Federated Learning

By Abhivardhan

• Definition

Federated Learning is a decentralised way to train machine learning models through iterations, where instead of employing mass data collection on a central server to improve the global model, the model is trained on the client's private data, which remains locally stored, with only the resulting model updates thereafter sent to the central server. With the objective of reducing privacy risks, each participating client first downloads the global model from the central server, trains it on its locally held data, and then sends back a summarised version of the model's updated configuration. These model updates are thereafter aggregated and integrated into the global model, enabling it to improve without requiring access to the underlying raw data.¹

There are multiple types² of federated learning, including Horizontal Federated Learning, Vertical Federated Learning, Cross-Device Federated Learning, and Cross-Silo Federated Learning.

Commentary:

• Origin of the term 

Authored by Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson and Blaise Agüera y Arcas, the paper Communication-Efficient Learning of Deep Networks from Decentralized Data,³ published in 2017, introduced the term 'Federated Learning' to describe a decentralised machine learning technique whereby, instead of transferring sensitive and voluminous user data to a central server for model training, the model is trained locally on users' devices. Only the locally computed model updates are transmitted to a central server, where they are aggregated to improve a shared global model. Aligning with the principle of data minimisation,⁴ this technique allows the underlying raw data to remain on the users' devices. Federated Learning can also be referred to as Collaborative Machine Learning, Distributed Machine Learning, Privacy-Preserving Machine Learning, Decentralised Artificial Intelligence, Federated Machine Learning, etc⁵ and is especially relevant in Healthcare, Defence, Financial Services, Insurance, Marketing, Mobile Applications among other sectors.

• Operation in Practice

In practice, federated learning is being used by companies such as Owkin,⁶ which collaborates with hospitals to train AI models for cancer research without centralising patient data.⁷ Google’s Gboard⁸ also uses federated learning to improve next-word prediction and typing suggestions by learning from users’ devices while sending only model updates back to Google’s servers.⁹

• RELEVANCE VIS-A-VIS COMPETITION

Federated Learning becomes relevant to competition law because it allows competing firms to collaboratively improve AI models without directly sharing raw data. While it may create free-riding concerns, where one firm benefits from the global model without contributing proportionately valuable data, studies also indicate that successful FL formation can create an “all-win” situation by benefiting participating firms, consumers and overall welfare.¹⁰ However, such outcomes depend on market conditions, data asymmetries and the intensity of competition.

---

1. “What Is Federated Learning?” (Stanford HAI) <https://hai.stanford.edu/ai-definitions/what-is-federated-learning> accessed August 14, 2026↩︎

2. Gooday A, ‘Understanding the Types of Federated Learning’ (OpenMined, 9 December 2024) <https://openmined.org/blog/federated-learning-types/> accessed 14 August 2026↩︎

3. H Brendan McMahan and others, 'Communication-Efficient Learning of Deep Networks from Decentralized Data' (2017) 54 JMLR: W&CP↩︎

4. Ganesh P and others, “The Data Minimization Principle in Machine Learning” (arXiv.org, May 29, 2024) <https://arxiv.org/abs/2405.19471> accessed August 14, 2026↩︎

5. “FEDERATED LEARNING Synonyms: 11 Similar Phrases” (Power Thesaurus) <https://www.powerthesaurus.org/federated_learning/synonyms> accessed August 14, 2026.↩︎

6. 'What Are Real-World Examples of Federated Learning in Action?' (Milvus) https://milvus.io/ai-quick-reference/what-are-realworld-examples-of-federated-learning-in-action accessed 14 August 2026↩︎

7. Joseph Lehár, 'Federated Learning at Owkin' (Application of Artificial Intelligence & Machine Learning for Precision Medicine, FDA/CERSI, 17 February 2023)↩︎

8. Zhang Y and others, “Private Federated Learning in Gboard” (arXiv.org, June 26, 2023) <https://arxiv.org/abs/2306.14793> accessed August 14, 2026↩︎

9. Zheng Xu and others, 'Federated Learning of Gboard Language Models with Differential Privacy' [2023] arXiv:2305.18465v2 https://arxiv.org/abs/2305.18465 accessed 14 August 2026↩︎

10. Jiajun Meng and others, ‘Federated Learning and Free-riding in a Competitive Market’ (2024) arXiv:2410.12723 https://arxiv.org/abs/2410.12723 accessed 14 August 2026.↩︎