning: Protecting Sensitive Data through Unlearning”, The Science Archive, 2025.
Machine Learning, Federated Learning, Vertical Federated Learning, Unlearning, Data Privacy, Data Security, Knowledge Distillation, Artificial Intelligence, Sensitive Information, Collaboration.
