Swarm Learning applied to Medicine

Swarm Learning describes a collective behavior of decentralized and self-organizing systems, and the concept is being employed in artificial intelligence. The inspiration comes from nature, especially from observing swarm phenomena.

As we know, birds, bees, bats, ants, and fish can move collectively and intelligently as a direct response to changes or occurrences in the environment

In the field of Artificial Intelligence, Swarm Learning is a decentralized Machine Learning framework based on blockchain technology and designed to analyze distributed data without the need for aggregation. At the same time, it can reserve privacy, guarantee data ownership, and sovereignty, and add blockchain-based security.

In the framework defined for Swarm Learning, both the model training operations and the inference of the results, using the previously trained model, occur entirely at the edges of the network. Or if you prefer, on the edge cloud

Thus, computing power is used close to or next to the data sources to run the Machine Learning algorithms, as shown in the figure below:

Briefly, a Swarm Learning architecture has five components connected in the form of a distributed network:

• Swarm Learning (SL) nodes. These nodes run a user-defined machine learning algorithm. This algorithm is called the Swarm Learning ML Program and it is responsible for iteratively training and updating the model.
• Swarm Network (SN) nodes. They form the blockchain network (currently based on Ethereum) and interact with each other to maintain global information about the model being trained and track progress.
• Swarm Learning Command Interface (SWCI) node. The SWCI node is the command interface for the Swarm Learning framework and through it, we can view the status, control, and manage the learning framework.
• Spiffe Spire Server nodes. They provide security for the entire network. The platform can run one or more Spire Server nodes that are connected to form a federation.
• License Server. The license to run the Swarm Learning platform is installed and managed by the License Server node. All Swarm Learning nodes must use the same Machine Learning platform: Keras (based on TensorFlow 2) or PyTorch.

Each Swarm Learning (SL) node participating in the network will train a common ML model using its own local data, represented by the organizations in the previous figure, which are not shared with the other nodes in the architecture.

This allows machine learning network participants to maintain the privacy and ownership of their data. Instead of sharing the data, the SL nodes only share the results obtained by training the common learning model locally.

The parameters shared by the participating SL nodes are then merged to obtain a global, collaborative predictive model. The process is repeated successively until the model reaches the desired precision.

The use of a blockchain-based security framework present in the SN nodes ensures that only legitimate participants join the distributed architecture and that each participant is bound by a smart contract both in terms of contribution and rewards.

Swarm Learning is then able to address several challenges inherent in the Healthcare industry: privacy, data ownership, global collaboration, security, and compliance with industry standards (HIPAA, GDPR, CCPA, and others).

The differentials in relation to ML centralized systems

Swarm Learning emerges as an alternative to the centralized architectures of Machine Learning. In these traditional models, data is all moved from the edge, or edge cloud, to a central location. This hub is typically a public cloud (like AWS Sage Maker and GCP Vertex AI) or a private cloud (Datacenter).

Thus, the data is all stored centrally, usually in a Data Lake, or processed in real-time. They are treated and transformed, and the desired learning models are applied. Once the model has been trained, it is again transported to the edges of the architecture for inference and production of results.

And this makes it difficult to circumvent the privacy, ownership, efficiency, and collaboration needs which are especially important in the healthcare area.

Additionally, Swarm Learning is an evolution over federation-based architectures such as Google’s “Federated Learning” project as it eliminates the need for a central leader: AI modeling is done completely by the devices at the edge of the network.

HPE Swarm Learning

On April 27, 2022, HPE announced its Swarm Learning solution. It was developed by its Research and Development division, HPE Labs, and promises to bring the technology not only to the healthcare sector, but also to meet the demands of the financial, public, and industrial sectors.

Press Release

Solution Page

White Paper

Practical application in Medicine

An article published in Nature on May 26, 2021, shows the potential of the Swarm Learning framework. 04 use cases applied to heterogeneous diseases (COVID-19, tuberculosis, leukemia, and pulmonary pathologies) were analyzed. The classifiers obtained by Swarm Learning outperformed those developed individually and fully complied with local confidentiality regulations.

https://www.nature.com/articles/s41586-021-03583-3

Rights and permissions: This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution, and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made: https://creativecommons.org/licenses/by/4.0/

Credits: Hewlett Packard Enterprise, Amazon Web Services, and Nature.