Tech Insights

HPE Swarm Learning: Increase accuracy and reduce bias in AI models

Are you dealing with AI models that are often inaccurate and biased due to data privacy and ownership rules limiting access to data? Here’s a new solution: HPE Swarm Learning is a decentralized machine learning solution that gives models access to larger sets of data without violating privacy and ownership rules. This in turn enables better business decisions through more accurate models. By Arshad Khan, AI Product Management at HPE

What’s driving AI adoption today?

The explosion of data and improved processing power is what's driving AI adoption today. Of course, good AI requires more data – and data is everywhere.

Let’s first take a moment to reflect back on AI and learning models, as the AI model training journey has continuously evolved. Initially, local learning was performed where local models were trained at each data source. This approach resulted in local data bias, plus inaccurate and suboptimal models creating inaccurate insights. Next came centralized learning to reduce bias and improve model accuracy. This improved model accuracy since larger data sets were used for training. However, data aggregation to a central location posed new challenges around data privacy, data ownership, and data movement. So along came federated learning to alleviate central learning challenges with a central custodian responsible for aggregating all learnings from multiple edge devices or data sources while also preserving privacy. The central custodian, however, poses challenges around resilience, scalability, and ownership. With the central custodian, local high-availability features may be used for resilience. But, if the central custodian goes down, training stops, and scalability is limited to what the client-server model in a predominantly star topology can support

It's time for a new modern approach to machine learning.

Introducing HPE Swarm LearningHPE-AI-Swarm Learning-BLOG.png

HPE Swarm Learning is a decentralized, privacy-preserving, and collaborative machine learning (ML) framework at the data source, obviating the need for a central server. Developed in HPE Labs, this model training software helps alleviate ML challenges around data aggregation and a central custodian. It’s easy-to-use API is built on a resilient, scalable architecture based on blockchain, and it runs on a heterogeneous infrastructure.

HPE Swarm Learning builds a swarm network of equal and like-minded peers to collaborate for model training by unlocking new business models in the form of consortiums. Uses benefit from being able to:

  • Improve operational efficiency to achieve business decisions with reduced bias
  • Boost efficiency by not having to move or duplicate data
  • Delivers scalability and resilience
  • Supports the development of new business models by enabling collaboration across orgs and geos

The architecture and design at work (data privacy is key)

Data privacy is a key point here. HPE Swarm Learning shares the learnings between the peers but not the underlying data – thus preserving data privacy.

The decentralized architecture enables secure onboarding of the participants of the swarm network. Blockchain enables dynamic leader election for merging the learnings. Participating peers send their learnings to the leader where the learnings are merged. Merged learnings are then communicated back to the swarm participants – and a new global-state model is generated. For each training interval, a new leader is elected via user-defined criteria such as which node completed the training first for a training interval. Each node participates as an equal peer in the network.

The architecture also enables the deployment of HPE Swarm Learning at each data source – as each data source then becomes a “swarm node.” Leveraging training data from multiple sources increases the size of the training data set, providing for improved models for better accuracy.

HPE Swarm Learning is designed to be flexible and easy to use. Simple APIs enable any model to be “swarmified” in three simple steps that reduce the complexity of the AI solution. A web UI is available for ease of installation as well as command-line interfaces for ease of management and batch operations.

HPE Swarm Learning applied to healthcare

Today's healthcare organizations requires improved accuracy for diagnosis and decisions from their AI solutions. To achieve accuracy, AI requires access to larger data sets and to reduce local data biases. Without data aggregation to one location, results from suboptimal models lead to inaccurate decisions.

Aggregating data in healthcare can be particularly challenging because:

  • Data privacy regulations such as HIPAA and GDPR inhibit sharing data.
  • Data ownership prevents data sharing between hospitals and across geographies.
  • Even when data can be aggregated data, inefficiencies occur due to the cost of moving very large diagnostic image data utilizing precious bandwidth, and the fact that data may have to be duplicated utilizing precious storage resources.

With HPE Swarm Learning, healthcare organizations can perform AI model training with large sets of distributed data at the data source – with no movement of data. Collaboration between data sources can happen without compromising privacy. And data ownership concerns are eliminated, as each data owner participates as an equal partner.

