Federated Learning or Federated machine learning (FML), a type of machine learning, allows multiple parties to train a single model while protecting the privacy of their respective data. In classical machine learning, data collection and storage are often centralized, that is, they happen all at once. On the other hand, when utilizing FML, the data is stored on the computers or servers of the involved parties, and the model is jointly trained without the need for raw data.
In FML, a local model is trained on the data of each party, and the local models are combined to form a global model. This procedure can be repeated in order to improve the global model’s accuracy. The main advantage of FML is that it protects customer privacy while allowing businesses to benefit from group intelligence.
Applications of Federated Learning:
Numerous practical applications of FML can be found in the banking, healthcare, and smart city industries. In the healthcare sector, for example, multiple organizations can collaborate to train a model that can detect illnesses without revealing their patients’ private health information. Financial institutions can collaborate to protect the privacy of their customers’ transaction data while training a fraud detection model.
Since FML is a rapidly evolving field, there are a number of problems that need to be resolved, such as model security, efficient communication, and heterogeneous data.
How does Federated Machine Learning (FML) work?
A subset of machine learning called federated machine learning allows multiple parties to collaborate on creating a machine learning model without disclosing any of their individual data. Instead of collecting data from all participants and centralizing it in one location, federated machine learning allows each person to maintain their data locally and train a model collectively by sharing only model updates rather than raw data. Below is a summary of how federated machine learning works:
With their datasets, the participants hope to develop a machine-learning model. Subsets of the datasets are separated out and kept private for training purposes.
Initialization of the model:
The central server sends the initialized version of a machine learning model (e.g., neural network to each participant.
To train the model, each participant uses their own local dataset. The local dataset is the subset of the data that the model sees during training. The model generates an update with parameters that can be adjusted in order to enhance the model’s fit to the local dataset.
Each participant’s changes are sent to the central server, which combines them to create a new model. The aggregation method is commonly used to compute the average of the updates, where weights are determined by the number of data points in each participant’s local dataset.
After receiving the updated model, each participant applies it to carry out the local training process once more. It is repeated for a predefined number of iterations or until a specific convergence criterion is met.
Federated machine learning, in general, enables various parties to collaborate on training a machine learning model without exchanging raw data. Rather, participants just exchange updates while the model is trained using their local data. When there are concerns about data ownership or privacy, or when the volume or dispersion of the data makes centralized data storage impractical, this tactic can be very useful.
Is federated learning supervised or unsupervised?
It is possible to use federated learning in supervised and unsupervised learning situations. Federated learning involves training a model without data exchange across several decentralized devices or servers that hold local data samples. The model can evolve over time while maintaining data privacy thanks to this collaborative learning methodology. Depending on the use case and application, the underlying learning task may be unsupervised (without labeled data) or supervised (with labeled data).
What are the three types of federated learning?
Depending on the type of learning task and device collaboration, federated learning can be divided into three main categories:
FedEx Horizontal Learning:
Each device (or server) in this type stores data instances from distinct samples within the same feature space. Training a model with the ability to generalize across various instances of the same features is the aim.
Federated Learning Vertically:
Each device (or server) in vertical federated learning contains data pertaining to various features of the same samples. The goal is to work together to create a model that can learn from the entire collection of features from all of the devices.
Federated Learning Transfer:
With this kind, a model is trained on a task or domain and then its knowledge is transferred to another task or domain that is related. The goal is to enhance the model’s performance in a distinct but related setting by utilizing the knowledge gained from one setting.
These three varieties facilitate cooperation and model enhancement amongst decentralized data sources by enabling federated learning to be tailored to diverse scenarios.