Federated Learning: Predictive Model Without Data Sharing

What is Federated Learning?  Machine Learning techniques are advancing technology rapidly, like Automated Machine Learning. They are undoubtedly creating new opportunities and advantages for organizations from different...

Federated Learning: Predictive Model Without Data Sharing

What is Federated Learning? 

Machine Learning techniques are advancing technology rapidly, like Automated Machine Learning. They are undoubtedly creating new opportunities and advantages for organizations from different sectors. However the adoption of these newer technologies implies giving third parties access to their business ecosystems and databases and is always accompanied by serious privacy concerns. Big companies wonder how their data will be processed. This is an obstacle to the decision of outsourcing for machine learning tasks. That’s why Federated Learning can often be a possible solution!

Traditional Machine Learning approaches require a centralized data center or one server where all the local datasets are uploaded to be trained by a machine learning model. Federated Learning enables multiple actors to collaboratively build a robust machine learning model. All the while keeping all the training data on their own servers. In a nutshell, Federated Learning allows one to train a model without sharing data with a central computational node. Companies are then able to create a shared model, overcoming all the critical issues about data privacy, data security, and data access rights.

How does it work? 

Let’s have a look step-by-step at how it works.

  1. Different users download the current model, so that there will be a local copy of the centralized machine learning model on all devices.
  2. Local models train themselves on local datasets of different devices.
  3. The devices, then, transfer only the training results back to the central server, without users data, which remain encrypted.
  4. Results are aggregated in the centralized server.
  5. The centralized server can now update its central machine learning model from the aggregated training results, getting far better than the previously deployed version.
  6. The users can now get the newer model version, a smarter model, built from their own data.


What are the benefits? 

First: Multiple organizations or institutions can work together to solve a machine-learning problem under the coordination of a central server. A common machine learning model is created and continually improved within a central unit, with no need to spend time gathering and aggregating data from different sources every time.

Second: Federated Learning allows the central model of a continuous learning on heterogeneous data. This great enrichment in terms of data diversity results in a high-accuracy model. Besides, collaborative learning can reduce the problem of periodically shortages or insufficient volumes of data in some institutions, or also, peak seasons with no time for data-labeling due to high volumes.
In the healthcare sector, algorithms are trained across multiple healthcare institutions, located in different geographic regions. Therefore, the set of data provided comes from patient populations that really differ for gender ratios, age distributions, ethnicity, and physical traits, contributing to the creation of better models.

Third: Federate Learning relieves the burden of aggregating data on a central and external server with the risk that it may be against the privacy policies of certain organizations and may make the data more vulnerable to data breaches. In the banking sector, for instance, there is a multitude of customer’s sensitive data exposed or data under confidentiality agreements and bank operators are rarely allowed to give access to them.


Use cases: Where can Federated Learning bring profit

Federated Learning applications are spread over a number of industries:


Financial institutions are increasingly adopting machine learning models for the detection of financial crimes or the creation of predictive models. Anti-money laundering programs can widely benefit from these models, detecting any suspicious money movement and illicit activity. However, different factors such as privacy policies and the cost of moving data, poses a risk to any model deployment. Federated learning solves the problem, as it allows to interrogate data sets, without a direct access to any sensitive customer information. This capability can revolutionize the entire Fintech sector, changing employees roles and re-prioritizing the counter-fraud efforts.


Alongside all the various ethical implications that medical staff go through, there is that regarding confidentiality and granting patient privacy. Federated Learning creates a global model without directly sharing datasets. Thus, patient privacy is maintained across sites. But another significant implication is that the clinical predictions accuracy increases as long as the amount of differentiated input data expands. Meaningful medical patterns, as similar diagnoses, rates, statistics are shared over large communities, with positive externalities for the whole society.



Recently, the attention on Federated Learning has increased exponentially. Federated Learning seems to have a lot of potential. Not only it secures user sensitive information, but also aggregates results and identifies common patterns from a lot of users. It’s not hard to imagine how collaborative learning can compete with the creation of communities sharing knowledge and expertise.

Want to learn more about AI?…

Related Articles

Gemmo's noise classification case study with Sonitus