A collaborative machine learning technique called federated learning uses real data without changing it. Federated learning teaches algorithms across numerous decentralised edge devices or servers, in contrast to typical machine learning systems, which require the training data to be centralised into a single computer or data centre.
With the help of this learning method, mobile devices can create a shared prediction model while keeping the training data locally and avoiding online storage.
The following resources can teach you more about federated learning.
The federated learning software framework Substra was developed by a collaborative research effort centred on the French company Owkin, which was launched in 2016. With an emphasis on the medical sector and objectives for data ownership and confidentiality. As part of the MELLODY programme, it is used nowadays in the pharmaceutical industry for drug development.
Substra
For distinct user types, Substra supports a variety of interfaces. For data scientists, it has a Python library, for administrators, command-line interfaces, and for project managers and other sophisticated users, graphical user interfaces. Each node must have a complex Kubernetes setup for Substra’s deployment.
PyGrid plus PySyft
An open-source Python 3 tool called PySyft uses federated learning for research by utilising differential privacy, FL, and encrypted computations. It was developed by the OpenMined community and largely uses TensorFlow and PyTorch for deep learning.
PySyft is capable of two different types of computations:
dynamic calculations with unforeseen data
Graphs of calculations that we can do later in a different computing environment are known as static computations.
A programming language called PySyft defines abstractions, machine learning techniques, and objects. Simple data science projects requiring network communication cannot be completed with PySyft. It would require the use of PyGrid, another module. In addition, PyGrid offers federated learning over a wide range of terminal types, including web, mobile, and edge devices. The API used to scale and maintain PySyft is called PyGrid. To control it, we can utilise PyGrid Admin.
OpenFL
To apply FL to sensitive data, Intel created the open-source Python 3 project known as Intel Open Federated Learning. Although OpenFL includes bash deployment scripts and secures communication with certificates, the user is still responsible for much of this.
Two components make up the library: the collaborator, which uses a local dataset to train global models, and the aggregator, which receives model changes and combines them to create the global model. Both a command-line interface and a Python API are included in OpenFL. The usage of mTLS for communication between nodes necessitates the use of certificates. Every node inside the federation needs to be accredited. OpenFL allows for both lossy and lossless data compression to lower communication costs. In OpenFL, programmers can modify the logging, data splitting, and aggregation algorithms.
The OpenFL design ethos is built on the Federated Learning (FL) Plan. The required collaborators, aggregators, connections, models, data, and any fundamental setup are defined in a YAML file. OpenFL runs in Docker containers to separate federation scenarios.
Federated Learning by IBM
On top of IBM Federated Learning, we can construct more complex features for FL. It is independent of any machine learning framework and supports a wide range of learning topologies, including protocols and a common aggregator.
In order to support a wide range of learning models, topologies, and learning models in commercial and hybrid-Cloud situations, it aims to provide a solid foundation for federated learning. Several machine learning models, including the following, are compatible with IBM Federated Learning:
Models are built using TensorFlow, PyTorch, and Keras.
Ridge regression, logistic regression, linear SVM, and other linear classifiers/regressions (with regularizers)
Various Deep Reinforcement Learning algorithms, include ID3 Decision Tree DQN, DDPG, PPO, and others
theorem of Bayes
Intel Clara
The application framework NVIDIA CLARA was created with use cases in healthcare in mind. It includes full-stack GPU-accelerated frameworks, SDKs, and reference applications to assist researchers, data scientists, and developers in creating federated learning systems that are scalable, secure, and real-time. For instance, the French startup Therapixel presently uses CLARA to improve the precision of a breast cancer diagnosis. The startup uses NVIDIA technology.
Following use cases are compatible with NVIDIA CLARA:
A medical device clarifier is called Clara AGX.
Clara’s Drug Development Discovery
Clara Hospital Protector
Medical imaging is Clara Imaging’s area of expertise.
For Genomic Research, Clara Parabricks
