Federated Learning: An Innovative Approach to Privacy and Collaboration in the Age of Artificial Intelligence

*By Patrícia Araújo de Oliveira and Kátia Adriana Cardoso

Federated Learning (FA) is a machine learning technique that allows collaboration in building Artificial Intelligence (AI) models without the need for direct data sharing. Unlike conventional methods that store data for training, AF is ideal for parameterized learning models, such as neural networks, allowing ethical and safe use of data. 

The transformative potential of AF was highlighted by Sheller et al. (2020), especially in healthcare, where it allows the safe and ethical use of patient data to train AI models, thus contributing to more accurate diagnoses and personalized treatments. The European Data Protection Supervisor (EDPS) reinforced the importance of THEFederated arrest  in your emerging technologies report. 

Benefits of using federated learning  from the perspective of Data Protection 

The integration of Artificial Intelligence (AI) into systems and processes represents a significant technological advance, with significant impacts on critical areas such as health, education and public safety and has become a key input in all sectors of the economy. However, the rise of AI has also raised concerns about the privacy and protection of personal data, triggering the search for solutions that balance innovation and data security. 

In this context, AF presents itself as a possible solution to achieve these objectives, as it allows different organizations to collaborate in the development of AI models without sharing raw data. This approach, where only the model parameters are shared, keeping the data on the original devices or servers, was demonstrated by McMahan et al. (2017) as effective in areas such as speech recognition and computer vision. Thus, it is possible to develop innovative solutions while respecting the privacy of user data. 

 Additionally, AF reduces the need for large data centers, increasing efficiency and decreasing vulnerability to cyberattacks. This is especially relevant under Brazil's General Data Protection Law (LGPD), which establishes clear guidelines for the processing of personal data and emphasizes the importance of technologies that ensure compliance with privacy principles. 

Challenges and Future Perspectives 

Despite its potential, AF faces challenges, such as efficiency in communication between devices, the heterogeneity of devices and data, in addition to important challenges related to privacy issues (Li et al., 2020). These are critical points for future research aimed at improving PA. 

The AF positions Brazil at the forefront of data protection, promoting the ethical use of AI for social benefit. It fosters collaboration and knowledge sharing, paving the way for advances in essential areas such as public health and scientific research. 

In summary, federated learning is a milestone at the intersection of AI and data protection, offering a promising path to secure and effective collaboration. As Brazil and the world advance in the adoption of AI, AF presents itself as a fundamental strategy for harmonizing technological progress and privacy, outlining a future where technology and ethics go hand in hand. 

*Patricia Araujo de Oliveira has a PhD in Computer Science from the University of Malaga (Spain), Researcher at the think tank from ABES, Adjunct Professor at the Federal University of Amapá and IT advisor at the National Cinema Agency.
*Kátia Adriana Cardoso de Oliveira is a PhD student in Law, with a focus on Artificial Intelligence and Data Protection at the Centro Universitário de Brasília (CEUB/DF), Researcher at the think tank from ABES in the area of privacy and data protection, OAB/DF lawyer specializing in digital law and data protection and federal public servant. The opinions expressed in this article do not necessarily reflect the positions of the Association.

Notice: The opinion presented in this article is the responsibility of its author and not of ABES - Brazilian Association of Software Companies

****

References 

• Bonawitz, K., et al. (2017). Practical Secure Aggregation for Privacy-Preserving Machine Learning. In Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security (pp. 1175-1191). 
• Li, T., Sahu, A. K., Talwalkar, A., & Smith, V. (2020). Federated learning: Challenges, methods, and future directions. IEEE signal processing magazine, 37(3), 50-60. 
• McMahan, H.B., et al. (2017). Communication-Efficient Learning of Deep Networks from Decentralized Data. In Artificial Intelligence and Statistics (pp. 1273-1282). PMLR. 
• Neto, HN, Mattos, DM, & Fernandes, NC (2020). User privacy in collaborative learning: Federated learning, from theory to practice. Brazilian Computing Society. 
• Sheller, B., Demiris, G., & Wiederhold, G. (2020). Federated Learning in Medicine: Facilitating Multi-Institutional Collaboration Without Sharing Patient Data. Scientific Bulletin, 5(1), 61-69.