The result? Improved models provide more accurate disease classification with reduced bias. Improved time-to-accuracy is achieved as the process of training is all achieved by HPE Swarm Learning, an easy-to-use ML system at the data source.

Reaching the cutting-edge of medical insight with swarm learning

There's no question that swarm learning will reshape healthcare and life sciences operations. The capabilities of swarm learning are driving improvements such as personalized patient care, more accurate diagnoses, decreased time-to-market for new drugs and therapies, and faster disease research. It is already being used to detect cancer, infectious diseases, and genetic conditions, and the list continues to grow.

Healthcare teams using swarm learning have applied the technology for disease diagnosis from images and other data sources. Specifically, when swarm learning is applied to decentralized and confidential clinical machine learning, it can be used to predict leukemias, identify tuberculosis, Covid-19, and more. What’s more, users have reported improved accuracy with swarm learning while preserving data privacy at the same time. They also reported equal or close accuracy to central learning and better results than local learning.

When a hospital applies AI and ML to isolated patient data, the results can be biased due to demographics or the types of cases being seen. Sharing insights across multiple hospitals allows each hospital to continue learning locally without having to share private patient data. At each cycle, swarm learning technology collects the learned weight of the ML peers, averages the inputs, and sends results back to all the hospitals. This gives organizations the ability to learn from everyone at the edge to remove biases and achieve fuel better clinical outcomes.

As just one example, the German Center for Neurodegenerative Diseases (DZNE) is working to develop cure for common diseases, including Alzheimer’s and Parkinson’s, by studying health data of 30,000 patients during several decades. DZNE scientists utilize swarm learning to manage the vast data pools their research requires and share health insights while following privacy laws and compliance regulations which vary between jurisdictions and specialties. As a result, scientists can access key insights to train AI and ML algorithms to drive faster, more informed decision-making in disease research.

Healthcare use case: lung disease detection based on chest x-rays

In a use case focused on examining lung diseases for multiple patients, the patient data was spread across three geographies. Models at one hospital in each location failed to detect infrequently observed diseases. The infrequently observed lung diseases mean that the data had local bias. The swarm learning model was able detect these diseases where the hospitals had limited data and remove local data bias for the category. Even with sufficiently available data, the swarm learning model is either better or at par with any individual model.

More specifically, lung x-ray images were pre-labeled for four diseases. The models were trained locally at each of the three hospitals. Each location with fewer images for the particular disease had a lower accuracy of ~10%. Even where sufficient images were availabile, the accuracy was ~60%. Swarm learning improves model accuracy and subsequent patient diagnosis and achieved an accuracy of ~70%. This is significantly better in cases where there were fewer images and close or at par where sufficient data was available.

Beyond healthcare

In additional to applications in healthcare and the life sciences, HPE Swarm Learning can be applied to all industries. In the federal government sector, examples of swarm learning applications include detecting anomalies and threats, as well as for research collaboration. Financial services organizations apply swarm learning to financial fraud detection activities. And in manufacturing and the oil and gas industries, swarm learning provides predictive maintenance capabilities.

Going forward with HPE Swarm Learning: the industry’s first privacy-preserving, decentralized ML framework

HPE Swarm Learning is the first and only completely decentralized, privacy-preserving, and edge training framework that doesn’t require a central custodian to aggregate and redistribute learning parameters

Unlike competitors, Only HPE Swarm Learning allows peer model learning, not underlying data, ensuring non-biased, secure, accurate model deployment

Read the press release: Hewlett Packard Enterprise Ushers in Next Era in AI Innovation with Swarm Learning Solution Built for the Edge and Distributed Sites

Meet Arshad Khan, AI Product Management at HPE

Arshad Khan-HPE AI.pngArshad specializes in evaluating new technologies and bringing them to the market. Arshad has a unique mix of customer-focused technical and business skills in data storage, telecommunications, healthcare, and the consulting industries with a progressively successful career spanning over 20+ years..


Insights Experts
Hewlett Packard Enterprise


0 Kudos
About the Author


Our team of HPE and other technology experts shares insights about relevant topics related to artificial intelligence, data analytics, IoT, and telco.

HPE Webinars
Find out about the latest live broadcasts and on-demand webinars
Read more
Online Expert Days
Visit this forum and get the schedules for online Expert Days where you can talk to HPE product experts, R&D and support team members and get answers...
Read more
View all